community pool / model answer
Qwen3.8-27B-FP8
chat · Qwen/Qwen3.8-27B-FP8
80.8 tok/s · model median · 24 runs in the model record.
| Reference rig | Quantization | Median result | Basis | TTFT (ms) | Peak VRAM | Max context | Engines |
|---|---|---|---|---|---|---|---|
| Arc Pro B70 32GB | 8-bit | 54.2 tok/s | measured | 177 | - GB | 32,768 | vllm |
| Arc Pro B70 32GB ×2 | 8-bit | 64.3 tok/s | measured | 94 | - GB | 33,024 | vllm |
| B200 192GB | 8-bit | 106.2 tok/s | reported | 37 | 158.6 GB | 32,768 | vllm |
| GB10 Grace Blackwell 128GB | 8-bit | 14.8 tok/s | reported | - | - GB | 65,536 | vllm |
| H200 SXM 141GB | 8-bit | 252.4 tok/s | measured | 45 | 111.4 GB | 8,192 | vllm |
| RTX 5090 32GB | 8-bit | 16.8 tok/s | reported | 144 | 30.9 GB | 8,192 | vllm |
| RTX PRO 6000 Blackwell 96GB | 8-bit | 100.4 tok/s | reported | 68 | - GB | 262,144 | vllm |
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.