GPUs & Macs › NVIDIA Tesla K80
NVIDIA Tesla K80: local LLM speed
Llama 2 7B (Q4_0) generates 14.3 tokens/s on the NVIDIA Tesla K80 (CUDA), and processes prompts at 133 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 14.3 | 133 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 13.8 | 133 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 9.5 | 88 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 9.4 | 89 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
What fits on the NVIDIA Tesla K80
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 14.3 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-IQ2_XXS | 7.3 GB | 7.5 estimate |
| gemma-4-12B-it | Q6_K | 9.8 GB | 5.6 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 5.7 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 12.2 estimate |
| gemma-4-26B-A4B-it | UD-IQ2_M | 10.0 GB | 17.7 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 5.7 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 6.7 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 12.7 estimate |
| gemma-4-31B-it | UD-IQ2_XXS | 8.5 GB | 6.4 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 8.1 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 45.4 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 11.0 estimate |
| DeepSeek-R1-0528-Qwen3-8B | Q8_0 | 8.7 GB | 6.3 estimate |
| Llama-3.2-3B-Instruct | Q8_0 | 3.4 GB | 15.9 estimate |
Similar hardware
NVIDIA Tesla P100 · NVIDIA Tesla T4 · NVIDIA Tesla P4 · NVIDIA Tesla V100 16GB · NVIDIA P106-100 · NVIDIA P102-100 · NVIDIA P104-100 · NVIDIA A30