Llama 2 7B on the NVIDIA Tesla K80
Best measured: 14.3 tokens/s generation and 133 tokens/s prompt processing (Q4_0, CUDA, FA on).
Every run
| Quant | Backend | Generate tok/s | Prompt tok/s | Test | Build / by / notes | Source |
|---|---|---|---|---|---|---|
| Q4_0 | CUDA FA on | 14.3 | 133 | tg128 | 32732f2 @pebaryan | llama.cpp CUDA scoreboard (#15013) |
| Q4_0 | CUDA FA off | 13.8 | 133 | tg128 | 32732f2 @pebaryan | llama.cpp CUDA scoreboard (#15013) |
| Q4_0 | Vulkan FA on | 9.5 | 88 | tg128 | 5d46bab Running on single GPU | llama.cpp Vulkan scoreboard (#10879) |
| Q4_0 | Vulkan FA off | 9.4 | 89 | tg128 | 5d46bab Running on single GPU | 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.
Rent one by the hour
Rent a GPU on RunPodAffiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.
All results for the NVIDIA Tesla K80 · Llama 2 7B on other hardware