Llama 2 7B on RTX 5090
300best gen tok/s
14,970best prompt tok/s
640max context tested
4results
Measured · all numbers below come from the linked sources
llama.cpp
| Quant | Gen tok/s | Prompt tok/s | Source | Context tested | TTFT | Build |
|---|---|---|---|---|---|---|
| Q4_0 | 300 | 14,970 | llama.cpp (CUDA) ↗ | 512+128 | — | llama.cpp CUDA @8cf6b42 |
| Q4_0 | 290 | 14,073 | llama.cpp (CUDA) ↗ | 512+128 | — | llama.cpp CUDA @8cf6b42 |
| Q4_0 | 274 | 11,796 | llama.cpp (Vulkan) ↗ | 512+128 | — | llama.cpp Vulkan @ca71fb9 — coopmat2 |
| Q4_0 | 264 | 10,382 | llama.cpp (Vulkan) ↗ | 512+128 | — | llama.cpp Vulkan @ca71fb9 — coopmat2 |
Ollama
No measured Ollama results yet. Submit yours →
MLX
No measured MLX results yet. Submit yours →
"Context tested" is the largest prompt+generation length actually benchmarked (tokens), not a memory-fit estimate.