Decode · tok/s
50
Prefill
—
tok/100W
13.9
Benchmarked Strata with Qwen3.8-Flash-Next IQ3_S on 128GB DDR4 & 1xRTX 5080 at 4K, 32K and 128K context with vision ON. Decode 50 tok/s, Prefill 1,783 - 3,006 tok/s
First-person 5080 box. Rank is the quoted 50 decode. Prefill range is not ranked. VRAM size is the 5080's 16GB; the quote names the card, not the GB.
