Llama 2 7B on the NVIDIA RTX A4000
Best measured: 85.2 tokens/s generation and 2,808 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 | 85.2 | 2,808 | tg128 | 65349f2 @TinyServal | llama.cpp CUDA scoreboard (#15013) |
| Q4_0 | CUDA FA off | 83.8 | 2,684 | tg128 | 65349f2 @TinyServal | llama.cpp CUDA scoreboard (#15013) |
| Q4_0 | Vulkan FA on | 76.1 | 2,483 | tg128 | f5245b5 coopmat2 | llama.cpp Vulkan scoreboard (#10879) |
| Q4_0 | Vulkan FA off | 73.7 | 2,248 | tg128 | f5245b5 coopmat2 | 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 NVIDIA RTX A4000 · Llama 2 7B on other hardware