GPUs & Macs › NVIDIA TITAN Xp
NVIDIA TITAN Xp: local LLM speed
Llama 2 7B (Q4_0) generates 76.1 tokens/s on the NVIDIA TITAN Xp (CUDA), and processes prompts at 1,155 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 76.1 | 1,155 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 73.8 | 1,218 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 3.2 1B Instruct | 160 | 4,221 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Qwen2.5 14B Instruct | 21.1 | 412 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA TITAN Xp
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 76.1 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 | 40.0 estimate |
| gemma-4-12B-it | Q6_K | 9.8 GB | 29.7 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 30.5 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 64.9 estimate |
| gemma-4-26B-A4B-it | UD-IQ2_M | 10.0 GB | 94.4 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 30.5 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 35.5 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 67.9 estimate |
| gemma-4-31B-it | UD-IQ2_XXS | 8.5 GB | 34.1 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 43.3 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 242 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 58.5 estimate |
| DeepSeek-R1-0528-Qwen3-8B | Q8_0 | 8.7 GB | 33.4 estimate |
| Llama-3.2-3B-Instruct | Q8_0 | 3.4 GB | 85.0 estimate |
Similar hardware
NVIDIA RTX 5070 · NVIDIA TITAN V · NVIDIA RTX 3060 · NVIDIA RTX 4070 Ti · NVIDIA RTX 4070 SUPER · NVIDIA RTX 4070 · NVIDIA RTX 3080 Ti · NVIDIA RTX 2080 Ti