AMD Instinct MI350X 288GBBudget pick
B108.7 tok/s decode
~$8,000 MSRP
Can it run?
Too heavy
Using Q4_K_M in llama.cpp
Fit status
Too heavy
Decode
12.5 tok/s
TTFT
15438 ms
Safe context
5K
Memory
152.5 GB / 48.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | F | Too heavy | 12.5 tok/s | 22456 ms | 10K |
| Chat | F | Too heavy | 12.5 tok/s | 8421 ms | 4K |
| Coding | F | Too heavy | 12.5 tok/s | 15438 ms | 5K |
| RAG | F | Too heavy | 12.5 tok/s | 28070 ms | 10K |
| Reasoning | F | Too heavy | 12.5 tok/s | 18245 ms | 5K |
How Qwen 3 235B A22B (235B params) fits at each quantization level on NVIDIA L40S 48GB (48.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 91.7 GB | Low | F0 |
Q3_K_S | 3 | 115.2 GB | Low | F0 |
NVFP4 | 4 | 131.6 GB | Medium | F0 |
Q4_K_M | 4 | 143.4 GB | Medium | F0 |
Q5_K_M | 5 | 169.2 GB | High | F0 |
Q6_K | 6 | 192.7 GB | High | F0 |
Q8_0 | 8 | 251.5 GB | Very High | F0 |
F16 | 16 | 481.7 GB | Maximum | F0 |
Upgrade options
~$8,000 MSRP
~$15,000 MSRP
B100 192GBNVIDIA upgrade