community pool / model answer
Qwen3.8-27B-NVFP4-BF16-LMHead
chat · RadixArk/Qwen3.8-27B-NVFP4-BF16-LMHead
183.6 tok/s · model median · 5 runs in the model record.
| Reference rig | Quantization | Median result | Basis | TTFT (ms) | Peak VRAM | Max context | Engines |
|---|---|---|---|---|---|---|---|
| GB10 Grace Blackwell 128GB | 4-bit | 25.3 tok/s | reported | 240 | - GB | 131,072 | sglang |
| RTX 5090 32GB ×2 | 4-bit | 189.6 tok/s | reported | 36 | 57.4 GB | 262,144 | sglang |
| RTX PRO 6000 Blackwell 96GB | 4-bit | 296 tok/s | reported | 47 | - GB | 262,144 | sglang |
Context tested
The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.