Can Radeon PRO W7900 48G run Devstral Small 1.1 24B?
Comfortable: GGUF Q6_K needs 20.9 GB at 4K, batch 1; headroom 27.1 GB.
Text input, FP16 KV cache, one card. Unified-memory capacity uses the GPU guide’s usable limit; negative headroom may require host offload. Speed source settings may differ from this 4K estimate.
Measured runs
No measured run for this pair. Sizing is an estimate, not a runtime test.
Quant memory at 4K
| Quant | Total GB | Verdict | Weight sizing |
|---|---|---|---|
| GGUF Q4_K_M | 15.7 | Comfortable | Verified file |
| GGUF Q5_K_M | 18.2 | Comfortable | Verified file |
| GGUF Q6_K | 20.9 | Comfortable | Verified file |
| AWQ INT4 | 13.0 | Comfortable | Estimated weights |
Context capacity · GGUF Q6_K
| Context | Total GB | Verdict |
|---|---|---|
| 4K | 20.9 | Comfortable |
| 16K | 22.9 | Comfortable |
| 32K | 25.7 | Comfortable |
| 128K | 42.2 | Comfortable |
Generation speed
GGUF Q6_K: 24–31 tok/s · Derived range