Reproduce “Uncovering the Memory Hierarchy of Modern NVIDIA GPUs for Accurate Simulation” 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
Ziwen Wang, Jiapeng Zhou, Changxi Liu, Dongsheng Wang, Yun Chen
GemAtlas improves GPU simulation by probing hidden cache behavior with controlled native instructions and incorporating the inferred mechanisms into a simulator. Its recovered models describe cache placement, eviction, and distributed L2 behavior. Against BL-Impv, average simulation MAPE across four GPUs falls from 31.69% to 22.24%, and for LLM operators from 34.04% to 23.98%; accuracy remains imperfect, especially for some kernels.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.