Llama 3.3 70B
32 records across 14 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GB300 | Llama 3.3 70B 70B dense | NVFP4 | vLLM | 11478.84 | – | – | – | – | 2026-06 | community | – | source |
| B300 | Llama 3.3 70B 70B dense | NVFP4 | vLLM | 162.29 | – | – | – | – | 2026-06 | community | – | source |
| GB300 | Llama 3.3 70B 70B dense | NVFP4 | vLLM | 143.25 | – | – | – | – | 2026-06 | community | – | source |
| RTX Pro 5000 Blackwell | Llama 3.3 70B 70B | Q4_K | llama.cpp | 24.0 | 1022.9 | – | – | 4096 | – | sourced | – | source |
| RTX Pro 5000 Blackwell | Llama 3.3 70B 70B | Q4_K | llama.cpp | 22.4 | 826.3 | – | – | 16384 | – | sourced | – | source |
| M4 Ultra (192 GB) | Llama 3.3 70B | Q4_K - Medium | MLX | 21 | – | – | 2 | – | – | sourced | – | source |
| RTX 6000 Ada | Llama 3.3 70B 70B | Q4_K | llama.cpp | 18.1 | 678.1 | – | – | 4096 | – | sourced | – | source |
| M3 Ultra (512 GB) | Llama 3.3 70B | 4bit | LM Studio | 15.5 | 150 | – | – | 7800 | – | sourced | – | source |
| M2 Ultra | Llama 3.3 70B | 4bit | oMLX | 15.3 | 114.0 | – | – | 1024 | 2026-10-17 | community | – | source |
| M2 Ultra | Llama 3.3 70B | 4bit | oMLX | 15.2 | 114.7 | – | – | 1024 | 2026-10-17 | community | – | source |
| M2 Ultra | Llama 3.3 70B | 4bit | oMLX | 14.7 | 113.4 | – | – | 4096 | 2026-10-17 | community | – | source |
| M2 Ultra | Llama 3.3 70B | 4bit | oMLX | 14.6 | 113.5 | – | – | 4096 | 2026-10-17 | community | – | source |
| M2 Ultra | Llama 3.3 70B | 4bit | oMLX | 14.6 | 113.5 | – | – | 4096 | 2026-10-17 | community | – | source |
| RTX 6000 Ada | Llama 3.3 70B 70B | Q4_K | llama.cpp | 13.7 | 526.0 | – | – | 16384 | – | sourced | – | source |
| RTX A6000 | Llama 3.3 70B 70B | Q4_K | llama.cpp | 12.6 | 423.0 | – | – | 4096 | – | sourced | – | source |
| M1 Ultra (64 GB) | Llama 3.3 70B | 4bit | LM Studio | 12.55 | – | – | 49.8 | 3991 | – | sourced | – | source |
| M2 Ultra | Llama 3.3 70B | 4bit | oMLX | 12.2 | 102.0 | – | – | 16384 | 2026-10-17 | community | – | source |
| M4 Max (128 GB) | Llama 3.3 70B | 4bit | LM Studio | 11.8 | – | – | – | – | – | sourced | – | source |
| M3 Ultra (512 GB) | Llama 3.3 70B | 4bit | LM Studio | 9.57 | 103 | – | – | 39686 | – | sourced | – | source |
| RTX A6000 | Llama 3.3 70B 70B | Q4_K | llama.cpp | 9.1 | 280.3 | – | – | 16384 | – | sourced | – | source |
| M3 Ultra | Llama 3.3 70B | 8bit | oMLX | 9.0 | 163.8 | – | – | 1024 | 2026-10-08 | community | – | source |
| M2 Max (38-core GPU, 64 GB) | Llama 3.3 70B | 4bit | LM Studio | 8.8 | – | – | – | – | – | sourced | – | source |
| M3 Ultra | Llama 3.3 70B | 8bit | oMLX | 8.7 | 164.5 | – | – | 4096 | 2026-10-08 | community | – | source |
| M3 Ultra (512 GB) | Llama 3.3 70B | 8bit | LM Studio | 8.5 | 150 | – | – | 7800 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Llama 3.3 70B | Q4_K - Medium | llama.cpp | 8.15 | 67.86 | – | – | 258 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Llama 3.3 70B | Q4_K - Medium | llama.cpp | 7.48 | 65.17 | – | – | 8013 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Llama 3.3 70B | Q4_K - Medium | llama.cpp | 7 | 59.5 | – | – | 16001 | – | sourced | – | source |
| M3 Ultra (512 GB) | Llama 3.3 70B | 8bit | LM Studio | 6.53 | 101 | – | – | 39686 | – | sourced | – | source |
| M4 Max (128 GB) | Llama 3.3 70B | 8bit | LM Studio | 6.5 | – | – | – | – | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Llama 3.3 70B | Q4_K - Medium | llama.cpp | 6.13 | 50.32 | – | – | 32170 | – | sourced | – | source |
| M4 Pro (64 GB) | Llama 3.3 70B | Q4_K - Medium | Ollama | 5 | – | – | 3.75 | – | – | sourced | – | source |
| M4 Max (128 GB) | Llama 3.3 70B | Q8_0 | llama.cpp | 4.5 | 58 | – | 600 | 35550 | – | sourced | – | source |