community pool / model answer

DeepSeek-V4-Flash-DSpark

chat · fraserprice/DeepSeek-V4-Flash-DSpark

281.5 tok/s · model median · 6 runs in the model record.

Reference rigQuantizationMedian resultBasisTTFT (ms)Peak VRAMMax contextEngines
GB10 Grace Blackwell 256GB4-bit65.1 tok/sreported46897.3 GB262,144vllm
RTX PRO 6000 Blackwell 96GB ×28-bit253.6 tok/sreported42- GB2,048vllm
RTX PRO 6000 Blackwell 96GB ×48-bit308.2 tok/smeasured39- GB2,048vllm

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.