Llama 2 7B › Apple M3 Max (40-core GPU)
Llama 2 7B on the Apple M3 Max (40-core GPU)
Best measured: 66.3 tokens/s generation and 760 tokens/s prompt processing (Q4_0, Metal, FA off).
Every run
| Quant | Backend | Generate tok/s | Prompt tok/s | Test | Build / by / notes | Source |
|---|---|---|---|---|---|---|
| Q4_0 | Metal FA off | 66.3 | 760 | tg128 | 8e672ef | llama.cpp Apple Silicon thread (#4167) |
| Q8_0 | Metal FA off | 42.8 | 758 | tg128 | 8e672ef | llama.cpp Apple Silicon thread (#4167) |
| F16 | Metal FA off | 25.1 | 779 | tg128 | 8e672ef | llama.cpp Apple Silicon thread (#4167) |
tg128 = generating 128 tokens; pp512 = reading a 512-token prompt (llama-bench). LocalScore averages several prompt and output lengths, so its numbers run lower than tg128.
No card yet?
Try it on a rented GPU (RunPod)Affiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.
All results for the Apple M3 Max (40-core GPU) · Llama 2 7B on other hardware