Thank you, @JeffreyBrown, for a wonderful interview with @YoYo_Ma at @PurdueConvos today. It was a beautiful conversation with music that I found inspiring. Also, thank you for acknowledging @PurdueBands marching and alumni bands who bring arts and fun together. Boiler Up!
@glennklockwood Huge congratulations to KAUST and Cerebras! "AI Accelerator" should be considered a marketing term. Yes, the architecture is good for AI, but the architecture should be called a programmable dataflow accelerator or massive micro-core mesh accelerator.
@science_dot @glennklockwood@HPC_Guru@SamPartee @MickeyShaughnes If you are referring to parallel global address space (PGAS) languages, then yes, it has a future in certain parallel computing communities. But doesn't Chapel/PGAS use MPI under the hood?
I will be giving a virtual seminar tomorrow Friday, April 7, as part of the Computational Research in Boston and Beyond (CRiBB) seminar series. The topic is AI Hardware Accelerators Survey. More details here: https://t.co/gpzLihV0aJ
@glennklockwood The SC22 workshop, on which this article is based, covered a wide range of HPC security topics, and many approaches were shared. It was a lively discussion, and it's not surprising that it sounds like a bunch of voices from different perspectives.
@glennklockwood The NIST/NSF effort, on the other hand, is focused on finding the commonalities - the "typical" - among HPC deployments so that NIST can provide guidance to secure each of the four function zones and systems as a whole.