GPUs & Macs › NVIDIA A100 80GB PCIe
NVIDIA A100 80GB PCIe: local LLM speed
Llama 2 7B (Q4_0) generates 201 tokens/s on the NVIDIA A100 80GB PCIe (CUDA), and processes prompts at 5,286 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 201 | 5,286 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 191 | 4,850 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 3.2 1B Instruct | 81.9 | 10,806 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Llama 3.1 8B Instruct | 37.1 | 2,601 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Qwen2.5 14B Instruct | 20.8 | 1,438 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the NVIDIA A100 80GB PCIe
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 201 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 | Q8_0 | 29.0 GB | 26.4 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 60.6 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 80.6 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 171 estimate |
| Qwen3.6-35B-A3B | Q8_0 | 36.9 GB | 121 estimate |
| Qwen3.8-Flash-Next | UD-IQ1_M | 74.5 GB | 10.3 estimate |
| gemma-4-26B-A4B-it | Q8_0 | 26.9 GB | 92.9 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 80.4 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 93.7 estimate |
| Muse-Glimmer-30B | Q8_0 | 29.6 GB | 25.9 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 179 estimate |
| gemma-4-31B-it | Q8_0 | 32.6 GB | 23.5 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 175 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 114 estimate |
Similar hardware
NVIDIA H100 SXM · NVIDIA H100 PCIe · NVIDIA A100 80GB SXM · NVIDIA L40 · NVIDIA A40 · NVIDIA L40S · NVIDIA Tesla V100 32GB · NVIDIA A30