Llama 2 7B on the NVIDIA L40
Best measured: 154 tokens/s generation and 10,098 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 | 154 | 10,098 | tg128 | ee09828 @Hedede | llama.cpp CUDA scoreboard (#15013) |
| Q4_0 | CUDA FA off | 152 | 8,870 | tg128 | ee09828 @Hedede | llama.cpp CUDA scoreboard (#15013) |
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 L40 · Llama 2 7B on other hardware