GPUs & Macs › NVIDIA P106-100
NVIDIA P106-100: local LLM speed
Llama 2 7B (Q4_0) generates 30.8 tokens/s on the NVIDIA P106-100 (Vulkan), and processes prompts at 183 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 30.8 | 183 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 30.6 | 438 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 30.4 | 407 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 29.8 | 184 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
| Llama 3.2 1B Instruct | 70.3 | 1,758 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA P106-100
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 30.8 tok/s (Vulkan, source). Fit assumes the whole model in GPU memory with ~8K context.
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| gemma-4-12B-it | UD-IQ2_M | 4.2 GB | 27.9 estimate |
| Qwen3.5-9B | UD-IQ3_XXS | 4.0 GB | 29.3 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 26.3 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q3_K_M | 4.5 GB | 26.3 estimate |
| gemma-4-E4B-it | Q3_K_M | 4.1 GB | 29.0 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 27.5 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 98.1 estimate |
| gemma-4-E2B-it | Q4_0 | 2.8 GB | 41.4 estimate |
| DeepSeek-R1-0528-Qwen3-8B | Q3_K_M | 4.1 GB | 28.5 estimate |
| Llama-3.2-3B-Instruct | Q8_0 | 3.4 GB | 34.4 estimate |
| Phi-4-mini-instruct | Q8_0 | 4.1 GB | 28.8 estimate |
| gemma-3-4b-it | Q8_0 | 4.1 GB | 28.5 estimate |
| Qwen3-14B | UD-IQ1_M | 4.1 GB | 29.0 estimate |
| gemma-3-12b-it | UD-IQ2_M | 4.4 GB | 27.0 estimate |
Similar hardware
NVIDIA P102-100 · NVIDIA Tesla P4 · NVIDIA P104-100 · NVIDIA Tesla K80 · NVIDIA Tesla P100 · NVIDIA Tesla T4 · NVIDIA Tesla V100 16GB · NVIDIA A30