AMD Instinct MI350X 288GBBudget pick
C40.7 tok/s decode
~$8,000 MSRP
Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
30.1 tok/s
TTFT
6424 ms
Safe context
20K
Memory
206.9 GB / 256.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs with offload | 30.6 tok/s | 9216 ms | 34K |
| Chat | C | Runs well | 30.6 tok/s | 3456 ms | 11K |
| Coding | C | Runs well | 30.1 tok/s | 6424 ms | 20K |
| RAG | C | Runs with offload | 30.6 tok/s | 11520 ms | 34K |
| Reasoning | C | Runs well | 30.6 tok/s | 7488 ms | 20K |
How Baichuan M3 235B (235B params) fits at each quantization level on AMD Instinct MI325X 256GB (256.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 91.7 GB | Low | D36 |
Q3_K_S | 3 | 115.2 GB | Low | D38 |
NVFP4 | 4 |
Upgrade options
~$8,000 MSRP
131.6 GB |
| Medium |
| D40 |
Q4_K_M | 4 | 143.4 GB | Medium | C41 |
Q5_K_M | 5 | 169.2 GB | High | C43 |
Q6_KBest for your GPU | 6 | 192.7 GB | High | C44 |
Q8_0 | 8 | 251.5 GB | Very High | C44 |
F16 | 16 | 481.7 GB | Maximum | F0 |