Radeon Instinct MI25
AMD
16GB VRAM · 436 GB/s bandwidth · 24.6 FP16 TFLOPS · 300W TDP
The Radeon Instinct MI25 has 16GB of VRAM with 436 GB/s memory bandwidth and 24.6 TFLOPS FP16 compute. At Q4 quantization, it can comfortably run Gemma 3 4B (98 tok/s), Qwen 2.5 7B (51 tok/s), Llama 3.1 8B (48 tok/s). Models larger than ~27B parameters won't fit even at Q4. Electricity cost is approximately $32/month at 300W TDP.
What LLMs can you run?
| Model | Params | Q4 Weight | Fit | Decode |
|---|---|---|---|---|
| Gemma 3 4B | 4.0B | 2 GB | comfortable | 98 tok/s |
| Qwen 2.5 7B | 7.6B | 4 GB | comfortable | 51 tok/s |
| Llama 3.1 8B | 8.0B | 4 GB | comfortable | 48 tok/s |
| Mistral Small 24B | 24.0B | 12 GB | tight | 16 tok/s |
| Gemma 3 27B | 27.4B | 14 GB | won't fit | |
| Qwen 2.5 Coder 32B | 32.5B | 16 GB | won't fit | |
| Llama 3.3 70B | 70.6B | 35 GB | won't fit | |
| Qwen 2.5 72B | 72.7B | 36 GB | won't fit | |
| Llama 3.1 405B | 405B | 202 GB | won't fit | |
| DeepSeek R1 671B | 671B | 336 GB | won't fit |
Similar GPUs
| GPU | VRAM | BW | TFLOPS | TDP |
|---|---|---|---|---|
| Quadro RTX 5000 | 16GB | 448 GB/s | 22.3 | 230W |
| GeForce RTX 5060 Ti 16 GB | 16GB | 448 GB/s | 23.7 | 180W |
| RTX A4000 | 16GB | 448 GB/s | 19.2 | 140W |
| RTX A4000H | 16GB | 448 GB/s | 19.2 | 140W |
| Arc A770 | 16GB | 512 GB/s | 39.3 | 225W |
Compare with another GPU
Select another GPU to compare specs and model performance side by side.