GPUs & Macs › Nvidia A100
Nvidia A100: local LLM speed
Llama 2 7B (Q4_0) generates 171 tokens/s on the Nvidia A100 (Vulkan), and processes prompts at 7,064 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 171 | 7,064 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 161 | 6,390 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
What fits on the Nvidia A100
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 171 tok/s (Vulkan, source). Fit assumes the whole model in GPU memory with ~8K context.
Memory size unknown for this device.
Similar hardware
Nvidia Tesla T10 · Nvidia Tesla K40 · NVIDIA Quadro P620 · NVIDIA GT 1030 · NVIDIA MX150 · NVIDIA GTX 980 · NVIDIA GTX 1050 Ti · NVIDIA GTX 1650 Ti