Llama 2 7B on the Apple M2 Ultra
Best measured: 36.1 tokens/s generation and 199 tokens/s prompt processing (Q4_0, Vulkan, FA on).
For comparison, our size-scaling estimate for this combination is 169 tok/s.
Every run
| Quant | Backend | Generate tok/s | Prompt tok/s | Test | Build / by / notes | Source |
|---|---|---|---|---|---|---|
| Q4_0 | Vulkan FA on | 36.1 | 199 | tg128 | dbb852b Asahi Linux | llama.cpp Vulkan scoreboard (#10879) |
| Q4_0 | Vulkan FA off | 34.3 | 206 | tg128 | dbb852b Asahi Linux | llama.cpp Vulkan scoreboard (#10879) |
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 M2 Ultra · Llama 2 7B on other hardware