community pool / model answer
Qwen3.8-27B-NVFP4
chat · RadixArk/Qwen3.8-27B-NVFP4
4,617.9 tok/s · model median · 8 runs in the model record.
| Reference rig | Quantization | Median result | Basis | TTFT (ms) | Peak VRAM | Max context | Engines |
|---|---|---|---|---|---|---|---|
| GB10 Grace Blackwell 128GB | 4-bit | 23.6 tok/s | measured | 180 | - GB | 131,072 | sglang, vllm |
| RTX 5090 32GB | 4-bit | 4,662.6 tok/s | reported | 177 | 29.5 GB | 4,096 | sglang |
| RTX PRO 6000 Blackwell 96GB | 4-bit | 97.4 tok/s | reported | 40 | - GB | 2,048 | 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.