community pool / model answer

DeepSeek-V4-Flash-0731-ROCmFP3

chat · Lucebox/DeepSeek-V4-Flash-0731-ROCmFP3

77.5 tok/s · model median · 22 runs in the model record.

Answer

78.9tok/s

measuredRTX 3090 24GB4-bit · 20 runs · decodebest cell in the pool

  1. RTX 3090 24GB78.9 tok/smeasured20 runs · peak VRAM no data yet · 24 GB rig memory
  2. Ryzen AI Max 395 128GB · unified memory30.1 tok/sreported2 runs · peak VRAM no data yet
Reference rigQuantizationMedian resultBasisTTFT (ms)Peak VRAMMax contextEngines
RTX 3090 24GB4-bit78.9 tok/smeasured-- GB131,072llama.cpp
Ryzen AI Max 395 128GB2-bit30.1 tok/sreported-- GB8,192lucebox

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.