Oak Ridge National LaboratoryのTeranishiさんをお招きして、情報工学科の近藤研と合同セミナーを開きました。
「Toward Productive HPC for the Post-Exascale Era」
講演者：Dr. Keita Teranishi (Oak Ridge National Laboratory)
日時： 12月2日(火) 10:00 – 11:00
場所：慶應義塾大学矢上キャンパス14-210 (DS 22)
With the U.S. Department of Energy’s Exascale Computing Project now complete, the challenge shifts from reaching exascale to advancing applications and software on rapidly evolving, heterogeneous systems. AI integration and performance portability are often treated as separate goals, but in practice they converge on a single imperative: productivity—time-to-insight per unit of human and machine effort. This seminar lays out a productivity-first roadmap. I’ll show how AI-assisted programming systems—ChatHPC for guided code synthesis, refactoring, and verification —shorten the loop from algorithm design to validated results while preserving scientific trust via automated testing and continuous integration.
I’ll also discuss the rise of Julia and Mojo as high-level languages that retain low-level control. With composable ecosystems, JIT/AOT options, and maturing multi-backend support, they enable single-source paths to CPUs, GPUs, and emerging accelerators, reducing porting costs and long-term maintenance without sacrificing performance. I’ll share recent ORNL activities evaluating their performance and portability across modern HPC platforms. We close with ongoing ORNL efforts: ChatHPC, JACC (Julia for Accelerators), and exploration of LLVM IR as a foundation for portable, robust HPC programming.
