=> "ORNL Site Update", 2024 HPC AI User Forum, Sep 4 hpcuserforum.com/wp-content/upl… "Discovery" (OLCF-6) targeted 2027-2028, RfP, Jul 19 (closed Aug 30) x.com/ogawa_tter/sta… DoE Integrated Research Infrastructure (IRI), ASCAC, May 29 Goodbye, Summit! x.com/OLCFGOV/status…

The US DOE has issued an RFI for its new FASST (Frontiers in AI for Science, Security, and Technology) initiative. Only 16 broad questions. I wish I could submit a response as an unaffiliated but concerned citizen! federalregister.gov/documents/2024…

I'll be speaking at Smoky Mountains Conference next week in Knoxville. At the "Exascale Accomplished! What’s next?" session, I'll be talking about how supercomputers like Eagle are a blueprint for how to stop worrying about post-Exascale. tinyurl.com/22dxm2o5

The first author of this astrophysics paper found that if he gave o1-preview the methods section, it was able to reproduce 10 months of work coding he did as a PhD in 5 prompts (a few caveats in the video) Side note: all of your methods sections are becoming instruction manuals.

I don’t think I’ve ever arrived at Knoxville when the sun is still shining. Getting here from California always takes a full day.

This is a great piece: “the story of AI capability has largely been a story of increasing model size, and the sizes of the models follow a generational approach.” And timely! I’ll be talking about how this intersects post-Exascale HPC on Thurs at SMC.

I'm confused by this. OCI said it's taking orders, but this article suggests Oracle is actually building this theoretical max-sized system. Let's be real - if you can afford a 130K B200 cluster, you aren't going to someone who is "taking orders" for them.

Had an amazing chat with Judy Khan Shaw on the NYSE 🏛 floor! 🗽 Supercomputers power breakthroughs in AI, medicine, & more—they deserve more love than just flashy numbers. Check out my first NextSilicon blog on why real-world efficiency is key! 🚀 nextsilicon.com/insights/elads…

Does it violate our corporate social media policy if I retweet a fellow employee?

My first few talks given in front of an audience were only two minutes long, but seemed like an eternity at the time (high school/undergrad). Now I feel like 20 minute talks are barely long enough to stand up and say hello.

This (from someone at OpenAI) is an important thing to understand The major AI labs including OpenAI are very much focused on racing for the future. It is almost accidental that their early products are making billions of dollars. Their goal Is explicitly AGI, not your use case

This reflects my experience. The naysayers often forget that many of the leading companies making foundation models aren't actually trying to make money. Their sponsors are.

An allegory for my career in storage

Glenn K. Lockwood Disk drive exit interviews are brutal

I didn’t expect to see so much HBM at SMC this week. Thanks Elad Raz for letting me have a close-up look at this big fella!

Gave my talk on Eagle and post-Exascale supercomputing in Azure during SMC24’s last session, and now I’ve got an evening to recover from the week in lovely Knoxville. Was great to catch up with many of my old pals in DOE and make some new ones. It’s an unrivaled community.

I’ve long hoped that our nation’s love of money (now pouring in from AI) will overcome its historic fears of nuclear and push zero-carbon base load energy innovation. This isn’t new nuclear, but it’s showing that AI is willing to pay for it.

When I think Intel training, my first thought is Habana, not GPUs. I guess those are going too? “Gelsinger previously said he will not compete with Nvidia and AMD in the AI training space.”