vLLM cuts a DeepSeek V4.1 Engram lookup from 54 ms to 5.1 ms on eight MI355X GPUs

Contributor Fangzhou Ai's 10.6x result covers a 589,824-row lookup against a 94 GiB shared table on eight MI355X GPUs, with the sort path enabled on ROCm from 32,768 rows.

By · Published

AI-generated illustration of black server racks with bundled cables and blue lights beside a glowing grid and streams of blue and orange squares.

Contributor Fangzhou Ai's 10.6x result covers a 589,824-row lookup against a 94 GiB shared table on eight MI355X GPUs, with the sort path enabled on ROCm from 32,768 rows.