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

Radeon Instinct MI250

AMD

128GB VRAM · 3280 GB/s bandwidth · 362.1 FP16 TFLOPS · 500W TDP

The Radeon Instinct MI250 has 128GB of VRAM with 3280 GB/s memory bandwidth and 362.1 TFLOPS FP16 compute. At Q4 quantization, it can comfortably run Gemma 3 4B (1066 tok/s), Qwen 2.5 7B (561 tok/s), Llama 3.1 8B (531 tok/s). Models larger than ~218B parameters won't fit even at Q4. Electricity cost is approximately $54/month at 500W TDP.

What LLMs can you run?

ModelParamsQ4 WeightFitDecode
Gemma 3 4B4.0B2 GBcomfortable1066 tok/s
Qwen 2.5 7B7.6B4 GBcomfortable561 tok/s
Llama 3.1 8B8.0B4 GBcomfortable531 tok/s
Mistral Small 24B24.0B12 GBcomfortable177 tok/s
Gemma 3 27B27.4B14 GBcomfortable155 tok/s
Qwen 2.5 Coder 32B32.5B16 GBcomfortable131 tok/s
Llama 3.3 70B70.6B35 GBcomfortable60 tok/s
Qwen 2.5 72B72.7B36 GBcomfortable58 tok/s
Llama 3.1 405B405B202 GBwon't fit
DeepSeek R1 671B671B336 GBwon't fit

Similar GPUs

GPUVRAMBWTFLOPSTDP
Radeon Instinct MI250X128GB3280 GB/s383.0500W
M1 Ultra 128GB128GB800 GB/s42.5120W
M2 Ultra 128GB128GB800 GB/s54.4120W
M3 Ultra 128GB128GB800 GB/s65.5120W
M4 Max 128GB128GB546 GB/s36.975W

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