Reproduce “The xPU-athalon: Quantifying the Competition of AI Acceleration” 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

Alicia Golden, Carole-Jean Wu, Gu-Yeon Wei, David Brooks

The authors compare eight accelerator platforms for large language model inference across end-to-end workloads and individual operations, measuring latency, throughput, power, and energy. They distinguish prompt processing from token generation and account for communication when models span devices. In a low-batch Llama-3.1-8B comparison, Cerebras’ latency per token was 22.89% of H100’s; this measured result does not establish a general winner.

Summary by alphaXiv.

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