community pool / model answer

Qwen3.8-27B-QUASAR-NVFP4

chat · QUASAR-QAT/Qwen3.8-27B-QUASAR-NVFP4

263.7 tok/s · model median · 2 runs in the model record.

Answer

427.3tok/s

reportedTesla V100 32GB ×44-bit · 1 run · decodebest cell in the pool

  1. Tesla V100 32GB ×4427.3 tok/sreported1 run · peak 117.4 GB · fits 128 GB rig memory
  2. RTX 5070 Ti 16GB ×2100.1 tok/sreported1 run · peak 29 GB · fits 32 GB rig memory
Reference rigQuantizationMedian resultBasisTTFT (ms)Peak VRAMMax contextEngines
RTX 5070 Ti 16GB ×24-bit100.1 tok/sreported-29 GB2,048sglang
Tesla V100 32GB ×44-bit427.3 tok/sreported315117.4 GB262,144vllm

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.