GPUs & Macs › NVIDIA Tesla T4
NVIDIA Tesla T4: local LLM speed
Llama 2 7B (Q4_0) generates 46.4 tokens/s on the NVIDIA Tesla T4 (CUDA), and processes prompts at 1,219 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 46.4 | 1,219 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 44.0 | 1,310 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Qwen2.5 14B Instruct | 13.3 | 450 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA Tesla T4
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 46.4 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-IQ3_XXS | 10.9 GB | 16.2 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 14.0 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 18.6 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 39.6 estimate |
| Qwen3.6-35B-A3B | UD-IQ3_XXS | 13.2 GB | 78.3 estimate |
| gemma-4-26B-A4B-it | UD-IQ4_XS | 13.6 GB | 42.4 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 18.6 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 21.6 estimate |
| Muse-Glimmer-30B | UD-IQ3_XXS | 13.1 GB | 13.5 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 41.4 estimate |
| gemma-4-31B-it | UD-IQ3_XXS | 11.8 GB | 15.0 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 26.4 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 148 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 35.7 estimate |
Similar hardware
NVIDIA Tesla P100 · NVIDIA Tesla V100 16GB · NVIDIA Tesla K80 · NVIDIA A30 · NVIDIA Tesla P40 · NVIDIA Tesla M40 · NVIDIA Tesla P4 · NVIDIA L4