Can it run?
Runs well
Using Q4_K_M in vLLM
Fit status
Runs well
Decode
82.0 tok/s
TTFT
2362 ms
Safe context
23K
Memory
175.2 GB / 256.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | B | Runs well | 82.0 tok/s | 3436 ms | 46K |
| Chat | B | Runs well | 82.0 tok/s | 1288 ms | 12K |
| Coding | B | Runs well | 82.0 tok/s | 2362 ms | 23K |
| RAG | B | Runs well | 82.0 tok/s | 4294 ms | 46K |
| Reasoning | B | Runs well | 82.0 tok/s | 2791 ms | 23K |
How DeepSeek V2.5 236B (236B params) fits at each quantization level on AMD Instinct MI325X 256GB (256.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 92.0 GB | Low | D37 |
Q3_K_S | 3 | 115.6 GB | Low | D39 |
NVFP4 | 4 | 132.2 GB | Medium | C40 |
Q4_K_M | 4 | 144.0 GB | Medium | C41 |
Q5_K_M | 5 | 169.9 GB | High | C44 |
Q6_KBest for your GPU | 6 | 193.5 GB | High | C44 |
Q8_0 | 8 | 252.5 GB | Very High | C44 |
F16 | 16 | 483.8 GB | Maximum | F0 |
ollama run deepseek-v2.5-236bhuggingface-cli download deepseek-v2.5-236b