AMD Instinct MI350X 288GBBudget pick
B109.3 tok/s decode
~$8,000 MSRP
Can it run?
Tight fit
Using Q4_K_M in vLLM
Fit status
Tight fit
Decode
107.8 tok/s
TTFT
1796 ms
Safe context
17K
Memory
167.6 GB / 180.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Tight fit | 125.8 tok/s | 2239 ms | 34K |
| Chat | C | Tight fit | 125.8 tok/s | 840 ms | 9K |
| Coding | C | Tight fit | 107.8 tok/s | 1796 ms | 17K |
| RAG | C | Tight fit | 125.8 tok/s | 2799 ms | 34K |
| Reasoning | C | Tight fit | 125.8 tok/s | 1819 ms | 17K |
How DeepSeek V2.5 236B (236B params) fits at each quantization level on NVIDIA B200 180GB (180.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 92.0 GB | Low | C40 |
Q3_K_S | 3 | 115.6 GB | Low | C43 |
NVFP4Best for your GPU |
ollama run deepseek-v2.5-236bhuggingface-cli download deepseek-v2.5-236bUpgrade options
~$8,000 MSRP
| 4 |
132.2 GB |
| Medium |
| C44 |
Q4_K_M | 4 | 144.0 GB | Medium | C44 |
Q5_K_M | 5 | 169.9 GB | High | C44 |
Q6_K | 6 | 193.5 GB | High | F0 |
Q8_0 | 8 | 252.5 GB | Very High | F0 |
F16 | 16 | 483.8 GB | Maximum | F0 |