GPUs & Macs › NVIDIA RTX 5090
NVIDIA RTX 5090: local LLM speed
Llama 2 7B (Q4_0) generates 300 tokens/s on the NVIDIA RTX 5090 (CUDA), and processes prompts at 14,970 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 300 | 14,970 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 290 | 14,073 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 274 | 11,796 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 264 | 10,382 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
| gpt-oss 20B | 283 | 9,848 | MXFP4 | CUDA FA on | tg128 | llama.cpp #15396 details |
| Llama 3.2 1B Instruct | 120 | 16,760 | Q4_K_M | llamafile | LocalScore avg · n=2 | LocalScore LocalScore details |
| Llama 3.1 8B Instruct | 66.3 | 6,297 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Qwen2.5 14B Instruct | 41.8 | 3,734 | Q4_K_M | llamafile | LocalScore avg · n=2 | LocalScore LocalScore details |
What fits on the NVIDIA RTX 5090
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 300 tok/s (CUDA, source). Fit assumes the whole model in GPU memory with ~8K context.
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| Qwen3.8-27B | Q8_0 | 29.0 GB | 39.5 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 90.6 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 121 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 256 estimate |
| Qwen3.6-35B-A3B | UD-Q6_K | 29.3 GB | 229 estimate |
| gemma-4-26B-A4B-it | Q8_0 | 26.9 GB | 139 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 120 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 140 estimate |
| Muse-Glimmer-30B | Q8_0 | 29.6 GB | 38.8 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 268 estimate |
| gemma-4-31B-it | Q6_K | 25.2 GB | 45.6 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 262 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 171 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 957 estimate |
Similar hardware
NVIDIA RTX 4090 D · NVIDIA RTX 4090 · NVIDIA RTX 3090 Ti · NVIDIA RTX 3090 · NVIDIA TITAN RTX · NVIDIA Quadro P6000 · NVIDIA RTX 5080 · NVIDIA RTX 5070 Ti