Ep. 027 - OpenAI Jalapeño: Better Than Nvidia Blackwell (Accelerators)

This week Bryan, Myron and Jordan discuss our recent article on OpenAI Jalapeño. They cover the performance, architecture, programming model, implications for NVIDIA and more.1:05 (youtube.com/watch?v=hj_ffVldQZA&t=65s) Jalapeno Overview4:22 (youtube.com/watch?v=hj_ffVldQZA&t=262s) Tokens Per Megawatt9:15 (youtube.com/watch?v=hj_ffVldQZA&t=555s) Benchmark Caveats13:48 (youtube.com/watch?v=hj_ffVldQZA&t=828s) The CUDA Moat21:12 (youtube.com/watch?v=hj_ffVldQZA&t=1272s) How OpenAI Did It32:13 (youtube.com/watch?v=hj_ffVldQZA&t=1933s) Samsung HBM442:10 (youtube.com/watch?v=hj_ffVldQZA&t=2530s) AI-Designed Silicon49:24 (youtube.com/watch?v=hj_ffVldQZA&t=2964s) Architecture Deep Dive56:58 (youtube.com/watch?v=hj_ffVldQZA&t=3418s) Doom and Wrap
Full Article: ⁠newsletter.semianalysis.com/p/openai-jalapeno-… (youtube.com/redirect?event=video_description&a…)