Published AI speed benchmarks
796 speeds that other people measured with llama.cpp and LocalScore, the open tools most local AI runs on. They cover Apple chips, graphics cards and laptop processors. We collected the numbers and link every one to its source and the date we read it.
Reported by others. These are not Findra scores. Each row is one run of the standard llama-bench test on a small 7-billion-parameter model (Llama 2 7B, 512-token prompt, 128 written tokens). Bigger models are slower, and settings differ between people. For a laptop that is not listed here, the finder estimates speed from its memory bandwidth. Download the data: reported-speeds.json (free to reuse, credit Findra).
Apple chips
Metal, the way Apple chips run local AI. Fastest first.
| Chip | Model and size | Writing (tokens per second) | Reading (tokens per second) | Source |
|---|---|---|---|---|
| Apple M5 Ultra 80 graphics cores | Llama 2 7B, Q4_0 | 179.1 | 4945 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Max 40 graphics cores | Llama 2 7B, Q4_0 | 119.9 | 3220 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Ultra 80 graphics cores | Llama 2 7B, Q8_0 | 119.0 | 4911 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Ultra 76 graphics cores | Llama 2 7B, Q4_0 | 94.3 | 1238 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Ultra 80 graphics cores | Llama 2 7B, Q4_0 | 92.1 | 1471 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Ultra 60 graphics cores | Llama 2 7B, Q4_0 | 88.6 | 1014 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Ultra 60 graphics cores | Llama 2 7B, Q4_0 | 88.4 | 1073 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Ultra 64 graphics cores | Llama 2 7B, Q4_0 | 83.7 | 1030 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Max 40 graphics cores | Llama 2 7B, Q4_0 | 83.1 | 886 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Ultra 48 graphics cores | Llama 2 7B, Q4_0 | 74.9 | 772 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Ultra 80 graphics cores | Llama 2 7B, F16 | 73.6 | 5207 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Max 40 graphics cores | Llama 2 7B, Q8_0 | 72.4 | 3144 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Max 32 graphics cores | Llama 2 7B, Q4_0 | 70.0 | 714 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Pro 16 graphics cores | Llama 2 7B, Q4_0 | 67.6 | 1340 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Ultra 76 graphics cores | Llama 2 7B, Q8_0 | 66.6 | 1249 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Pro 20 graphics cores | Llama 2 7B, Q4_0 | 66.3 | 1621 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Max 40 graphics cores | Llama 2 7B, Q4_0 | 66.3 | 760 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Max 38 graphics cores | Llama 2 7B, Q4_0 | 66.0 | 671 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Ultra 80 graphics cores | Llama 2 7B, Q8_0 | 63.9 | 1488 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Ultra 60 graphics cores | Llama 2 7B, Q8_0 | 63.5 | 1086 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Ultra 60 graphics cores | Llama 2 7B, Q8_0 | 62.1 | 1003 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Max 32 graphics cores | Llama 2 7B, Q4_0 | 61.2 | 530 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Max 30 graphics cores | Llama 2 7B, Q4_0 | 61.0 | 538 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Ultra 64 graphics cores | Llama 2 7B, Q8_0 | 59.9 | 1043 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Max 30 graphics cores | Llama 2 7B, Q4_0 | 56.6 | 568 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Ultra 48 graphics cores | Llama 2 7B, Q8_0 | 55.7 | 783 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Max 24 graphics cores | Llama 2 7B, Q4_0 | 54.6 | 400 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Max 40 graphics cores | Llama 2 7B, Q8_0 | 54.0 | 892 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Pro 20 graphics cores | Llama 2 7B, Q4_0 | 50.7 | 440 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Pro 16 graphics cores | Llama 2 7B, Q4_0 | 49.6 | 364 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Max 32 graphics cores | Llama 2 7B, Q8_0 | 43.9 | 719 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Max 40 graphics cores | Llama 2 7B, Q8_0 | 42.8 | 758 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Ultra 60 graphics cores | Llama 2 7B, F16 | 42.2 | 1122 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Max 38 graphics cores | Llama 2 7B, Q8_0 | 41.8 | 678 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Ultra 76 graphics cores | Llama 2 7B, F16 | 41.0 | 1402 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Max 32 graphics cores | Llama 2 7B, Q8_0 | 40.2 | 537 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Max 30 graphics cores | Llama 2 7B, Q8_0 | 40.0 | 540 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Ultra 60 graphics cores | Llama 2 7B, F16 | 39.9 | 1129 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Ultra 80 graphics cores | Llama 2 7B, F16 | 39.8 | 1538 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Pro 16 graphics cores | Llama 2 7B, Q8_0 | 38.9 | 1303 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Pro 20 graphics cores | Llama 2 7B, Q8_0 | 38.9 | 1554 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Pro 19 graphics cores | Llama 2 7B, Q4_0 | 38.9 | 341 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Pro 16 graphics cores | Llama 2 7B, Q4_0 | 37.9 | 294 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Max 24 graphics cores | Llama 2 7B, Q8_0 | 37.8 | 406 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Max 40 graphics cores | Llama 2 7B, F16 | 37.1 | 3158 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Ultra 64 graphics cores | Llama 2 7B, F16 | 37.0 | 1169 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Pro 16 graphics cores | Llama 2 7B, Q4_0 | 36.4 | 266 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M6 8 graphics cores | Llama 2 7B, Q4_0 | 35.7 | 989 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Pro 14 graphics cores | Llama 2 7B, Q4_0 | 35.5 | 233 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Max 30 graphics cores | Llama 2 7B, Q8_0 | 34.3 | 566 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Ultra 48 graphics cores | Llama 2 7B, F16 | 33.9 | 876 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Max 40 graphics cores | Llama 2 7B, F16 | 31.6 | 923 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Pro 18 graphics cores | Llama 2 7B, Q4_0 | 30.7 | 342 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Pro 20 graphics cores | Llama 2 7B, Q8_0 | 30.7 | 450 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Pro 16 graphics cores | Llama 2 7B, Q8_0 | 30.5 | 367 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Max 40 graphics cores | Llama 2 7B, F16 | 25.1 | 779 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Max 38 graphics cores | Llama 2 7B, F16 | 24.6 | 756 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Max 32 graphics cores | Llama 2 7B, F16 | 24.3 | 736 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Max 30 graphics cores | Llama 2 7B, F16 | 24.2 | 600 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 10 graphics cores | Llama 2 7B, Q4_0 | 24.1 | 221 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Max 32 graphics cores | Llama 2 7B, F16 | 23.0 | 600 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Pro 19 graphics cores | Llama 2 7B, Q8_0 | 23.0 | 344 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Pro 16 graphics cores | Llama 2 7B, Q8_0 | 22.7 | 288 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Max 24 graphics cores | Llama 2 7B, F16 | 22.6 | 453 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Pro 16 graphics cores | Llama 2 7B, Q8_0 | 22.3 | 270 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Pro 14 graphics cores | Llama 2 7B, Q8_0 | 21.9 | 235 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 10 graphics cores | Llama 2 7B, Q4_0 | 21.9 | 180 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 8 graphics cores | Llama 2 7B, Q4_0 | 21.7 | 146 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Pro 20 graphics cores | Llama 2 7B, F16 | 21.6 | 1589 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M5 Pro 16 graphics cores | Llama 2 7B, F16 | 21.4 | 1311 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M6 8 graphics cores | Llama 2 7B, Q8_0 | 21.0 | 970 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Max 30 graphics cores | Llama 2 7B, F16 | 19.5 | 589 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Pro 18 graphics cores | Llama 2 7B, Q8_0 | 17.5 | 345 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Pro 16 graphics cores | Llama 2 7B, F16 | 17.2 | 381 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 Pro 20 graphics cores | Llama 2 7B, F16 | 17.2 | 464 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 7 graphics cores | Llama 2 7B, Q4_0 | 14.2 | 108 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 8 graphics cores | Llama 2 7B, Q4_0 | 14.2 | 118 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 10 graphics cores | Llama 2 7B, Q8_0 | 13.5 | 224 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Pro 19 graphics cores | Llama 2 7B, F16 | 13.1 | 384 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Pro 14 graphics cores | Llama 2 7B, F16 | 12.8 | 263 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 Pro 16 graphics cores | Llama 2 7B, F16 | 12.8 | 302 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 Pro 16 graphics cores | Llama 2 7B, F16 | 12.5 | 313 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 10 graphics cores | Llama 2 7B, Q8_0 | 12.2 | 181 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 8 graphics cores | Llama 2 7B, Q8_0 | 12.2 | 147 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M6 8 graphics cores | Llama 2 7B, F16 | 11.6 | 953 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M3 Pro 18 graphics cores | Llama 2 7B, F16 | 9.9 | 357 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 7 graphics cores | Llama 2 7B, Q8_0 | 7.9 | 108 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M1 8 graphics cores | Llama 2 7B, Q8_0 | 7.9 | 117 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M4 10 graphics cores | Llama 2 7B, F16 | 7.4 | 230 | llama.cpp discussion 4167 read 2026-10-10 |
| Apple M2 10 graphics cores | Llama 2 7B, F16 | 6.7 | 201 | llama.cpp discussion 4167 read 2026-10-10 |
Graphics cards and other processors
CUDA, Vulkan, ROCm, OpenCL and processor-only runs. Fastest first. A desktop graphics card is not the same part as its laptop version.
| Chip or card | Backend | Model and size | Writing (tokens per second) | Reading (tokens per second) | Source |
|---|---|---|---|---|---|
| RTX 5090 | CUDA | Llama 2 7B, Q4_0 | 300.4 | 14970 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5090 | CUDA | Llama 2 7B, Q4_0 | 290.0 | 14073 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX PRO 6000 Blackwell | CUDA | Llama 2 7B, Q4_0 | 281.1 | 16619 | llama.cpp discussion 15013 read 2026-10-10 |
| H100 80 GB | CUDA | Llama 2 7B, Q4_0 | 280.7 | 11263 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX PRO 6000 Blackwell | CUDA | Llama 2 7B, Q4_0 | 274.2 | 14855 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5090 | Vulkan | Llama 2 7B, Q4_0 | 273.7 | 11796 | llama.cpp discussion 10879 read 2026-10-20 |
| H100 80 GB | CUDA | Llama 2 7B, Q4_0 | 267.8 | 9918 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5090 | Vulkan | Llama 2 7B, Q4_0 | 263.6 | 10382 | llama.cpp discussion 10879 read 2026-10-20 |
| Instinct MI300X | ROCm (HIP) | Llama 2 7B, Q4_0 | 232.9 | 11476 | llama.cpp discussion 15021 read 2026-10-10 |
| Instinct MI300X | ROCm (HIP) | Llama 2 7B, Q4_0 | 218.5 | 11946 | llama.cpp discussion 15021 read 2026-10-10 |
| A100 80 GB | CUDA | Llama 2 7B, Q4_0 | 200.9 | 5286 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5080 | Vulkan | Llama 2 7B, Q4_0 | 192.2 | 8055 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4090 D | CUDA | Llama 2 7B, Q4_0 | 191.6 | 12507 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 7900 XTX | Vulkan | Llama 2 7B, Q4_0 | 190.9 | 3890 | llama.cpp discussion 10879 read 2026-10-20 |
| A100 80 GB | CUDA | Llama 2 7B, Q4_0 | 190.9 | 4850 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4090 | Vulkan | Llama 2 7B, Q4_0 | 190.1 | 10830 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4090 D | CUDA | Llama 2 7B, Q4_0 | 189.3 | 10294 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4090 | CUDA | Llama 2 7B, Q4_0 | 189.0 | 14771 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4090 | Vulkan | Llama 2 7B, Q4_0 | 188.0 | 9452 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4090 | CUDA | Llama 2 7B, Q4_0 | 186.2 | 11993 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5080 | Vulkan | Llama 2 7B, Q4_0 | 185.1 | 7445 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 5080 | CUDA | Llama 2 7B, Q4_0 | 184.7 | 9488 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 7900 XTX | Vulkan | Llama 2 7B, Q4_0 | 182.6 | 3727 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 5070 Ti | CUDA | Llama 2 7B, Q4_0 | 182.4 | 8420 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5080 | CUDA | Llama 2 7B, Q4_0 | 182.0 | 8297 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 6000 Ada | CUDA | Llama 2 7B, Q4_0 | 179.5 | 10577 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5070 Ti | CUDA | Llama 2 7B, Q4_0 | 176.8 | 6952 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5070 Ti | Vulkan | Llama 2 7B, Q4_0 | 176.2 | 7568 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 6000 Ada | CUDA | Llama 2 7B, Q4_0 | 176.1 | 9229 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 3090 Ti | CUDA | Llama 2 7B, Q4_0 | 172.3 | 6924 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 3090 | Vulkan | Llama 2 7B, Q4_0 | 171.6 | 5201 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3090 Ti | CUDA | Llama 2 7B, Q4_0 | 171.2 | 6567 | llama.cpp discussion 15013 read 2026-10-10 |
| A100 | Vulkan | Llama 2 7B, Q4_0 | 170.6 | 7064 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7900 XTX | ROCm (HIP) | Llama 2 7B, Q4_0 | 170.1 | 3874 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX PRO 4500 Blackwell | CUDA | Llama 2 7B, Q4_0 | 168.9 | 7252 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 7900 XTX | ROCm (HIP) | Llama 2 7B, Q4_0 | 167.1 | 3552 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX 3090 | Vulkan | Llama 2 7B, Q4_0 | 164.1 | 4666 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3090 | CUDA | Llama 2 7B, Q4_0 | 161.9 | 5560 | llama.cpp discussion 15013 read 2026-10-10 |
| A100 | Vulkan | Llama 2 7B, Q4_0 | 160.8 | 6390 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3090 | CUDA | Llama 2 7B, Q4_0 | 158.2 | 5175 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5090 Laptop | CUDA | Llama 2 7B, Q4_0 | 158.1 | 7642 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5090 Laptop | CUDA | Llama 2 7B, Q4_0 | 156.5 | 6667 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5090 Mobile | Vulkan | Llama 2 7B, Q4_0 | 154.3 | 6011 | llama.cpp discussion 10879 read 2026-10-20 |
| L40 | CUDA | Llama 2 7B, Q4_0 | 153.8 | 10098 | llama.cpp discussion 15013 read 2026-10-10 |
| AI Pro R9700 | Vulkan | Llama 2 7B, Q4_0 | 152.7 | 5908 | llama.cpp discussion 10879 read 2026-10-20 |
| L40 | CUDA | Llama 2 7B, Q4_0 | 152.0 | 8870 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4080 Super | Vulkan | Llama 2 7B, Q4_0 | 150.2 | 8007 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 5090 Mobile | Vulkan | Llama 2 7B, Q4_0 | 149.4 | 5546 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4080 SUPER | CUDA | Llama 2 7B, Q4_0 | 148.3 | 8125 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4080 SUPER | CUDA | Llama 2 7B, Q4_0 | 147.5 | 9439 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4080 Super | Vulkan | Llama 2 7B, Q4_0 | 147.1 | 7101 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3080 | Vulkan | Llama 2 7B, Q4_0 | 145.7 | 4914 | llama.cpp discussion 10879 read 2026-10-20 |
| AI Pro R9700 | Vulkan | Llama 2 7B, Q4_0 | 145.7 | 5610 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX A6000 | CUDA | Llama 2 7B, Q4_0 | 144.9 | 5662 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4080 | CUDA | Llama 2 7B, Q4_0 | 143.5 | 9206 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4080 | CUDA | Llama 2 7B, Q4_0 | 142.5 | 8032 | llama.cpp discussion 15013 read 2026-10-10 |
| Tesla V100 | Vulkan | Llama 2 7B, Q4_0 | 142.1 | 1411 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX A5000 | Vulkan | Llama 2 7B, Q4_0 | 140.4 | 4071 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3080 | CUDA | Llama 2 7B, Q4_0 | 139.9 | 5570 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX A5000 | Vulkan | Llama 2 7B, Q4_0 | 139.9 | 3642 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3080 | CUDA | Llama 2 7B, Q4_0 | 139.7 | 5014 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 3080 | Vulkan | Llama 2 7B, Q4_0 | 139.2 | 4287 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX A6000 | CUDA | Llama 2 7B, Q4_0 | 138.7 | 4914 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 9070 XT | Vulkan | Llama 2 7B, Q4_0 | 137.1 | 5036 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX PRO 4000 Blackwell | CUDA | Llama 2 7B, Q4_0 | 136.4 | 5674 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX A5000 | CUDA | Llama 2 7B, Q4_0 | 135.8 | 4552 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4070 Ti Super | Vulkan | Llama 2 7B, Q4_0 | 135.8 | 6801 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 5070 Ti | Vulkan | Llama 2 7B, Q4_0 | 135.6 | 6214 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla V100 | CUDA | Llama 2 7B, Q4_0 | 134.8 | 2974 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4070 Ti SUPER | CUDA | Llama 2 7B, Q4_0 | 132.8 | 7612 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4070 Ti SUPER | CUDA | Llama 2 7B, Q4_0 | 132.3 | 6925 | llama.cpp discussion 15013 read 2026-10-10 |
| A30 | CUDA | Llama 2 7B, Q4_0 | 131.9 | 3069 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX PRO 4000 Blackwell | CUDA | Llama 2 7B, Q4_0 | 131.7 | 4993 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 9070 XT | Vulkan | Llama 2 7B, Q4_0 | 131.5 | 5048 | llama.cpp discussion 10879 read 2026-10-20 |
| Instinct MI210 | ROCm (HIP) | Llama 2 7B, Q4_0 | 130.2 | 2572 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX A5000 | CUDA | Llama 2 7B, Q4_0 | 130.1 | 4028 | llama.cpp discussion 15013 read 2026-10-10 |
| Tesla V100 | Vulkan | Llama 2 7B, Q4_0 | 129.6 | 1391 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4070 Ti Super | Vulkan | Llama 2 7B, Q4_0 | 129.4 | 6099 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla V100 | CUDA | Llama 2 7B, Q4_0 | 129.1 | 3043 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5070 | CUDA | Llama 2 7B, Q4_0 | 128.2 | 5783 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 5070 | CUDA | Llama 2 7B, Q4_0 | 127.5 | 5185 | llama.cpp discussion 15013 read 2026-10-10 |
| Pro W7900 | ROCm (HIP) | Llama 2 7B, Q4_0 | 127.4 | 3473 | llama.cpp discussion 15021 read 2026-10-10 |
| A40 | CUDA | Llama 2 7B, Q4_0 | 126.2 | 5256 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 7800 XT | Vulkan | Llama 2 7B, Q4_0 | 124.9 | 2197 | llama.cpp discussion 10879 read 2026-10-20 |
| A30 | CUDA | Llama 2 7B, Q4_0 | 124.8 | 2767 | llama.cpp discussion 15013 read 2026-10-10 |
| Instinct MI210 | ROCm (HIP) | Llama 2 7B, Q4_0 | 124.5 | 2486 | llama.cpp discussion 15021 read 2026-10-10 |
| A40 | CUDA | Llama 2 7B, Q4_0 | 124.1 | 4609 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 7900 XT | Vulkan | Llama 2 7B, Q4_0 | 123.2 | 2942 | llama.cpp discussion 10879 read 2026-10-20 |
| Pro W7900 | ROCm (HIP) | Llama 2 7B, Q4_0 | 121.2 | 3213 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 7900 XT | Vulkan | Llama 2 7B, Q4_0 | 120.6 | 2701 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 9070 | Vulkan | Llama 2 7B, Q4_0 | 119.7 | 3164 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 9070 | Vulkan | Llama 2 7B, Q4_0 | 119.5 | 2860 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7800 XT | Vulkan | Llama 2 7B, Q4_0 | 118.3 | 2017 | llama.cpp discussion 10879 read 2026-10-20 |
| Instinct MI50 | Vulkan | Llama 2 7B, Q4_0 | 117.9 | 1127 | llama.cpp discussion 10879 read 2026-10-20 |
| A30 | Vulkan | Llama 2 7B, Q4_0 | 117.4 | 3158 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7900 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 116.2 | 3098 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 7900 GRE | Vulkan | Llama 2 7B, Q4_0 | 116.1 | 2336 | llama.cpp discussion 10879 read 2026-10-20 |
| Apple M3 Ultra | Vulkan | Llama 2 7B, Q4_0 | 115.5 | 1117 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 9070 | ROCm (HIP) | Llama 2 7B, Q4_0 | 115.3 | 2453 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 9070 | ROCm (HIP) | Llama 2 7B, Q4_0 | 114.5 | 2382 | llama.cpp discussion 15021 read 2026-10-10 |
| Intel Arc Pro B70 | Vulkan | Llama 2 7B, Q4_0 | 114.2 | 3151 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4070 Ti | Vulkan | Llama 2 7B, Q4_0 | 113.5 | 5515 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7900 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 112.3 | 3262 | llama.cpp discussion 15021 read 2026-10-10 |
| Titan V | CUDA | Llama 2 7B, Q4_0 | 112.2 | 2481 | llama.cpp discussion 15013 read 2026-10-10 |
| Intel Arc Pro B70 | Vulkan | Llama 2 7B, Q4_0 | 112.0 | 3379 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4070 Super | Vulkan | Llama 2 7B, Q4_0 | 112.0 | 5199 | llama.cpp discussion 10879 read 2026-10-20 |
| A30 | Vulkan | Llama 2 7B, Q4_0 | 111.5 | 3291 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon VII | Vulkan | Llama 2 7B, Q4_0 | 110.9 | 1007 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4070 Ti | Vulkan | Llama 2 7B, Q4_0 | 110.5 | 4981 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7900 GRE | Vulkan | Llama 2 7B, Q4_0 | 110.5 | 2251 | llama.cpp discussion 10879 read 2026-10-20 |
| Instinct MI100 | ROCm (HIP) | Llama 2 7B, Q4_0 | 110.5 | 2733 | llama.cpp discussion 15021 read 2026-10-10 |
| Titan V | Vulkan | Llama 2 7B, Q4_0 | 109.2 | 793 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 2080 Ti | CUDA | Llama 2 7B, Q4_0 | 109.2 | 3108 | llama.cpp discussion 15013 read 2026-10-10 |
| Titan V | CUDA | Llama 2 7B, Q4_0 | 108.8 | 2617 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4070 Super | Vulkan | Llama 2 7B, Q4_0 | 108.7 | 4608 | llama.cpp discussion 10879 read 2026-10-20 |
| Instinct MI50 | Vulkan | Llama 2 7B, Q4_0 | 108.5 | 1120 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro VII | Vulkan | Llama 2 7B, Q4_0 | 108.5 | 784 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 6900 XT | Vulkan | Llama 2 7B, Q4_0 | 108.0 | 1901 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 2080 Ti | CUDA | Llama 2 7B, Q4_0 | 107.5 | 2891 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 6800 | Vulkan | Llama 2 7B, Q4_0 | 106.5 | 1598 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 6900 XT | Vulkan | Llama 2 7B, Q4_0 | 106.2 | 1762 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro VII | Vulkan | Llama 2 7B, Q4_0 | 106.0 | 912 | llama.cpp discussion 10879 read 2026-10-20 |
| Instinct MI50 | ROCm (HIP) | Llama 2 7B, Q4_0 | 105.8 | 1129 | llama.cpp discussion 15021 read 2026-10-10 |
| Titan V | Vulkan | Llama 2 7B, Q4_0 | 105.1 | 796 | llama.cpp discussion 10879 read 2026-10-20 |
| Instinct MI100 | ROCm (HIP) | Llama 2 7B, Q4_0 | 104.7 | 2755 | llama.cpp discussion 15021 read 2026-10-10 |
| Quadro RTX 6000 | CUDA | Llama 2 7B, Q4_0 | 104.4 | 3054 | llama.cpp discussion 15013 read 2026-10-10 |
| Quadro RTX 8000 | CUDA | Llama 2 7B, Q4_0 | 103.6 | 3052 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX A4500 | CUDA | Llama 2 7B, Q4_0 | 103.0 | 3453 | llama.cpp discussion 15013 read 2026-10-10 |
| Quadro RTX 6000 | CUDA | Llama 2 7B, Q4_0 | 102.8 | 2751 | llama.cpp discussion 15013 read 2026-10-10 |
| Quadro RTX 8000 | CUDA | Llama 2 7B, Q4_0 | 102.7 | 2710 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 6800 | Vulkan | Llama 2 7B, Q4_0 | 101.5 | 1594 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 9070 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 101.3 | 5055 | llama.cpp discussion 15021 read 2026-10-10 |
| Radeon VII | Vulkan | Llama 2 7B, Q4_0 | 101.2 | 1059 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 2080 Ti | Vulkan | Llama 2 7B, Q4_0 | 101.0 | 1936 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7800 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 100.9 | 2152 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 6800 XT | Vulkan | Llama 2 7B, Q4_0 | 100.5 | 1705 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 6800 XT | Vulkan | Llama 2 7B, Q4_0 | 100.3 | 1753 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro W6800X Duo | Vulkan | Llama 2 7B, Q4_0 | 100.1 | 795 | llama.cpp discussion 10879 read 2026-10-20 |
| Instinct MI50 | ROCm (HIP) | Llama 2 7B, Q4_0 | 99.0 | 1057 | llama.cpp discussion 15021 read 2026-10-10 |
| AI PRO R9700 | ROCm (HIP) | Llama 2 7B, Q4_0 | 98.0 | 4773 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX 2080 Ti | Vulkan | Llama 2 7B, Q4_0 | 97.6 | 1888 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7900 GRE | ROCm (HIP) | Llama 2 7B, Q4_0 | 97.5 | 1599 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX A4500 | CUDA | Llama 2 7B, Q4_0 | 97.3 | 2827 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 9070 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 97.3 | 4904 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX 5060 Ti | Vulkan | Llama 2 7B, Q4_0 | 97.0 | 3913 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7900 GRE | ROCm (HIP) | Llama 2 7B, Q4_0 | 96.1 | 1457 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 7800 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 96.0 | 2305 | llama.cpp discussion 15021 read 2026-10-10 |
| Radeon Pro W6800X Duo | Vulkan | Llama 2 7B, Q4_0 | 94.8 | 688 | llama.cpp discussion 10879 read 2026-10-20 |
| AI PRO R9700 | ROCm (HIP) | Llama 2 7B, Q4_0 | 93.8 | 4444 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX 5060 Ti | Vulkan | Llama 2 7B, Q4_0 | 93.5 | 3461 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 5060 Ti | CUDA | Llama 2 7B, Q4_0 | 93.5 | 4196 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4070 | Vulkan | Llama 2 7B, Q4_0 | 92.3 | 3179 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4070 | Vulkan | Llama 2 7B, Q4_0 | 91.5 | 4294 | llama.cpp discussion 10879 read 2026-10-20 |
| Instinct MI60 | ROCm (HIP) | Llama 2 7B, Q4_0 | 91.5 | 1289 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX 5060 Ti | CUDA | Llama 2 7B, Q4_0 | 90.9 | 3737 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 3060 Ti | Vulkan | Llama 2 7B, Q4_0 | 90.3 | 2645 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 6900 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 88.5 | 1890 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX 2070 SUPER | CUDA | Llama 2 7B, Q4_0 | 88.1 | 2088 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 2070 SUPER | CUDA | Llama 2 7B, Q4_0 | 87.7 | 2293 | llama.cpp discussion 15013 read 2026-10-10 |
| Pro VII | ROCm (HIP) | Llama 2 7B, Q4_0 | 87.5 | 1065 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX 3060 Ti | Vulkan | Llama 2 7B, Q4_0 | 86.7 | 2361 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro W6800X | Vulkan | Llama 2 7B, Q4_0 | 86.5 | 511 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX A4000 | CUDA | Llama 2 7B, Q4_0 | 85.2 | 2808 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 6900 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 85.0 | 1948 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 6800 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 83.9 | 1447 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 6700 XT | Vulkan | Llama 2 7B, Q4_0 | 83.9 | 1051 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX A4000 | CUDA | Llama 2 7B, Q4_0 | 83.8 | 2684 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 6750 XT | Vulkan | Llama 2 7B, Q4_0 | 82.3 | 997 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 6750 XT | Vulkan | Llama 2 7B, Q4_0 | 82.0 | 1041 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 6700 XT | Vulkan | Llama 2 7B, Q4_0 | 81.9 | 1011 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro V620 | Vulkan | Llama 2 7B, Q4_0 | 81.8 | 1595 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3060 | Vulkan | Llama 2 7B, Q4_0 | 80.6 | 2013 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro V620 | Vulkan | Llama 2 7B, Q4_0 | 79.2 | 1556 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3070 | Vulkan | Llama 2 7B, Q4_0 | 78.7 | 2113 | llama.cpp discussion 10879 read 2026-10-20 |
| Instinct MI60 | Vulkan | Llama 2 7B, Q4_0 | 78.2 | 369 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3060 | CUDA | Llama 2 7B, Q4_0 | 76.9 | 2408 | llama.cpp discussion 15013 read 2026-10-10 |
| Titan Xp | CUDA | Llama 2 7B, Q4_0 | 76.1 | 1155 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX A4000 | Vulkan | Llama 2 7B, Q4_0 | 76.1 | 2483 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla T10 | Vulkan | Llama 2 7B, Q4_0 | 76.0 | 1840 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3060 | Vulkan | Llama 2 7B, Q4_0 | 75.9 | 1816 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3060 | CUDA | Llama 2 7B, Q4_0 | 75.6 | 2138 | llama.cpp discussion 15013 read 2026-10-10 |
| Apple M4 Max | Vulkan | Llama 2 7B, Q4_0 | 75.0 | 725 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla T10 | Vulkan | Llama 2 7B, Q4_0 | 75.0 | 1693 | llama.cpp discussion 10879 read 2026-10-20 |
| Pro V620 | ROCm (HIP) | Llama 2 7B, Q4_0 | 74.7 | 1804 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 5700 XT | Vulkan | Llama 2 7B, Q4_0 | 74.4 | 538 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 9060 XT | Vulkan | Llama 2 7B, Q4_0 | 74.0 | 3228 | llama.cpp discussion 10879 read 2026-10-20 |
| Titan Xp | CUDA | Llama 2 7B, Q4_0 | 73.8 | 1218 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX A4000 | Vulkan | Llama 2 7B, Q4_0 | 73.7 | 2248 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Arc B580 | Vulkan | Llama 2 7B, Q4_0 | 72.0 | 419 | llama.cpp discussion 10879 read 2026-10-20 |
| Apple M4 Max | Vulkan | Llama 2 7B, Q4_0 | 71.8 | 557 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1080 Ti | Vulkan | Llama 2 7B, Q4_0 | 71.6 | 648 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro W5700 | Vulkan | Llama 2 7B, Q4_0 | 71.3 | 447 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 9060 XT | Vulkan | Llama 2 7B, Q4_0 | 71.1 | 2845 | llama.cpp discussion 10879 read 2026-10-20 |
| Pro V620 | ROCm (HIP) | Llama 2 7B, Q4_0 | 70.8 | 1257 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 5700 XT | Vulkan | Llama 2 7B, Q4_0 | 70.7 | 530 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Arc Pro B60 | Vulkan | Llama 2 7B, Q4_0 | 70.5 | 275 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Arc B580 | Vulkan | Llama 2 7B, Q4_0 | 70.1 | 621 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro V540 | Vulkan | Llama 2 7B, Q4_0 | 69.6 | 584 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Arc Pro B60 | Vulkan | Llama 2 7B, Q4_0 | 68.5 | 522 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro W5700 | Vulkan | Llama 2 7B, Q4_0 | 68.5 | 450 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1080 Ti | Vulkan | Llama 2 7B, Q4_0 | 67.8 | 586 | llama.cpp discussion 10879 read 2026-10-20 |
| RX Vega 64 | Vulkan | Llama 2 7B, Q4_0 | 67.7 | 585 | llama.cpp discussion 10879 read 2026-10-20 |
| Quadro RTX 4000 | CUDA | Llama 2 7B, Q4_0 | 67.6 | 1663 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 9060 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 67.6 | 1420 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 5700 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 67.5 | 354 | llama.cpp discussion 15021 read 2026-10-10 |
| P102-100 | Vulkan | Llama 2 7B, Q4_0 | 66.5 | 575 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla P100 | Vulkan | Llama 2 7B, Q4_0 | 66.5 | 686 | llama.cpp discussion 10879 read 2026-10-20 |
| Quadro RTX 4000 | CUDA | Llama 2 7B, Q4_0 | 65.6 | 1537 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 9060 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 65.4 | 1479 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX 2070 Super | Vulkan | Llama 2 7B, Q4_0 | 64.6 | 1199 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 6650 XT | Vulkan | Llama 2 7B, Q4_0 | 64.5 | 1089 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4060 Ti | CUDA | Llama 2 7B, Q4_0 | 64.0 | 3803 | llama.cpp discussion 15013 read 2026-10-10 |
| Instinct MI25 | ROCm (HIP) | Llama 2 7B, Q4_0 | 63.9 | 410 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX 4060 Ti | CUDA | Llama 2 7B, Q4_0 | 63.9 | 3395 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 3070 Mobile | Vulkan | Llama 2 7B, Q4_0 | 63.6 | 1689 | llama.cpp discussion 10879 read 2026-10-20 |
| RX Vega 64 | Vulkan | Llama 2 7B, Q4_0 | 63.4 | 575 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla P100 | Vulkan | Llama 2 7B, Q4_0 | 63.2 | 678 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD BC-250 | Vulkan | Llama 2 7B, Q4_0 | 63.1 | 357 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3070 Mobile | Vulkan | Llama 2 7B, Q4_0 | 62.9 | 1832 | llama.cpp discussion 10879 read 2026-10-20 |
| P102-100 | Vulkan | Llama 2 7B, Q4_0 | 62.8 | 511 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1080 Ti | CUDA | Llama 2 7B, Q4_0 | 62.5 | 1084 | llama.cpp discussion 15013 read 2026-10-10 |
| AMD BC-250 | Vulkan | Llama 2 7B, Q4_0 | 62.3 | 371 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 6650 XT | Vulkan | Llama 2 7B, Q4_0 | 62.1 | 1030 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 5700 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 62.0 | 314 | llama.cpp discussion 15021 read 2026-10-10 |
| Tesla P100 | CUDA | Llama 2 7B, Q4_0 | 62.0 | 787 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX A4000 Ada | CUDA | Llama 2 7B, Q4_0 | 61.8 | 2780 | llama.cpp discussion 15013 read 2026-10-10 |
| GTX 1080 Ti | CUDA | Llama 2 7B, Q4_0 | 61.4 | 1138 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX A4000 Ada | CUDA | Llama 2 7B, Q4_0 | 61.4 | 3172 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 2060 SUPER | CUDA | Llama 2 7B, Q4_0 | 61.1 | 1564 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 7600 | Vulkan | Llama 2 7B, Q4_0 | 60.9 | 1353 | llama.cpp discussion 10879 read 2026-10-20 |
| GB10 | Vulkan | Llama 2 7B, Q4_0 | 60.4 | 3322 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 2060 SUPER | CUDA | Llama 2 7B, Q4_0 | 60.0 | 1420 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 4060 Mobile | Vulkan | Llama 2 7B, Q4_0 | 60.0 | 2358 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4060 Mobile | Vulkan | Llama 2 7B, Q4_0 | 59.5 | 2136 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla P40 | Vulkan | Llama 2 7B, Q4_0 | 59.4 | 488 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla P40 | Vulkan | Llama 2 7B, Q4_0 | 59.2 | 484 | llama.cpp discussion 10879 read 2026-10-20 |
| GB10 | Vulkan | Llama 2 7B, Q4_0 | 58.5 | 2790 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla P100 | CUDA | Llama 2 7B, Q4_0 | 58.4 | 761 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 7600 | Vulkan | Llama 2 7B, Q4_0 | 58.0 | 1167 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1660 Ti Mobile | Vulkan | Llama 2 7B, Q4_0 | 57.3 | 514 | llama.cpp discussion 10879 read 2026-10-20 |
| DGX Spark | CUDA | Llama 2 7B, Q4_0 | 57.2 | 3062 | llama.cpp discussion 15013 read 2026-10-10 |
| DGX Spark | CUDA | Llama 2 7B, Q4_0 | 56.7 | 3661 | llama.cpp discussion 15013 read 2026-10-10 |
| GTX 1660 Ti Mobile | Vulkan | Llama 2 7B, Q4_0 | 56.6 | 512 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7600 XT | Vulkan | Llama 2 7B, Q4_0 | 56.1 | 1024 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD FirePro S9300 x2 | Vulkan | Llama 2 7B, Q4_0 | 55.6 | 243 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla P40 | CUDA | Llama 2 7B, Q4_0 | 54.7 | 1007 | llama.cpp discussion 15013 read 2026-10-10 |
| Instinct MI25 | Vulkan | Llama 2 7B, Q4_0 | 54.7 | 439 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen AI Max+ 300 Series | Vulkan | Llama 2 7B, Q4_0 | 54.4 | 1704 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 6600 XT | Vulkan | Llama 2 7B, Q4_0 | 53.9 | 575 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla P40 | CUDA | Llama 2 7B, Q4_0 | 53.7 | 1080 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 6600 | Vulkan | Llama 2 7B, Q4_0 | 53.2 | 805 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Arc A770 | Vulkan | Llama 2 7B, Q4_0 | 53.1 | 1120 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7600 XT | Vulkan | Llama 2 7B, Q4_0 | 53.0 | 841 | llama.cpp discussion 10879 read 2026-10-20 |
| RX Vega 56 | Vulkan | Llama 2 7B, Q4_0 | 52.7 | 429 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Arc A770 | Vulkan | Llama 2 7B, Q4_0 | 52.6 | 1074 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD FirePro S9300 x2 | Vulkan | Llama 2 7B, Q4_0 | 51.9 | 247 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen AI Max+ 300 Series | Vulkan | Llama 2 7B, Q4_0 | 51.8 | 1487 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 2000 Ada | CUDA | Llama 2 7B, Q4_0 | 50.7 | 2250 | llama.cpp discussion 15013 read 2026-10-10 |
| RTX 2000 Ada | CUDA | Llama 2 7B, Q4_0 | 50.6 | 1956 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 6600 | Vulkan | Llama 2 7B, Q4_0 | 50.6 | 758 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Arc B570 | Vulkan | Llama 2 7B, Q4_0 | 50.5 | 289 | llama.cpp discussion 10879 read 2026-10-20 |
| RX Vega 56 | Vulkan | Llama 2 7B, Q4_0 | 50.2 | 440 | llama.cpp discussion 10879 read 2026-10-20 |
| AI Max+ 395 | ROCm (HIP) | Llama 2 7B, Q4_0 | 50.0 | 911 | llama.cpp discussion 15021 read 2026-10-10 |
| AI Max+ 395 | ROCm (HIP) | Llama 2 7B, Q4_0 | 49.9 | 1004 | llama.cpp discussion 15021 read 2026-10-10 |
| Intel Arc B570 | Vulkan | Llama 2 7B, Q4_0 | 49.6 | 914 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 6800M | Vulkan | Llama 2 7B, Q4_0 | 49.1 | 784 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 3060 Mobile | Vulkan | Llama 2 7B, Q4_0 | 49.0 | 1060 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon 8060S | ROCm (HIP) | Llama 2 7B, Q4_0 | 49.0 | 366 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 6800M | Vulkan | Llama 2 7B, Q4_0 | 48.7 | 862 | llama.cpp discussion 10879 read 2026-10-20 |
| P104-100 | Vulkan | Llama 2 7B, Q4_0 | 48.6 | 325 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7600 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 48.6 | 1100 | llama.cpp discussion 15021 read 2026-10-10 |
| RX Vega 64 | ROCm (HIP) | Llama 2 7B, Q4_0 | 48.5 | 241 | llama.cpp discussion 15021 read 2026-10-10 |
| RX 6600M | Vulkan | Llama 2 7B, Q4_0 | 48.2 | 606 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon 8060S | ROCm (HIP) | Llama 2 7B, Q4_0 | 48.0 | 351 | llama.cpp discussion 15021 read 2026-10-10 |
| Intel Arc A770M | Vulkan | Llama 2 7B, Q4_0 | 47.7 | 876 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Arc Pro B50 | Vulkan | Llama 2 7B, Q4_0 | 47.7 | 1123 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7600 XT | ROCm (HIP) | Llama 2 7B, Q4_0 | 47.6 | 1199 | llama.cpp discussion 15021 read 2026-10-10 |
| Radeon Pro V340 | Vulkan | Llama 2 7B, Q4_0 | 47.5 | 360 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla T4 | CUDA | Llama 2 7B, Q4_0 | 46.4 | 1219 | llama.cpp discussion 15013 read 2026-10-10 |
| Intel Arc Pro B50 | Vulkan | Llama 2 7B, Q4_0 | 46.3 | 1186 | llama.cpp discussion 10879 read 2026-10-20 |
| P104-100 | Vulkan | Llama 2 7B, Q4_0 | 46.2 | 312 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX A2000 | Vulkan | Llama 2 7B, Q4_0 | 45.7 | 1362 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Arc A770M | Vulkan | Llama 2 7B, Q4_0 | 45.7 | 385 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX A2000 | Vulkan | Llama 2 7B, Q4_0 | 45.5 | 1245 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 7600M XT | Vulkan | Llama 2 7B, Q4_0 | 45.3 | 459 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro V340 | Vulkan | Llama 2 7B, Q4_0 | 45.2 | 375 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla T4 | CUDA | Llama 2 7B, Q4_0 | 44.0 | 1310 | llama.cpp discussion 15013 read 2026-10-10 |
| Intel Arc A750 | Vulkan | Llama 2 7B, Q4_0 | 44.0 | 303 | llama.cpp discussion 10879 read 2026-10-20 |
| RTX 4050 Laptop | CUDA | Llama 2 7B, Q4_0 | 43.7 | 1726 | llama.cpp discussion 15013 read 2026-10-10 |
| GTX 1070 Ti | Vulkan | Llama 2 7B, Q4_0 | 43.4 | 293 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 470 | Vulkan | Llama 2 7B, Q4_0 | 43.4 | 211 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1070 | Vulkan | Llama 2 7B, Q4_0 | 43.3 | 331 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1070 Ti | Vulkan | Llama 2 7B, Q4_0 | 42.9 | 298 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Arc A750 | Vulkan | Llama 2 7B, Q4_0 | 42.7 | 1076 | llama.cpp discussion 10879 read 2026-10-20 |
| RX Vega 64 | ROCm (HIP) | Llama 2 7B, Q4_0 | 42.5 | 153 | llama.cpp discussion 15021 read 2026-10-10 |
| RTX 4050 Mobile | Vulkan | Llama 2 7B, Q4_0 | 41.9 | 1154 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla M40 | Vulkan | Llama 2 7B, Q4_0 | 41.7 | 93 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1070 | Vulkan | Llama 2 7B, Q4_0 | 41.5 | 322 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1660 | CUDA | Llama 2 7B, Q4_0 | 41.4 | 154 | llama.cpp discussion 15013 read 2026-10-10 |
| GTX 1660 | CUDA | Llama 2 7B, Q4_0 | 41.4 | 149 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 570 | Vulkan | Llama 2 7B, Q4_0 | 40.2 | 229 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 470 | Vulkan | Llama 2 7B, Q4_0 | 40.1 | 207 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla M40 | CUDA | Llama 2 7B, Q4_0 | 40.0 | 290 | llama.cpp discussion 15013 read 2026-10-10 |
| Tesla M40 | Vulkan | Llama 2 7B, Q4_0 | 39.4 | 92 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 580 | Vulkan | Llama 2 7B, Q4_0 | 39.3 | 258 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla M40 | CUDA | Llama 2 7B, Q4_0 | 38.0 | 283 | llama.cpp discussion 15013 read 2026-10-10 |
| GTX 1070 Ti | CUDA | Llama 2 7B, Q4_0 | 37.9 | 791 | llama.cpp discussion 15013 read 2026-10-10 |
| GTX 1070 Ti | CUDA | Llama 2 7B, Q4_0 | 37.8 | 714 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 570 | Vulkan | Llama 2 7B, Q4_0 | 37.4 | 226 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 480 | Vulkan | Llama 2 7B, Q4_0 | 37.2 | 195 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro W5500 | Vulkan | Llama 2 7B, Q4_0 | 36.8 | 315 | llama.cpp discussion 10879 read 2026-10-20 |
| Apple M2 Ultra | Vulkan | Llama 2 7B, Q4_0 | 36.1 | 199 | llama.cpp discussion 10879 read 2026-10-20 |
| Jetson AGX Orin | CUDA | Llama 2 7B, Q4_0 | 35.9 | 1172 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 480 | Vulkan | Llama 2 7B, Q4_0 | 34.7 | 249 | llama.cpp discussion 10879 read 2026-10-20 |
| Apple M2 Ultra | Vulkan | Llama 2 7B, Q4_0 | 34.3 | 206 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 980 | Vulkan | Llama 2 7B, Q4_0 | 34.2 | 181 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 980 | Vulkan | Llama 2 7B, Q4_0 | 33.9 | 186 | llama.cpp discussion 10879 read 2026-10-20 |
| Jetson AGX Orin | CUDA | Llama 2 7B, Q4_0 | 33.6 | 991 | llama.cpp discussion 15013 read 2026-10-10 |
| Tesla P4 | CUDA | Llama 2 7B, Q4_0 | 33.3 | 515 | llama.cpp discussion 15013 read 2026-10-10 |
| Tesla P4 | CUDA | Llama 2 7B, Q4_0 | 33.1 | 530 | llama.cpp discussion 15013 read 2026-10-10 |
| P106-100 | Vulkan | Llama 2 7B, Q4_0 | 30.8 | 183 | llama.cpp discussion 10879 read 2026-10-20 |
| P106-100 | CUDA | Llama 2 7B, Q4_0 | 30.6 | 438 | llama.cpp discussion 15013 read 2026-10-10 |
| P106-100 | CUDA | Llama 2 7B, Q4_0 | 30.4 | 407 | llama.cpp discussion 15013 read 2026-10-10 |
| P106-100 | Vulkan | Llama 2 7B, Q4_0 | 29.8 | 184 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD FirePro W8100 | Vulkan | Llama 2 7B, Q4_0 | 29.5 | 155 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD FirePro W8100 | Vulkan | Llama 2 7B, Q4_0 | 29.3 | 141 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla P4 | Vulkan | Llama 2 7B, Q4_0 | 28.4 | 287 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1060 | CUDA | Llama 2 7B, Q4_0 | 28.2 | 446 | llama.cpp discussion 15013 read 2026-10-10 |
| Tesla P4 | Vulkan | Llama 2 7B, Q4_0 | 28.0 | 266 | llama.cpp discussion 10879 read 2026-10-20 |
| Quadro T1000 | CUDA | Llama 2 7B, Q4_0 | 27.8 | 79 | llama.cpp discussion 15013 read 2026-10-10 |
| RX 6500 XT | Vulkan | Llama 2 7B, Q4_0 | 27.8 | 255 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1060 | CUDA | Llama 2 7B, Q4_0 | 27.8 | 417 | llama.cpp discussion 15013 read 2026-10-10 |
| Apple M3 | Vulkan | Llama 2 7B, Q4_0 | 26.4 | 264 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD FirePro S10000 | Vulkan | Llama 2 7B, Q4_0 | 25.3 | 95 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Core Ultra 200 Series | Vulkan | Llama 2 7B, Q4_0 | 25.1 | 867 | llama.cpp discussion 10879 read 2026-10-20 |
| Qualcomm Snapdragon X Elite (12-core) | CPU | Llama 2 7B, Q4_0 | 24.8 | 178 | llama.cpp discussion 8336 read 2026-10-20 |
| Intel Core Ultra 300 Series | Vulkan | Llama 2 7B, Q4_0 | 24.4 | 1380 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Core Ultra 200 Series | Vulkan | Llama 2 7B, Q4_0 | 24.4 | 865 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen AI 300 Series | Vulkan | Llama 2 7B, Q4_0 | 23.9 | 533 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Core Ultra 300 Series | Vulkan | Llama 2 7B, Q4_0 | 23.9 | 1332 | llama.cpp discussion 10879 read 2026-10-20 |
| Quadro P2000 | Vulkan | Llama 2 7B, Q4_0 | 23.8 | 182 | llama.cpp discussion 10879 read 2026-10-20 |
| Quadro P2000 | CUDA | Llama 2 7B, Q4_0 | 23.8 | 312 | llama.cpp discussion 15013 read 2026-10-10 |
| Quadro P2000 | CUDA | Llama 2 7B, Q4_0 | 23.6 | 309 | llama.cpp discussion 15013 read 2026-10-10 |
| Quadro P2000 | Vulkan | Llama 2 7B, Q4_0 | 23.1 | 170 | llama.cpp discussion 10879 read 2026-10-20 |
| Qualcomm Snapdragon X Elite (12-core) | CPU | Llama 2 7B, Q4_0 | 22.6 | 177 | llama.cpp discussion 8336 read 2026-10-20 |
| AMD Ryzen AI 300 Series | Vulkan | Llama 2 7B, Q4_0 | 22.4 | 479 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen 8000 Series | Vulkan | Llama 2 7B, Q4_0 | 21.3 | 311 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen 6000 Series | Vulkan | Llama 2 7B, Q4_0 | 21.3 | 241 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen 6000 Series | Vulkan | Llama 2 7B, Q4_0 | 21.1 | 278 | llama.cpp discussion 10879 read 2026-10-20 |
| Apple M2 Pro | Vulkan | Llama 2 7B, Q4_0 | 21.0 | 59 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1050 Ti | Vulkan | Llama 2 7B, Q4_0 | 21.0 | 136 | llama.cpp discussion 10879 read 2026-10-20 |
| Apple M2 Pro | Vulkan | Llama 2 7B, Q4_0 | 20.9 | 63 | llama.cpp discussion 10879 read 2026-10-20 |
| Qualcomm Snapdragon X Elite (12-core) | CPU | Llama 2 7B, Q4_0 | 20.7 | 67 | llama.cpp discussion 8273 read 2026-10-20 |
| AMD Ryzen 7000 Series | Vulkan | Llama 2 7B, Q4_0 | 20.1 | 313 | llama.cpp discussion 10879 read 2026-10-20 |
| GTX 1050 Ti | Vulkan | Llama 2 7B, Q4_0 | 20.1 | 128 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen 7000 Series | Vulkan | Llama 2 7B, Q4_0 | 19.9 | 282 | llama.cpp discussion 10879 read 2026-10-20 |
| Qualcomm Snapdragon X Elite (12-core) | CPU | Llama 2 7B, Q4_0 | 19.9 | 166 | llama.cpp discussion 8273 read 2026-10-20 |
| Qualcomm Snapdragon X Elite (12-core) | CPU | Llama 2 7B, Q4_0 | 19.8 | 63 | llama.cpp discussion 8273 read 2026-10-20 |
| AMD Ryzen 8000 Series | Vulkan | Llama 2 7B, Q4_0 | 19.8 | 267 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen AI 400 Series | Vulkan | Llama 2 7B, Q4_0 | 19.0 | 150 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen Z1 Extreme | Vulkan | Llama 2 7B, Q4_0 | 18.8 | 199 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Steam Deck | Vulkan | Llama 2 7B, Q4_0 | 18.4 | 164 | llama.cpp discussion 10879 read 2026-10-20 |
| Qualcomm Snapdragon X Plus X1P-64-100 | CPU | Llama 2 7B, Q4_0 | 18.2 | 59 | llama.cpp discussion 8273 read 2026-10-20 |
| Qualcomm Adreno X1-85 GPU (Snapdragon X Elite) | OpenCL | Llama 2 7B, Q4_0 | 17.9 | 101 | llama.cpp discussion 8336 read 2026-10-20 |
| AMD Ryzen AI 400 Series | Vulkan | Llama 2 7B, Q4_0 | 17.7 | 131 | llama.cpp discussion 10879 read 2026-10-20 |
| Qualcomm Adreno X1-85 GPU (Snapdragon X Elite) | OpenCL | Llama 2 7B, Q4_0 | 17.6 | 102 | llama.cpp discussion 8273 read 2026-10-20 |
| AMD Steam Deck | Vulkan | Llama 2 7B, Q4_0 | 17.5 | 144 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD FirePro D700 | Vulkan | Llama 2 7B, Q4_0 | 16.6 | 70 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon Pro WX 4100 | Vulkan | Llama 2 7B, Q4_0 | 16.6 | 76 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon 880M | ROCm (HIP) | Llama 2 7B, Q4_0 | 16.2 | 213 | llama.cpp discussion 15021 read 2026-10-10 |
| Radeon Pro WX 4100 | Vulkan | Llama 2 7B, Q4_0 | 16.1 | 79 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla K80 | CUDA | Llama 2 7B, Q4_0 | 14.3 | 133 | llama.cpp discussion 15013 read 2026-10-10 |
| Quadro P1000 | CUDA | Llama 2 7B, Q4_0 | 14.0 | 183 | llama.cpp discussion 15013 read 2026-10-10 |
| Tesla K80 | CUDA | Llama 2 7B, Q4_0 | 13.8 | 133 | llama.cpp discussion 15013 read 2026-10-10 |
| Qualcomm Snapdragon X Plus X1P-64-100 | CPU | Llama 2 7B, Q4_0 | 13.7 | 62 | llama.cpp discussion 8273 read 2026-10-20 |
| Quadro P1000 | CUDA | Llama 2 7B, Q4_0 | 13.7 | 174 | llama.cpp discussion 15013 read 2026-10-10 |
| Qualcomm Snapdragon X Plus X1P-64-100 | CPU | Llama 2 7B, Q8_0 | 13.5 | 59 | llama.cpp discussion 8273 read 2026-10-20 |
| Apple M2 | Vulkan | Llama 2 7B, Q4_0 | 13.5 | 51 | llama.cpp discussion 10879 read 2026-10-20 |
| Radeon 880M | ROCm (HIP) | Llama 2 7B, Q4_0 | 13.0 | 163 | llama.cpp discussion 15021 read 2026-10-10 |
| Apple M1 | Vulkan | Llama 2 7B, Q4_0 | 12.8 | 36 | llama.cpp discussion 10879 read 2026-10-20 |
| Qualcomm Snapdragon X Elite (12-core) | CPU | Llama 2 7B, Q8_0 | 12.7 | 64 | llama.cpp discussion 8273 read 2026-10-20 |
| Apple M1 | Vulkan | Llama 2 7B, Q4_0 | 12.5 | 38 | llama.cpp discussion 10879 read 2026-10-20 |
| Apple M2 | Vulkan | Llama 2 7B, Q4_0 | 12.2 | 47 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen 5000 Series | Vulkan | Llama 2 7B, Q4_0 | 11.0 | 91 | llama.cpp discussion 10879 read 2026-10-20 |
| Qualcomm Adreno X1-85 GPU (Snapdragon X Elite) | OpenCL | Llama 2 7B, Q4_0 | 10.9 | 140 | llama.cpp discussion 8273 read 2026-10-20 |
| AMD Ryzen 5000 Series | Vulkan | Llama 2 7B, Q4_0 | 10.8 | 79 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Core 1100 Series | Vulkan | Llama 2 7B, Q4_0 | 10.6 | 175 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Core 1100 Series | Vulkan | Llama 2 7B, Q4_0 | 10.4 | 187 | llama.cpp discussion 10879 read 2026-10-20 |
| RX 550 | Vulkan | Llama 2 7B, Q4_0 | 10.2 | 53 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla K40 | Vulkan | Llama 2 7B, Q4_0 | 9.9 | 64 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen 4000 Series | Vulkan | Llama 2 7B, Q4_0 | 9.9 | 113 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen 4000 Series | Vulkan | Llama 2 7B, Q4_0 | 9.6 | 104 | llama.cpp discussion 10879 read 2026-10-20 |
| Qualcomm Snapdragon X Plus X1P-64-100 | CPU | Llama 2 7B, Q8_0 | 9.6 | 62 | llama.cpp discussion 8273 read 2026-10-20 |
| Tesla K80 | Vulkan | Llama 2 7B, Q4_0 | 9.5 | 88 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla K80 | Vulkan | Llama 2 7B, Q4_0 | 9.4 | 89 | llama.cpp discussion 10879 read 2026-10-20 |
| Tesla K40 | Vulkan | Llama 2 7B, Q4_0 | 9.3 | 64 | llama.cpp discussion 10879 read 2026-10-20 |
| MediaTek Dimensity 9400 | Vulkan | Llama 2 7B, Q4_0 | 8.9 | 38 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen 3000 Series | Vulkan | Llama 2 7B, Q4_0 | 8.5 | 49 | llama.cpp discussion 10879 read 2026-10-20 |
| AMD Ryzen 5 3000 Series | Vulkan | Llama 2 7B, Q4_0 | 8.5 | 47 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Core Ultra 100 Series | Vulkan | Llama 2 7B, Q4_0 | 8.2 | 186 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Core Ultra 100 Series | Vulkan | Llama 2 7B, Q4_0 | 7.8 | 78 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Core 1000 Series | Vulkan | Llama 2 7B, Q4_0 | 4.2 | 26 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Core 8000 Series | Vulkan | Llama 2 7B, Q4_0 | 3.4 | 25 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel Core 8000 Series | Vulkan | Llama 2 7B, Q4_0 | 3.4 | 26 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel N150 | Vulkan | Llama 2 7B, Q4_0 | 2.9 | 29 | llama.cpp discussion 10879 read 2026-10-20 |
| Intel N150 | Vulkan | Llama 2 7B, Q4_0 | 2.9 | 26 | llama.cpp discussion 10879 read 2026-10-20 |
Rendering speed (Blender Open Data)
Median Cycles render score per chip from results people submitted to Blender Open Data (CC0 data). Higher is faster. Processors are scored on the processor; graphics cards on their own rendering backend. Grouped by what the render ran on, fastest first. Scores from different Blender versions are not directly comparable, so each row names its version.
| Chip | Rendered on | Score | Source |
|---|---|---|---|
| AMD Radeon RX 7800M | AMD HIP, Blender 5.2.0 | 2259 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 6800M | AMD HIP, Blender 5.1.1 | 1658 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 6850M XT | AMD HIP, Blender 4.4.0 | 1395 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 7600M XT | AMD HIP, Blender 5.2.0 | 1092 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 6800S | AMD HIP, Blender 5.2.0 | 1081 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 7600M | AMD HIP, Blender 4.2.0 | 1060 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 7600S | AMD HIP, Blender 5.2.0 | 1037 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 6700M | AMD HIP, Blender 5.1.1 | 807 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 7700S | AMD HIP, Blender 5.1.1 | 758 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 6650M | AMD HIP, Blender 4.5.0 | 733 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 6700S | AMD HIP, Blender 5.0.0 | 586 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 6600M | AMD HIP, Blender 5.1.1 | 560 | Blender Open Data read 2026-10-22 |
| AMD Radeon RX 6550M | AMD HIP, Blender 4.2.0 | 381 | Blender Open Data read 2026-10-22 |
| Apple M5 Max (40-core GPU) | Apple Metal, Blender 5.2.0 | 8156 | Blender Open Data read 2026-10-22 |
| Apple M5 Max (32-core GPU) | Apple Metal, Blender 5.2.0 | 6462 | Blender Open Data read 2026-10-22 |
| Apple M4 Max (16-core CPU, 40-core GPU) | Apple Metal, Blender 5.2.0 | 5874 | Blender Open Data read 2026-10-22 |
| Apple M4 Max (14-core CPU, 32-core GPU) | Apple Metal, Blender 5.2.0 | 4881 | Blender Open Data read 2026-10-22 |
| Apple M3 Max (16-core CPU, 40-core GPU) | Apple Metal, Blender 5.2.0 | 4432 | Blender Open Data read 2026-10-22 |
| Apple M3 Max (14-core CPU, 30-core GPU) | Apple Metal, Blender 5.2.0 | 3613 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 5090 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 8074 | Blender Open Data read 2026-10-22 |
| NVIDIA RTX PRO 5000 Blackwell Generation Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 7445 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 5080 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 6834 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 4090 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 6291 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 4080 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 5342 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 5070 Ti Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 5087 | Blender Open Data read 2026-10-22 |
| NVIDIA RTX PRO 3000 Blackwell Generation Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 4218 | Blender Open Data read 2026-10-22 |
| NVIDIA RTX PRO 4000 Blackwell Generation Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 3712 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 4070 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 3426 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 5060 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 3258 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 3080 Ti Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 3246 | Blender Open Data read 2026-10-22 |
| NVIDIA RTX PRO 2000 Blackwell Generation Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 2967 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 4060 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 2905 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 3070 Ti Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 2637 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 5050 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 2571 | Blender Open Data read 2026-10-22 |
| NVIDIA RTX PRO 1000 Blackwell Generation Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 2495 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 3070 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 2471 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 4050 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 2416 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 3060 Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 1957 | Blender Open Data read 2026-10-22 |
| NVIDIA RTX PRO 500 Blackwell Generation Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 1839 | Blender Open Data read 2026-10-22 |
| NVIDIA GeForce RTX 3050 Ti Laptop GPU | NVIDIA OptiX, Blender 5.2.0 | 1086 | Blender Open Data read 2026-10-22 |
| Apple M5 Ultra | processor, Blender 5.2.0 | 1134 | Blender Open Data read 2026-10-22 |
| Apple M3 Ultra | processor, Blender 5.2.0 | 679 | Blender Open Data read 2026-10-22 |
| Apple M2 Ultra | processor, Blender 5.2.0 | 501 | Blender Open Data read 2026-10-22 |
| Apple M5 Pro | processor, Blender 5.2.0 | 500 | Blender Open Data read 2026-10-22 |
| Qualcomm Snapdragon X2 Elite Extreme X2E-96-100 | processor, Blender 5.2.0 | 456 | Blender Open Data read 2026-10-22 |
| Qualcomm Snapdragon X2 Elite Extreme X2E-94-100 | processor, Blender 5.2.0 | 452 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI Max+ 392 | processor, Blender 5.2.0 | 430 | Blender Open Data read 2026-10-22 |
| Apple M1 Ultra | processor, Blender 5.2.0 | 406 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI Max+ 395 | processor, Blender 5.2.0 | 391 | Blender Open Data read 2026-10-22 |
| Apple M4 Pro | processor, Blender 5.2.0 | 362 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI Max 390 | processor, Blender 4.5.0 | 352 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 9 HX 470 | processor, Blender 5.2.0 | 340 | Blender Open Data read 2026-10-22 |
| Apple M6 | processor, Blender 5.2.0 | 335 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 9 HX 375 | processor, Blender 5.2.0 | 330 | Blender Open Data read 2026-10-22 |
| Qualcomm Snapdragon X2 Elite X2E-88-100 | processor, Blender 5.1.1 | 326 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 9 465 | processor, Blender 5.2.0 | 302 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI Max 385 | processor, Blender 4.5.0 | 302 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 9 HX 370 | processor, Blender 5.2.0 | 293 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI Max+ 388 | processor, Blender 5.2.0 | 288 | Blender Open Data read 2026-10-22 |
| Qualcomm Snapdragon X Elite X1E-78-100 | processor, Blender 5.1.0 | 284 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 9 386H (Series 3) | processor, Blender 5.2.0 | 284 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 356H (Series 3) | processor, Blender 5.2.0 | 272 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 9 285H (Series 2) | processor, Blender 5.2.0 | 266 | Blender Open Data read 2026-10-22 |
| Qualcomm Snapdragon X Elite X1E-84-100 | processor, Blender 4.5.0 | 261 | Blender Open Data read 2026-10-22 |
| Qualcomm Snapdragon X2 Elite X2E-90-100 | processor, Blender 5.2.0 | 261 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 366H (Series 3) | processor, Blender 5.2.0 | 248 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra X7 358H (Series 3) | processor, Blender 5.2.0 | 244 | Blender Open Data read 2026-10-22 |
| Apple M2 Max | processor, Blender 5.2.0 | 244 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra X9 388H (Series 3) | processor, Blender 5.2.0 | 243 | Blender Open Data read 2026-10-22 |
| Apple M5 | processor, Blender 5.2.0 | 240 | Blender Open Data read 2026-10-22 |
| AMD Ryzen 7 8845HS | processor, Blender 5.2.0 | 236 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 7 450 | processor, Blender 5.1.1 | 230 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 7 350 | processor, Blender 5.2.0 | 222 | Blender Open Data read 2026-10-22 |
| AMD Ryzen 7 7840HS | processor, Blender 5.2.0 | 217 | Blender Open Data read 2026-10-22 |
| Qualcomm Snapdragon X Plus X1P-64-100 | processor, Blender 5.2.0 | 216 | Blender Open Data read 2026-10-22 |
| AMD Ryzen 9 8945HS | processor, Blender 5.2.0 | 215 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 165H (Series 1) | processor, Blender 5.2.0 | 214 | Blender Open Data read 2026-10-22 |
| Apple M3 Pro | processor, Blender 5.2.0 | 214 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 338H (Series 3) | processor, Blender 5.2.0 | 213 | Blender Open Data read 2026-10-22 |
| Apple M4 | processor, Blender 5.2.0 | 213 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 265H (Series 2) | processor, Blender 5.1.1 | 212 | Blender Open Data read 2026-10-22 |
| AMD Ryzen 9 7940HS | processor, Blender 5.2.0 | 208 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 255H (Series 2) | processor, Blender 5.2.0 | 207 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 9 365 | processor, Blender 5.2.0 | 207 | Blender Open Data read 2026-10-22 |
| Apple M1 Max | processor, Blender 5.2.0 | 201 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra X7 368H (Series 3) | processor, Blender 5.2.0 | 198 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 235H (Series 2) | processor, Blender 4.5.0 | 191 | Blender Open Data read 2026-10-22 |
| Apple M2 Pro | processor, Blender 5.2.0 | 191 | Blender Open Data read 2026-10-22 |
| AMD Ryzen 7 7840U | processor, Blender 5.2.0 | 191 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 9 185H (Series 1) | processor, Blender 5.2.0 | 188 | Blender Open Data read 2026-10-22 |
| Apple M1 Pro | processor, Blender 5.2.0 | 185 | Blender Open Data read 2026-10-22 |
| Qualcomm Snapdragon X Plus X1P-42-100 | processor, Blender 5.1.1 | 182 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 225H (Series 2) | processor, Blender 5.2.0 | 181 | Blender Open Data read 2026-10-22 |
| AMD Ryzen 7 8840U | processor, Blender 5.2.0 | 179 | Blender Open Data read 2026-10-22 |
| AMD Ryzen 5 7640HS | processor, Blender 5.2.0 | 175 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 135H (Series 1) | processor, Blender 5.2.0 | 172 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 5 340 | processor, Blender 5.2.0 | 170 | Blender Open Data read 2026-10-22 |
| Qualcomm Snapdragon X Elite X1E-80-100 | processor, Blender 5.1.1 | 164 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 155H (Series 1) | processor, Blender 5.2.0 | 164 | Blender Open Data read 2026-10-22 |
| Qualcomm Snapdragon X2 Plus X2P-42-100 | processor, Blender 5.2.0 | 164 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 268V (Series 2) | processor, Blender 5.2.0 | 156 | Blender Open Data read 2026-10-22 |
| Apple M3 | processor, Blender 5.2.0 | 151 | Blender Open Data read 2026-10-22 |
| AMD Ryzen 5 8645HS | processor, Blender 5.2.0 | 148 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 165U (Series 1) | processor, Blender 4.4.0 | 143 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 125H (Series 1) | processor, Blender 5.2.0 | 138 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 266V (Series 2) | processor, Blender 5.1.1 | 132 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 258V (Series 2) | processor, Blender 5.2.0 | 130 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 255U (Series 2) | processor, Blender 5.2.0 | 130 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 7 445 | processor, Blender 5.2.0 | 127 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 236V (Series 2) | processor, Blender 4.5.0 | 127 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 238V (Series 2) | processor, Blender 5.2.0 | 122 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 9 288V (Series 2) | processor, Blender 5.2.0 | 121 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 235U (Series 2) | processor, Blender 5.1.1 | 115 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 265U (Series 2) | processor, Blender 4.5.0 | 115 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 135U (Series 1) | processor, Blender 4.5.0 | 112 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 226V (Series 2) | processor, Blender 5.2.0 | 110 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 125U (Series 1) | processor, Blender 5.2.0 | 109 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 256V (Series 2) | processor, Blender 5.2.0 | 107 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 5 430 | processor, Blender 5.1.1 | 103 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 225U (Series 2) | processor, Blender 5.2.0 | 103 | Blender Open Data read 2026-10-22 |
| Apple M2 | processor, Blender 5.2.0 | 100 | Blender Open Data read 2026-10-22 |
| Apple M1 | processor, Blender 5.2.0 | 99 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 7 155U (Series 1) | processor, Blender 5.1.0 | 98 | Blender Open Data read 2026-10-22 |
| AMD Ryzen AI 5 330 | processor, Blender 5.2.0 | 88 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra X9 378H (Series 3) | processor, Blender 5.2.0 | 80 | Blender Open Data read 2026-10-22 |
| Intel Core Ultra 5 228V (Series 2) | processor, Blender 5.1.1 | 76 | Blender Open Data read 2026-10-22 |
LocalScore results
LocalScore runs its own mixed test (prompts of 16 to 4,096 tokens) on a quantized Llama 3.1 8B and Qwen2.5 14B. These averages are not comparable with the llama-bench rows above. Sorted by model, then fastest first.
| Chip or card | Model and size | Writing (tokens per second) | Reading (tokens per second) | Source |
|---|---|---|---|---|
| NVIDIA RTX 6000 Ada Generation | Meta Llama 3.1 8B Instruct, Q4_K_M | 121.0 | 6808 | LocalScore read 2026-10-22 |
| NVIDIA H100 PCIe | Meta Llama 3.1 8B Instruct, Q4_K_M | 120.0 | 5202 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4090 | Meta Llama 3.1 8B Instruct, Q4_K_M | 119.6 | 7594 | LocalScore read 2026-10-22 |
| NVIDIA A100-SXM4-80GB | Meta Llama 3.1 8B Instruct, Q4_K_M | 110.1 | 3571 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3090 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 109.8 | 4024 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5090 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 105.1 | 4463 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3090 | Meta Llama 3.1 8B Instruct, Q4_K_M | 96.3 | 3550 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4090 | Meta Llama 3.1 8B Instruct, Q4_K_M | 91.0 | 6697 | LocalScore read 2026-10-22 |
| Tesla V100-SXM2-16GB | Meta Llama 3.1 8B Instruct, Q4_K_M | 90.8 | 2256 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5090 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 90.6 | 3829 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3080 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 90.5 | 3515 | LocalScore read 2026-10-22 |
| NVIDIA RTX A6000 | Meta Llama 3.1 8B Instruct, Q4_K_M | 90.5 | 3447 | LocalScore read 2026-10-22 |
| NVIDIA RTX A5000 | Meta Llama 3.1 8B Instruct, Q4_K_M | 89.0 | 2891 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4080 | Meta Llama 3.1 8B Instruct, Q4_K_M | 88.3 | 4792 | LocalScore read 2026-10-22 |
| Tesla V100-SXM2-32GB | Meta Llama 3.1 8B Instruct, Q4_K_M | 88.2 | 2177 | LocalScore read 2026-10-22 |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | Meta Llama 3.1 8B Instruct, Q4_K_M | 85.8 | 6346 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3070 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 83.2 | 2509 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3080 | Meta Llama 3.1 8B Instruct, Q4_K_M | 81.7 | 3031 | LocalScore read 2026-10-22 |
| NVIDIA A40 | Meta Llama 3.1 8B Instruct, Q4_K_M | 80.6 | 3270 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3080 | Meta Llama 3.1 8B Instruct, Q4_K_M | 80.6 | 3019 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4080 SUPER | Meta Llama 3.1 8B Instruct, Q4_K_M | 78.7 | 4979 | LocalScore read 2026-10-22 |
| NVIDIA RTX A4500 | Meta Llama 3.1 8B Instruct, Q4_K_M | 77.9 | 2498 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4070 | Meta Llama 3.1 8B Instruct, Q4_K_M | 76.3 | 3192 | LocalScore read 2026-10-22 |
| Quadro RTX 8000 | Meta Llama 3.1 8B Instruct, Q4_K_M | 74.0 | 1677 | LocalScore read 2026-10-22 |
| NVIDIA TITAN V | Meta Llama 3.1 8B Instruct, Q4_K_M | 73.3 | 1823 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 2080 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 72.9 | 2027 | LocalScore read 2026-10-22 |
| Apple M5 Max 6P+12E+40GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 71.2 | 718 | LocalScore read 2026-10-22 |
| NVIDIA TITAN RTX | Meta Llama 3.1 8B Instruct, Q4_K_M | 71.2 | 2226 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5090 | Meta Llama 3.1 8B Instruct, Q4_K_M | 66.3 | 6297 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5070 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 65.5 | 3657 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5070 Ti Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 65.4 | 2874 | LocalScore read 2026-10-22 |
| Quadro RTX 6000 | Meta Llama 3.1 8B Instruct, Q4_K_M | 64.9 | 1891 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4070 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 64.2 | 3769 | LocalScore read 2026-10-22 |
| Apple M3 Ultra 24P+8E+80GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 63.3 | 1062 | LocalScore read 2026-10-22 |
| Apple M3 Ultra 24P+8E+80GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 62.7 | 1109 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3070 Ti Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 62.4 | 1910 | LocalScore read 2026-10-22 |
| Apple M5 Max 6P+12E+32GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 62.2 | 637 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3060 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 60.2 | 1839 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 2080 SUPER | Meta Llama 3.1 8B Instruct, Q4_K_M | 59.8 | 1693 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3070 | Meta Llama 3.1 8B Instruct, Q4_K_M | 59.6 | 2006 | LocalScore read 2026-10-22 |
| Apple M2 Ultra 16P+8E+60GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 59.4 | 706 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5060 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 59.0 | 2365 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 2080 | Meta Llama 3.1 8B Instruct, Q4_K_M | 58.4 | 1546 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5070 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 57.7 | 2030 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 2070 SUPER | Meta Llama 3.1 8B Instruct, Q4_K_M | 57.6 | 1444 | LocalScore read 2026-10-22 |
| NVIDIA RTX A4000 | Meta Llama 3.1 8B Instruct, Q4_K_M | 56.5 | 1931 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5070 | Meta Llama 3.1 8B Instruct, Q4_K_M | 55.9 | 2927 | LocalScore read 2026-10-22 |
| Apple M4 Max 12P+4E+40GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 54.5 | 656 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4080 SUPER | Meta Llama 3.1 8B Instruct, Q4_K_M | 54.4 | 4461 | LocalScore read 2026-10-22 |
| Apple M1 Ultra 16P+4E+64GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 54.3 | 668 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4070 Ti SUPER | Meta Llama 3.1 8B Instruct, Q4_K_M | 53.9 | 3561 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3080 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 53.9 | 2052 | LocalScore read 2026-10-22 |
| NVIDIA RTX 4000 Ada Generation | Meta Llama 3.1 8B Instruct, Q4_K_M | 53.5 | 2608 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4070 SUPER | Meta Llama 3.1 8B Instruct, Q4_K_M | 53.4 | 3336 | LocalScore read 2026-10-22 |
| Apple M4 Max 12P+4E+40GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 52.9 | 632 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 2070 | Meta Llama 3.1 8B Instruct, Q4_K_M | 52.4 | 1252 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3080 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 52.3 | 1885 | LocalScore read 2026-10-22 |
| NVIDIA RTX 6000 Ada Generation | Meta Llama 3.1 8B Instruct, Q4_K_M | 51.3 | 5488 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3060 | Meta Llama 3.1 8B Instruct, Q4_K_M | 51.3 | 1483 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5060 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 51.2 | 2314 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 2060 SUPER | Meta Llama 3.1 8B Instruct, Q4_K_M | 49.9 | 1205 | LocalScore read 2026-10-22 |
| Apple M1 Ultra 16P+4E+48GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 48.9 | 534 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4060 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 48.3 | 2233 | LocalScore read 2026-10-22 |
| Apple M4 Max 12P+4E+40GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 47.9 | 566 | LocalScore read 2026-10-22 |
| Apple M2 Max 8P+4E+38GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 46.4 | 484 | LocalScore read 2026-10-22 |
| NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | Meta Llama 3.1 8B Instruct, Q4_K_M | 46.2 | 5544 | LocalScore read 2026-10-22 |
| NVIDIA L40S | Meta Llama 3.1 8B Instruct, Q4_K_M | 45.9 | 5328 | LocalScore read 2026-10-22 |
| Apple M3 Max 12P+4E+40GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 45.8 | 583 | LocalScore read 2026-10-22 |
| Apple M4 Max 10P+4E+32GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 45.5 | 540 | LocalScore read 2026-10-22 |
| NVIDIA RTX A5000 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 45.5 | 1497 | LocalScore read 2026-10-22 |
| Radeon RX 7900 XTX | Meta Llama 3.1 8B Instruct, Q4_K_M | 45.3 | 885 | LocalScore read 2026-10-22 |
| NVIDIA L40 | Meta Llama 3.1 8B Instruct, Q4_K_M | 45.1 | 5065 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5080 | Meta Llama 3.1 8B Instruct, Q4_K_M | 44.9 | 4428 | LocalScore read 2026-10-22 |
| Apple M5 Pro 6P+12E+20GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 44.9 | 411 | LocalScore read 2026-10-22 |
| Apple M2 Max 8P+4E+38GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 44.7 | 474 | LocalScore read 2026-10-22 |
| Apple M5 Pro 6P+12E+20GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 44.5 | 410 | LocalScore read 2026-10-22 |
| NVIDIA RTX 4000 SFF Ada Generation | Meta Llama 3.1 8B Instruct, Q4_K_M | 44.4 | 2004 | LocalScore read 2026-10-22 |
| Apple M5 Pro 5P+10E+16GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 43.7 | 341 | LocalScore read 2026-10-22 |
| Apple M5 Pro 5P+10E+16GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 43.2 | 336 | LocalScore read 2026-10-22 |
| NVIDIA L4 | Meta Llama 3.1 8B Instruct, Q4_K_M | 42.9 | 2056 | LocalScore read 2026-10-22 |
| AMD Radeon RX 6900 XT | Meta Llama 3.1 8B Instruct, Q4_K_M | 42.4 | 1082 | LocalScore read 2026-10-22 |
| NVIDIA H100 80GB HBM3 | Meta Llama 3.1 8B Instruct, Q4_K_M | 42.4 | 4687 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3070 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 41.9 | 1176 | LocalScore read 2026-10-22 |
| NVIDIA RTX 3500 Ada Generation Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 41.8 | 2445 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4070 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 40.9 | 2131 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3060 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 40.8 | 1253 | LocalScore read 2026-10-22 |
| NVIDIA RTX PRO 2000 Blackwell Generation Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 40.4 | 1453 | LocalScore read 2026-10-22 |
| Quadro RTX 4000 | Meta Llama 3.1 8B Instruct, Q4_K_M | 40.0 | 992 | LocalScore read 2026-10-22 |
| AMD Radeon RX 7900 XTX | Meta Llama 3.1 8B Instruct, Q4_K_M | 39.2 | 865 | LocalScore read 2026-10-22 |
| AMD Radeon RX 7800 XT | Meta Llama 3.1 8B Instruct, Q4_K_M | 38.9 | 524 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4060 | Meta Llama 3.1 8B Instruct, Q4_K_M | 38.3 | 1544 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4060 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 38.3 | 2087 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 2060 | Meta Llama 3.1 8B Instruct, Q4_K_M | 38.1 | 960 | LocalScore read 2026-10-22 |
| Apple M3 Max 10P+4E+30GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 37.7 | 457 | LocalScore read 2026-10-22 |
| AMD Radeon RX 6800 XT | Meta Llama 3.1 8B Instruct, Q4_K_M | 37.6 | 951 | LocalScore read 2026-10-22 |
| NVIDIA RTX 2000 Ada Generation | Meta Llama 3.1 8B Instruct, Q4_K_M | 37.5 | 1357 | LocalScore read 2026-10-22 |
| Apple M3 Max 10P+4E+30GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 37.5 | 444 | LocalScore read 2026-10-22 |
| NVIDIA A100 80GB PCIe | Meta Llama 3.1 8B Instruct, Q4_K_M | 37.1 | 2601 | LocalScore read 2026-10-22 |
| NVIDIA H100 NVL | Meta Llama 3.1 8B Instruct, Q4_K_M | 37.1 | 4299 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3060 | Meta Llama 3.1 8B Instruct, Q4_K_M | 37.0 | 1340 | LocalScore read 2026-10-22 |
| AMD Radeon RX 7800 XT | Meta Llama 3.1 8B Instruct, Q4_K_M | 36.7 | 403 | LocalScore read 2026-10-22 |
| Apple M1 Max 8P+2E+32GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 36.7 | 372 | LocalScore read 2026-10-22 |
| NVIDIA A10 | Meta Llama 3.1 8B Instruct, Q4_K_M | 36.5 | 2085 | LocalScore read 2026-10-22 |
| Apple M2 Max 8P+4E+30GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 36.2 | 362 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4060 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 36.0 | 1365 | LocalScore read 2026-10-22 |
| Apple M1 Max 8P+2E+32GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 35.9 | 366 | LocalScore read 2026-10-22 |
| NVIDIA RTX A2000 12GB | Meta Llama 3.1 8B Instruct, Q4_K_M | 35.7 | 976 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3050 OEM | Meta Llama 3.1 8B Instruct, Q4_K_M | 35.0 | 965 | LocalScore read 2026-10-22 |
| NVIDIA GB10 | Meta Llama 3.1 8B Instruct, Q4_K_M | 34.7 | 2010 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5070 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 34.7 | 2575 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3050 | Meta Llama 3.1 8B Instruct, Q4_K_M | 34.5 | 979 | LocalScore read 2026-10-22 |
| NVIDIA GB10 | Meta Llama 3.1 8B Instruct, Q4_K_M | 33.6 | 1986 | LocalScore read 2026-10-22 |
| Tesla P100-SXM2-16GB | Meta Llama 3.1 8B Instruct, Q4_K_M | 33.5 | 427 | LocalScore read 2026-10-22 |
| Apple M4 Pro 10P+4E+20GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 32.9 | 349 | LocalScore read 2026-10-22 |
| Apple M4 Pro 10P+4E+20GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 32.7 | 361 | LocalScore read 2026-10-22 |
| Apple M4 Pro 10P+4E+20GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 32.5 | 360 | LocalScore read 2026-10-22 |
| Apple M1 Max 8P+2E+24GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 32.1 | 306 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4050 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 31.4 | 1201 | LocalScore read 2026-10-22 |
| Tesla P100-PCIE-16GB | Meta Llama 3.1 8B Instruct, Q4_K_M | 31.1 | 372 | LocalScore read 2026-10-22 |
| AMD Radeon RX 6800 | Meta Llama 3.1 8B Instruct, Q4_K_M | 30.9 | 737 | LocalScore read 2026-10-22 |
| Apple M4 Pro 8P+4E+16GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 30.6 | 302 | LocalScore read 2026-10-22 |
| Apple M4 Pro 8P+4E+16GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 29.7 | 293 | LocalScore read 2026-10-22 |
| NVIDIA RTX 2000 Ada Generation Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 29.5 | 1151 | LocalScore read 2026-10-22 |
| Quadro RTX 3000 | Meta Llama 3.1 8B Instruct, Q4_K_M | 29.5 | 606 | LocalScore read 2026-10-22 |
| NVIDIA RTX A2000 | Meta Llama 3.1 8B Instruct, Q4_K_M | 28.7 | 758 | LocalScore read 2026-10-22 |
| NVIDIA RTX A2000 8GB Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 27.5 | 790 | LocalScore read 2026-10-22 |
| Apple M2 Pro 8P+4E+19GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 26.3 | 262 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5080 | Meta Llama 3.1 8B Instruct, Q4_K_M | 26.1 | 3662 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3050 | Meta Llama 3.1 8B Instruct, Q4_K_M | 25.9 | 778 | LocalScore read 2026-10-22 |
| Apple M3 Max 12P+4E+40GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 25.4 | 377 | LocalScore read 2026-10-22 |
| AMD Radeon RX 5700 XT | Meta Llama 3.1 8B Instruct, Q4_K_M | 25.4 | 256 | LocalScore read 2026-10-22 |
| AMD Radeon RX 6650 XT | Meta Llama 3.1 8B Instruct, Q4_K_M | 25.2 | 536 | LocalScore read 2026-10-22 |
| Apple M6 2P+4E+12GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 25.2 | 263 | LocalScore read 2026-10-22 |
| Apple M2 Pro 6P+4E+16GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 24.3 | 225 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3050 6GB Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 24.2 | 790 | LocalScore read 2026-10-22 |
| NVIDIA RTX 1000 Ada Generation Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 24.2 | 955 | LocalScore read 2026-10-22 |
| Apple M2 Pro 6P+4E+16GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 24.1 | 216 | LocalScore read 2026-10-22 |
| Apple M3 Pro 6P+6E+18GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 22.1 | 284 | LocalScore read 2026-10-22 |
| Apple M1 Pro 8P+2E+16GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 21.9 | 205 | LocalScore read 2026-10-22 |
| Apple M1 Pro 8P+2E+16GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 21.7 | 200 | LocalScore read 2026-10-22 |
| Apple M5 4P+6E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 21.6 | 202 | LocalScore read 2026-10-22 |
| AMD Radeon RX 6600 XT | Meta Llama 3.1 8B Instruct, Q4_K_M | 21.5 | 604 | LocalScore read 2026-10-22 |
| Apple M3 Pro 5P+6E+14GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 21.5 | 223 | LocalScore read 2026-10-22 |
| Apple M3 Pro 5P+6E+14GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 21.1 | 218 | LocalScore read 2026-10-22 |
| Apple M3 Pro 6P+6E+18GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 20.8 | 279 | LocalScore read 2026-10-22 |
| Apple M5 4P+6E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 20.2 | 189 | LocalScore read 2026-10-22 |
| NVIDIA GeForce GTX 1080 | Meta Llama 3.1 8B Instruct, Q4_K_M | 20.2 | 652 | LocalScore read 2026-10-22 |
| Apple M1 Pro 6P+2E+14GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 20.1 | 177 | LocalScore read 2026-10-22 |
| Apple M5 4P+6E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 20.0 | 177 | LocalScore read 2026-10-22 |
| Apple M1 Pro 6P+2E+14GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 20.0 | 173 | LocalScore read 2026-10-22 |
| NVIDIA RTX PRO 1000 Blackwell Generation Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 19.9 | 906 | LocalScore read 2026-10-22 |
| Apple M2 4P+4E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 19.4 | 168 | LocalScore read 2026-10-22 |
| Apple M2 4P+4E+8GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 17.3 | 144 | LocalScore read 2026-10-22 |
| Apple M4 4P+6E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 16.9 | 175 | LocalScore read 2026-10-22 |
| Apple M4 4P+6E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 16.8 | 166 | LocalScore read 2026-10-22 |
| AMD Radeon RX 6700 XT | Meta Llama 3.1 8B Instruct, Q4_K_M | 16.4 | 597 | LocalScore read 2026-10-22 |
| Apple A18 Pro 2P+4E+5GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 15.8 | 116 | LocalScore read 2026-10-22 |
| Apple M4 4P+6E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 15.4 | 148 | LocalScore read 2026-10-22 |
| Apple M4 4P+6E+8GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 15.3 | 134 | LocalScore read 2026-10-22 |
| Apple M2 4P+4E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 14.7 | 142 | LocalScore read 2026-10-22 |
| Apple M2 4P+4E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 14.7 | 141 | LocalScore read 2026-10-22 |
| Apple M1 4P+4E+8GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 14.6 | 134 | LocalScore read 2026-10-22 |
| Apple M3 4P+4E+8GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 13.7 | 125 | LocalScore read 2026-10-22 |
| Apple M3 4P+4E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 13.5 | 139 | LocalScore read 2026-10-22 |
| Apple M1 4P+4E+7GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 13.4 | 109 | LocalScore read 2026-10-22 |
| Apple M2 4P+4E+8GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 12.9 | 114 | LocalScore read 2026-10-22 |
| AMD Radeon RX 6600 | Meta Llama 3.1 8B Instruct, Q4_K_M | 10.2 | 419 | LocalScore read 2026-10-22 |
| Apple M3 4P+4E+10GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 10.2 | 110 | LocalScore read 2026-10-22 |
| NVIDIA GeForce GTX 1660 SUPER | Meta Llama 3.1 8B Instruct, Q4_K_M | 9.6 | 129 | LocalScore read 2026-10-22 |
| Apple M1 4P+4E+7GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 9.4 | 84 | LocalScore read 2026-10-22 |
| Apple M1 4P+4E+8GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 9.3 | 93 | LocalScore read 2026-10-22 |
| NVIDIA GeForce GTX 1660 | Meta Llama 3.1 8B Instruct, Q4_K_M | 9.0 | 129 | LocalScore read 2026-10-22 |
| AMD Radeon RX 6950 XT | Meta Llama 3.1 8B Instruct, Q4_K_M | 6.1 | 1024 | LocalScore read 2026-10-22 |
| AMD Radeon 780M Graphics | Meta Llama 3.1 8B Instruct, Q4_K_M | 5.6 | 186 | LocalScore read 2026-10-22 |
| AMD Radeon RX 7600 | Meta Llama 3.1 8B Instruct, Q4_K_M | 5.5 | 317 | LocalScore read 2026-10-22 |
| NVIDIA T1000 8GB | Meta Llama 3.1 8B Instruct, Q4_K_M | 5.2 | 74 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3050 Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 4.8 | 280 | LocalScore read 2026-10-22 |
| AMD Radeon™ RX 7700S | Meta Llama 3.1 8B Instruct, Q4_K_M | 4.6 | 289 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3050 Ti Laptop GPU | Meta Llama 3.1 8B Instruct, Q4_K_M | 4.3 | 290 | LocalScore read 2026-10-22 |
| AMD Radeon(TM) 780M | Meta Llama 3.1 8B Instruct, Q4_K_M | 3.8 | 112 | LocalScore read 2026-10-22 |
| AMD Radeon 780M Graphics | Meta Llama 3.1 8B Instruct, Q4_K_M | 3.5 | 110 | LocalScore read 2026-10-22 |
| AMD Radeon 780M Graphics | Meta Llama 3.1 8B Instruct, Q4_K_M | 3.3 | 94 | LocalScore read 2026-10-22 |
| AMD Radeon 780M Graphics | Meta Llama 3.1 8B Instruct, Q4_K_M | 3.3 | 103 | LocalScore read 2026-10-22 |
| NVIDIA GeForce GTX 1650 Ti | Meta Llama 3.1 8B Instruct, Q4_K_M | 2.8 | 66 | LocalScore read 2026-10-22 |
| Quadro P620 | Meta Llama 3.1 8B Instruct, Q4_K_M | 1.3 | 45 | LocalScore read 2026-10-22 |
| Apple M3 4P+4E+10GPU | Qwen2.5 14B Instruct, Q4_K_M | 783.9 | 4916 | LocalScore read 2026-10-22 |
| NVIDIA H100 PCIe | Qwen2.5 14B Instruct, Q4_K_M | 72.2 | 2973 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4090 D | Qwen2.5 14B Instruct, Q4_K_M | 70.5 | 3905 | LocalScore read 2026-10-22 |
| NVIDIA RTX 6000 Ada Generation | Qwen2.5 14B Instruct, Q4_K_M | 70.0 | 3552 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4090 | Qwen2.5 14B Instruct, Q4_K_M | 68.8 | 4208 | LocalScore read 2026-10-22 |
| NVIDIA A100-SXM4-80GB | Qwen2.5 14B Instruct, Q4_K_M | 66.3 | 2069 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3090 Ti | Qwen2.5 14B Instruct, Q4_K_M | 64.2 | 2202 | LocalScore read 2026-10-22 |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | Qwen2.5 14B Instruct, Q4_K_M | 64.0 | 4416 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3090 | Qwen2.5 14B Instruct, Q4_K_M | 56.2 | 1950 | LocalScore read 2026-10-22 |
| Tesla V100-SXM2-16GB | Qwen2.5 14B Instruct, Q4_K_M | 53.0 | 1184 | LocalScore read 2026-10-22 |
| NVIDIA RTX A6000 | Qwen2.5 14B Instruct, Q4_K_M | 52.0 | 1883 | LocalScore read 2026-10-22 |
| NVIDIA RTX A5000 | Qwen2.5 14B Instruct, Q4_K_M | 50.4 | 1532 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4090 | Qwen2.5 14B Instruct, Q4_K_M | 49.1 | 3506 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3080 | Qwen2.5 14B Instruct, Q4_K_M | 46.6 | 1741 | LocalScore read 2026-10-22 |
| NVIDIA A40 | Qwen2.5 14B Instruct, Q4_K_M | 46.2 | 1825 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4080 SUPER | Qwen2.5 14B Instruct, Q4_K_M | 46.1 | 2803 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3080 Ti | Qwen2.5 14B Instruct, Q4_K_M | 46.1 | 1793 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5070 | Qwen2.5 14B Instruct, Q4_K_M | 45.7 | 1758 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5090 | Qwen2.5 14B Instruct, Q4_K_M | 45.5 | 3678 | LocalScore read 2026-10-22 |
| NVIDIA RTX A4500 | Qwen2.5 14B Instruct, Q4_K_M | 44.0 | 1331 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4080 SUPER | Qwen2.5 14B Instruct, Q4_K_M | 43.6 | 2681 | LocalScore read 2026-10-22 |
| Apple M5 Max 6P+12E+40GPU | Qwen2.5 14B Instruct, Q4_K_M | 42.1 | 424 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5090 Laptop GPU | Qwen2.5 14B Instruct, Q4_K_M | 40.5 | 1550 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4070 Ti | Qwen2.5 14B Instruct, Q4_K_M | 40.2 | 2176 | LocalScore read 2026-10-22 |
| NVIDIA A40-48Q | Qwen2.5 14B Instruct, Q4_K_M | 39.5 | 1707 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5090 | Qwen2.5 14B Instruct, Q4_K_M | 38.0 | 3790 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4070 | Qwen2.5 14B Instruct, Q4_K_M | 37.9 | 1692 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5070 Ti | Qwen2.5 14B Instruct, Q4_K_M | 37.0 | 2193 | LocalScore read 2026-10-22 |
| Apple M5 Max 6P+12E+40GPU | Qwen2.5 14B Instruct, Q4_K_M | 36.9 | 360 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4070 Ti SUPER | Qwen2.5 14B Instruct, Q4_K_M | 36.8 | 2147 | LocalScore read 2026-10-22 |
| Apple M3 Ultra 24P+8E+80GPU | Qwen2.5 14B Instruct, Q4_K_M | 36.7 | 568 | LocalScore read 2026-10-22 |
| Apple M2 Ultra 16P+8E+76GPU | Qwen2.5 14B Instruct, Q4_K_M | 36.6 | 471 | LocalScore read 2026-10-22 |
| Apple M2 Ultra 16P+8E+76GPU | Qwen2.5 14B Instruct, Q4_K_M | 36.5 | 469 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5070 Ti Laptop GPU | Qwen2.5 14B Instruct, Q4_K_M | 36.3 | 1474 | LocalScore read 2026-10-22 |
| Apple M3 Ultra 24P+8E+80GPU | Qwen2.5 14B Instruct, Q4_K_M | 35.8 | 580 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4080 | Qwen2.5 14B Instruct, Q4_K_M | 35.2 | 2339 | LocalScore read 2026-10-22 |
| Apple M3 Ultra 20P+8E+60GPU | Qwen2.5 14B Instruct, Q4_K_M | 34.4 | 445 | LocalScore read 2026-10-22 |
| Apple M2 Ultra 16P+8E+60GPU | Qwen2.5 14B Instruct, Q4_K_M | 34.2 | 381 | LocalScore read 2026-10-22 |
| Apple M5 Max 6P+12E+32GPU | Qwen2.5 14B Instruct, Q4_K_M | 34.1 | 343 | LocalScore read 2026-10-22 |
| Apple M2 Ultra 16P+8E+60GPU | Qwen2.5 14B Instruct, Q4_K_M | 33.9 | 380 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5080 | Qwen2.5 14B Instruct, Q4_K_M | 33.1 | 2416 | LocalScore read 2026-10-22 |
| Apple M1 Ultra 16P+4E+64GPU | Qwen2.5 14B Instruct, Q4_K_M | 32.4 | 372 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4090 | Qwen2.5 14B Instruct, Q4_K_M | 31.6 | 3056 | LocalScore read 2026-10-22 |
| Apple M4 Max 12P+4E+40GPU | Qwen2.5 14B Instruct, Q4_K_M | 30.2 | 336 | LocalScore read 2026-10-22 |
| Apple M4 Max 12P+4E+40GPU | Qwen2.5 14B Instruct, Q4_K_M | 28.9 | 328 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3080 Laptop GPU | Qwen2.5 14B Instruct, Q4_K_M | 28.2 | 858 | LocalScore read 2026-10-22 |
| NVIDIA RTX 6000 Ada Generation | Qwen2.5 14B Instruct, Q4_K_M | 28.1 | 2936 | LocalScore read 2026-10-22 |
| Apple M1 Ultra 16P+4E+48GPU | Qwen2.5 14B Instruct, Q4_K_M | 27.8 | 290 | LocalScore read 2026-10-22 |
| NVIDIA RTX A5000 Laptop GPU | Qwen2.5 14B Instruct, Q4_K_M | 27.6 | 839 | LocalScore read 2026-10-22 |
| Apple M4 Max 12P+4E+40GPU | Qwen2.5 14B Instruct, Q4_K_M | 25.6 | 288 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5070 Ti | Qwen2.5 14B Instruct, Q4_K_M | 25.6 | 1795 | LocalScore read 2026-10-22 |
| Apple M4 Max 10P+4E+32GPU | Qwen2.5 14B Instruct, Q4_K_M | 25.6 | 285 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 5080 | Qwen2.5 14B Instruct, Q4_K_M | 25.5 | 2097 | LocalScore read 2026-10-22 |
| Apple M3 Max 12P+4E+40GPU | Qwen2.5 14B Instruct, Q4_K_M | 25.5 | 302 | LocalScore read 2026-10-22 |
| Apple M2 Max 8P+4E+38GPU | Qwen2.5 14B Instruct, Q4_K_M | 25.2 | 252 | LocalScore read 2026-10-22 |
| NVIDIA RTX 3500 Ada Generation Laptop GPU | Qwen2.5 14B Instruct, Q4_K_M | 25.1 | 1298 | LocalScore read 2026-10-22 |
| NVIDIA L40 | Qwen2.5 14B Instruct, Q4_K_M | 24.7 | 2713 | LocalScore read 2026-10-22 |
| Apple M5 Pro 6P+12E+20GPU | Qwen2.5 14B Instruct, Q4_K_M | 24.2 | 217 | LocalScore read 2026-10-22 |
| Apple M5 Pro 6P+12E+20GPU | Qwen2.5 14B Instruct, Q4_K_M | 24.0 | 218 | LocalScore read 2026-10-22 |
| Apple M5 Pro 6P+12E+20GPU | Qwen2.5 14B Instruct, Q4_K_M | 23.7 | 216 | LocalScore read 2026-10-22 |
| NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | Qwen2.5 14B Instruct, Q4_K_M | 23.7 | 2623 | LocalScore read 2026-10-22 |
| Apple M3 Max 12P+4E+40GPU | Qwen2.5 14B Instruct, Q4_K_M | 23.6 | 287 | LocalScore read 2026-10-22 |
| NVIDIA H100 80GB HBM3 | Qwen2.5 14B Instruct, Q4_K_M | 23.4 | 2609 | LocalScore read 2026-10-22 |
| Apple M5 Pro 5P+10E+16GPU | Qwen2.5 14B Instruct, Q4_K_M | 22.5 | 177 | LocalScore read 2026-10-22 |
| Apple M2 Max 8P+4E+38GPU | Qwen2.5 14B Instruct, Q4_K_M | 22.0 | 223 | LocalScore read 2026-10-22 |
| Apple M5 Pro 5P+10E+16GPU | Qwen2.5 14B Instruct, Q4_K_M | 21.6 | 169 | LocalScore read 2026-10-22 |
| Apple M3 Max 10P+4E+30GPU | Qwen2.5 14B Instruct, Q4_K_M | 20.8 | 239 | LocalScore read 2026-10-22 |
| Apple M2 Max 8P+4E+38GPU | Qwen2.5 14B Instruct, Q4_K_M | 20.6 | 226 | LocalScore read 2026-10-22 |
| NVIDIA H100 NVL | Qwen2.5 14B Instruct, Q4_K_M | 20.5 | 2354 | LocalScore read 2026-10-22 |
| Apple M1 Max 8P+2E+32GPU | Qwen2.5 14B Instruct, Q4_K_M | 20.2 | 197 | LocalScore read 2026-10-22 |
| Apple M1 Max 8P+2E+32GPU | Qwen2.5 14B Instruct, Q4_K_M | 19.4 | 191 | LocalScore read 2026-10-22 |
| Apple M3 Max 10P+4E+30GPU | Qwen2.5 14B Instruct, Q4_K_M | 19.4 | 216 | LocalScore read 2026-10-22 |
| NVIDIA L40S | Qwen2.5 14B Instruct, Q4_K_M | 19.3 | 2868 | LocalScore read 2026-10-22 |
| Apple M1 Max 8P+2E+24GPU | Qwen2.5 14B Instruct, Q4_K_M | 18.2 | 155 | LocalScore read 2026-10-22 |
| Apple M4 Pro 10P+4E+20GPU | Qwen2.5 14B Instruct, Q4_K_M | 18.0 | 183 | LocalScore read 2026-10-22 |
| Apple M4 Pro 10P+4E+20GPU | Qwen2.5 14B Instruct, Q4_K_M | 18.0 | 190 | LocalScore read 2026-10-22 |
| NVIDIA RTX PRO 6000 Blackwell Server Edition | Qwen2.5 14B Instruct, Q4_K_M | 17.9 | 2581 | LocalScore read 2026-10-22 |
| Apple M4 Pro 10P+4E+20GPU | Qwen2.5 14B Instruct, Q4_K_M | 17.6 | 186 | LocalScore read 2026-10-22 |
| Apple M3 Max 12P+4E+40GPU | Qwen2.5 14B Instruct, Q4_K_M | 17.2 | 233 | LocalScore read 2026-10-22 |
| Apple M4 Pro 8P+4E+16GPU | Qwen2.5 14B Instruct, Q4_K_M | 16.7 | 157 | LocalScore read 2026-10-22 |
| NVIDIA RTX 5000 Ada Generation Laptop GPU | Qwen2.5 14B Instruct, Q4_K_M | 16.4 | 952 | LocalScore read 2026-10-22 |
| Apple M4 Pro 8P+4E+16GPU | Qwen2.5 14B Instruct, Q4_K_M | 16.1 | 151 | LocalScore read 2026-10-22 |
| Apple M4 Pro 8P+4E+16GPU | Qwen2.5 14B Instruct, Q4_K_M | 16.1 | 151 | LocalScore read 2026-10-22 |
| Apple M1 Max 8P+2E+24GPU | Qwen2.5 14B Instruct, Q4_K_M | 15.1 | 140 | LocalScore read 2026-10-22 |
| Quadro P5000 | Qwen2.5 14B Instruct, Q4_K_M | 14.9 | 303 | LocalScore read 2026-10-22 |
| Apple M2 Max 8P+4E+30GPU | Qwen2.5 14B Instruct, Q4_K_M | 14.5 | 149 | LocalScore read 2026-10-22 |
| Apple M2 Pro 8P+4E+19GPU | Qwen2.5 14B Instruct, Q4_K_M | 14.1 | 137 | LocalScore read 2026-10-22 |
| Apple M2 Pro 6P+4E+16GPU | Qwen2.5 14B Instruct, Q4_K_M | 13.3 | 118 | LocalScore read 2026-10-22 |
| Apple M3 Pro 5P+6E+14GPU | Qwen2.5 14B Instruct, Q4_K_M | 12.1 | 120 | LocalScore read 2026-10-22 |
| Apple M3 Pro 6P+6E+18GPU | Qwen2.5 14B Instruct, Q4_K_M | 11.9 | 147 | LocalScore read 2026-10-22 |
| Apple M1 Pro 8P+2E+16GPU | Qwen2.5 14B Instruct, Q4_K_M | 11.9 | 107 | LocalScore read 2026-10-22 |
| Apple M3 Pro 6P+6E+18GPU | Qwen2.5 14B Instruct, Q4_K_M | 11.8 | 145 | LocalScore read 2026-10-22 |
| Apple M3 Pro 5P+6E+14GPU | Qwen2.5 14B Instruct, Q4_K_M | 11.6 | 115 | LocalScore read 2026-10-22 |
| Apple M1 Pro 8P+2E+16GPU | Qwen2.5 14B Instruct, Q4_K_M | 11.6 | 103 | LocalScore read 2026-10-22 |
| Apple M5 4P+6E+10GPU | Qwen2.5 14B Instruct, Q4_K_M | 11.5 | 110 | LocalScore read 2026-10-22 |
| Apple M1 Pro 6P+2E+14GPU | Qwen2.5 14B Instruct, Q4_K_M | 10.8 | 92 | LocalScore read 2026-10-22 |
| Apple M1 Pro 8P+2E+14GPU | Qwen2.5 14B Instruct, Q4_K_M | 10.4 | 89 | LocalScore read 2026-10-22 |
| Apple M4 4P+6E+10GPU | Qwen2.5 14B Instruct, Q4_K_M | 9.1 | 92 | LocalScore read 2026-10-22 |
| Apple M4 4P+6E+10GPU | Qwen2.5 14B Instruct, Q4_K_M | 9.0 | 87 | LocalScore read 2026-10-22 |
| Apple M1 Pro 6P+2E+14GPU | Qwen2.5 14B Instruct, Q4_K_M | 8.9 | 83 | LocalScore read 2026-10-22 |
| Apple M5 4P+6E+10GPU | Qwen2.5 14B Instruct, Q4_K_M | 8.8 | 81 | LocalScore read 2026-10-22 |
| Apple M4 4P+6E+10GPU | Qwen2.5 14B Instruct, Q4_K_M | 8.6 | 79 | LocalScore read 2026-10-22 |
| NVIDIA RTX A4500 Laptop GPU | Qwen2.5 14B Instruct, Q4_K_M | 7.9 | 302 | LocalScore read 2026-10-22 |
| Apple M2 4P+4E+10GPU | Qwen2.5 14B Instruct, Q4_K_M | 7.7 | 72 | LocalScore read 2026-10-22 |
| Apple M2 4P+4E+10GPU | Qwen2.5 14B Instruct, Q4_K_M | 7.3 | 70 | LocalScore read 2026-10-22 |
| Apple M4 4P+6E+8GPU | Qwen2.5 14B Instruct, Q4_K_M | 7.3 | 61 | LocalScore read 2026-10-22 |
| Apple M2 4P+4E+8GPU | Qwen2.5 14B Instruct, Q4_K_M | 7.0 | 60 | LocalScore read 2026-10-22 |
| Apple M3 4P+4E+10GPU | Qwen2.5 14B Instruct, Q4_K_M | 6.1 | 66 | LocalScore read 2026-10-22 |
| Apple M1 4P+4E+8GPU | Qwen2.5 14B Instruct, Q4_K_M | 5.4 | 53 | LocalScore read 2026-10-22 |
| Apple M1 4P+4E+7GPU | Qwen2.5 14B Instruct, Q4_K_M | 4.8 | 41 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 4070 Laptop GPU | Qwen2.5 14B Instruct, Q4_K_M | 3.8 | 289 | LocalScore read 2026-10-22 |
| AMD Radeon 780M Graphics | Qwen2.5 14B Instruct, Q4_K_M | 3.0 | 99 | LocalScore read 2026-10-22 |
| NVIDIA RTX PRO 2000 Blackwell Generation Laptop GPU | Qwen2.5 14B Instruct, Q4_K_M | 2.1 | 230 | LocalScore read 2026-10-22 |
| AMD Radeon 780M Graphics | Qwen2.5 14B Instruct, Q4_K_M | 1.9 | 57 | LocalScore read 2026-10-22 |
| NVIDIA GeForce RTX 3060 Laptop GPU | Qwen2.5 14B Instruct, Q4_K_M | 1.1 | 83 | LocalScore read 2026-10-22 |
Where the numbers come from
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- LocalScore
- llama.cpp discussion 10879
- llama.cpp discussion 15013
- llama.cpp discussion 15021
- llama.cpp discussion 4167
- llama.cpp discussion 8273
- llama.cpp discussion 8336
We store numbers and facts only, with the link. How the finder estimates speed.