GPUs & Macs › NVIDIA RTX 4090 D
NVIDIA RTX 4090 D: local LLM speed
Llama 2 7B (Q4_0) generates 192 tokens/s on the NVIDIA RTX 4090 D (CUDA), and processes prompts at 12,507 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 192 | 12,507 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 189 | 10,294 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 3.2 1B Instruct | 359 | 24,549 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Qwen2.5 14B Instruct | 70.5 | 3,905 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA RTX 4090 D
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 192 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 | UD-Q6_K | 22.0 GB | 33.3 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 57.8 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 76.9 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 163 estimate |
| Qwen3.6-35B-A3B | UD-Q4_K_S | 20.9 GB | 204 estimate |
| gemma-4-26B-A4B-it | UD-Q5_K_M | 21.2 GB | 113 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 76.7 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 89.4 estimate |
| Muse-Glimmer-30B | UD-Q5_K_M | 19.2 GB | 38.2 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 171 estimate |
| gemma-4-31B-it | Q5_K_M | 21.7 GB | 33.8 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 167 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 109 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 610 estimate |
Similar hardware
NVIDIA RTX 4090 · NVIDIA RTX 3090 Ti · NVIDIA RTX 3090 · NVIDIA TITAN RTX · NVIDIA Quadro P6000 · NVIDIA RTX 5090 · NVIDIA RTX 5080 · NVIDIA RTX 5070 Ti