GPUs & Macs › NVIDIA A30
NVIDIA A30: local LLM speed
Llama 2 7B (Q4_0) generates 132 tokens/s on the NVIDIA A30 (CUDA), and processes prompts at 3,069 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 132 | 3,069 | Q4_0 | CUDA FA on | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 125 | 2,767 | Q4_0 | CUDA FA off | tg128 | llama.cpp #15013 details |
| Llama 2 7B | 117 | 3,158 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 111 | 3,291 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
What fits on the NVIDIA A30
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 132 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-Q6_K | 22.0 GB | 22.9 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 39.8 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 52.9 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 113 estimate |
| Qwen3.6-35B-A3B | UD-Q4_K_S | 20.9 GB | 141 estimate |
| gemma-4-26B-A4B-it | UD-Q5_K_M | 21.2 GB | 77.5 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 52.8 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 61.6 estimate |
| Muse-Glimmer-30B | UD-Q5_K_M | 19.2 GB | 26.3 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 118 estimate |
| gemma-4-31B-it | Q5_K_M | 21.7 GB | 23.3 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 115 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 75.1 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 420 estimate |
Similar hardware
NVIDIA Tesla P40 · NVIDIA Tesla M40 · NVIDIA L4 · NVIDIA A10 · NVIDIA Tesla V100 32GB · NVIDIA Tesla P100 · NVIDIA Tesla T4 · NVIDIA Tesla V100 16GB