vram.run Models Hardware Providers Cloud State of Inference
API provider data is live · Hardware & cloud pricing curated 2026-02-23

A800 PCIe 40 GB

NVIDIA

40GB VRAM · 1560 GB/s bandwidth · 78.0 FP16 TFLOPS · 250W TDP

The A800 PCIe 40 GB has 40GB of VRAM with 1560 GB/s memory bandwidth and 78.0 TFLOPS FP16 compute. At Q4 quantization, it can comfortably run Gemma 3 4B (483 tok/s), Qwen 2.5 7B (254 tok/s), Llama 3.1 8B (240 tok/s). Models larger than ~68B parameters won't fit even at Q4. Electricity cost is approximately $27/month at 250W TDP.

What LLMs can you run?

ModelParamsQ4 WeightFitDecode
Gemma 3 4B4.0B2 GBcomfortable483 tok/s
Qwen 2.5 7B7.6B4 GBcomfortable254 tok/s
Llama 3.1 8B8.0B4 GBcomfortable240 tok/s
Mistral Small 24B24.0B12 GBcomfortable80 tok/s
Gemma 3 27B27.4B14 GBcomfortable70 tok/s
Qwen 2.5 Coder 32B32.5B16 GBcomfortable59 tok/s
Llama 3.3 70B70.6B35 GBwon't fit
Qwen 2.5 72B72.7B36 GBwon't fit
Llama 3.1 405B405B202 GBwon't fit
DeepSeek R1 671B671B336 GBwon't fit

Similar GPUs

GPUVRAMBWTFLOPSTDP
A100 PCIe 40 GB40GB1560 GB/s78.0250W
A100 SXM4 40 GB40GB1560 GB/s78.0400W
M4 Max 36GB36GB273 GB/s36.975W
M3 Max 36GB36GB200 GB/s32.875W
M3 Pro 36GB36GB150 GB/s14.830W

Compare with another GPU

Select another GPU to compare specs and model performance side by side.

Install CLI [email protected] Raw data · MIT · API data: live · HW/Cloud data: curated 2026-02-23 · v0.6.0