Reproduce “WaferTetris: Efficient Freeform Mapping on Wafer-Scale Accelerators” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Dayou Du, Congjie He, Andrew Boutros, Luo Mai, Vaughn Betz, Jianyi Cheng

WaferTetris maps PE tasks onto irregular wafer-scale layouts while preserving task code and logical connections. It uses dependency-aware performance estimates to place workloads on scattered free processing elements. On tested Cerebras WSE-3, it reaches up to 4.42× CU.POKer’s system throughput across concurrent decoding requests. These mapping and throughput results are limited to tested WSE-3 workloads.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.