Microsoft announces #Fairwater #AI #supercomputer Over 100,000 GPUs in single system in custom liquid-cooled datacenter 10x performance of world’s fastest public supercomputer Plus attached CPU supercomputer with millions of cores And exabytes of storage. blogs.microsoft.com/blog/2025/09...

Microsoft announces $30B investment in UK, including 23,000 #GPUs of #supercomputers for #AI blogs.microsoft.com/on-the-issue...

Boarding #BritishAirways for Seattle … oh, nope, crew got on then back off again to say the aircraft hadn’t been cleaned at all, so now we all waiting at gate for estimated 40mins delay before boarding while aircraft gets cleaned.

Views of London through the clouds on way in to Heathrow earlier today.

Hearing the announcement: “boarding complete” for on time departure = good. Followed by pilot: “there is a fault with the aircraft so there will be a delay before we can go …” = less good.

Only 2 seats left! The 2025 HPC Leadership Institute in Barcelona is nearly full - register now if you want to secure one of last seats at the most recognized training for current or future leaders across #HPC including #AI, #cloud, #supercomputing, ... www.bsc.es/news/events/...

Learn about Ultra Ethernet 1.0 from some of the authors. Abdul Kabbani from Microsoft and I will be speaking about details of the 1.0 specification on Mon Sep. 8th 10am PST. Join in-person at Microsoft Redmond building 99/1915 (open to public) or online buff.ly/vWIGYls More info: buff.ly/ijiW5so

Bright rainbow yesterday afternoon

Launching Fall 2025: Duke’s Computational & Digital Health Certificate 🎓 Training future leaders at the intersection of AI, modeling, wearables & ethics. Learn more: duke.is/certificate #DigitalHealth #AI @dukeengineering.bsky.social

Wait… HPC does that?! Roco’s back with a new blog exploring the surprising ways #HPC shapes our world, from Formula 1 to farming, perfect pints to dinosaur research. 🦖🍺 Read here ➡️ bit.ly/4mYMQBb

The HPC Leadership Training Institute (Sept. 30-Oct. 2) will be held in Europe this year — hosted by the Barcelona Supercomputing Center in collaboration with TACC, SURF, the University of Luxembourg, and Microsoft. 👉 Learn more and register now! hpc-li.yellenge.nl 👈

Is teaching #programming in hand-coded #Fortran a bit like teaching assembly now? Should all programming teaching now be based on how to write software with intelligent use of #AI coding tools?

Fresh banana chopped on bran flakes with semi-skimmed milk. Yes or no?

Checking my email is a trap

What fraction of work messages you send are you telling someone (politely or not) they are stuck in past / wrong / an idiot? What fraction is that of the messages you wish you could send but can’t / shouldn’t? And what % messages to you are someone trying to tell you the same?

Moon photo using iPhone with cheap telephoto lens attachment

Darwin leaves with a career #LFC p90 xG + xAG of .86 and G + A of 0.70. All anyone will remember is the price tag and the -7.9 np:G-xG underperformance. But consider: Mané's p90 G + A for Liverpool was 0.67. Shearer at Newcastle was 0.63. Darwin's "failure" is better than most people's "success".

It still amazes me how we quickly life went from fear of any sound waking my kids at the crack of dawn on a Saturday to “my dudes, it’s almost noon, get the hell out of bed!”

A question for chip gurus to explain (ie argue 😁) … Purely in architectural terms (ie ignore current success / adoption etc) - which is/was the best architecture: #x86 vs #ARM vs #RISCV vs #MIPS vs #SPARC vs #IA64 vs Alpha - and why? What pros/cons of each make the choice?

As fun as it is to dunk on AI for science, the problem of self-serving science benchmarks is as old as HPC itself. Look at Gordon Bell prizes. It’s computer scientists patting themselves on the back for achieving science outcomes that domain scientists don’t actually care about.

Thought experiment: You have a #supercomputer with infinite FLOPS (FP64 or whatever else you prefer) and infinite memory bandwidth - what is now the biggest bottleneck on your science advances? (Or business value generation, or whatever your supercomputing mission is) #HPC #AI

Procrastinating thought about #supercomputing … When does the world of #HPC lose the most productivity temporarily - during November when everyone is focusing on #SC25 - or during July/August when large numbers of people are taking holidays?

You know when you watched a recording of a meeting back at 2x speed? Anyone else feel like their whole week is being played at 2x speed and many days/meetings are just racing past?

Proclaiming that #AI is a bubble / is useless / is not real etc. because you had recent poor experiences with an #LLM/chatbot is a bit like saying electricity will never be useful because your electric toaster burnt your bread this morning.

If it trains on less than 100,000 #GPUs, is it even real #AI? If it's not written in #Fortran77, is it even real #HPC? If it doesn't use tapes, is it even real #archive #storage? If it doesn't recycle at least 3 previously discarded ideas, is it even credible as "the new game-changer AI chip"?

HPCwire.com is posting the plenary talk summaries and the TPC25.org site will have full video recodings of all the talks.

Yup. The runway/growth for #AI is only just starting. People who mistakenly assume AI is going nowhere because of their limited experience with some chatbots or specific LLM are missing the (much) bigger picture.

When you need to quickly check a spec on some #HPC tech … and @glennklockwood.com’s blog comes higher up the search results page than the vendor’s own website(s) - and presents the required info simply and quickly too.

Podcast: “Supercomputing used to be about going fast in one direction. In AI, that just drives you off a cliff faster.” — @glennklockwood.com Lockwood on why flexibility, not just performance, defines next-gen infrastructure. Now at VAST. shared-everything.simplecast.com/episodes/fro...

#HPC Leadership scores so far Q1: 97% of people got it correct Q2: 76% Q3: 73% Q4: 84% Q5: 89% Q6: 88% Q7: 65% Q8: 78% Q9: 90% Q10: 70% The most selected wrong answer (33% of people) is for Q7. Nearly 20% of people got 10/10. Still nobody who says they don't need the course got 10/10.

So far, none of the people who ticked the Q11 box for “I am already experienced enough, I don't need this training” have scored 10/10. Of those who have scored 10/10, all of them also marked one of the Q11 options saying they would like to attend (even if not possible this year).

Bidding on known-lost causes is really about the opportunity cost. If an RFP requires a team to focus for weeks or months, don’t bother if you know the upside isn’t there.

Some #HPC fun for the week ahead: a quiz on HPC leadership topics forms.office.com/r/vCcEX690h2 (and no mention of #ai or #quantum!)

Sunset over the beach

In the few days I have between jobs, I wanted to share an unvarnished perspective on what I've learned after spending three years working on supercomputing in the cloud. It's hastily written and lightly edited, but I hope others find it interesting: blog.glennklockwood.com/2025/07/less... #HPC

And from last night’s once in a decade hotel to tonight’s hotel that is much more frequent such that some of the staff and I recognise each other now …

When a hotel reception welcomes you back … and then says it is ten years since your last visit!

Has HPC Changed Forever? And What Is Next? Looking forward to giving this keynote talk at the HPDC conference at Notre Dame later this month hpdc.sci.utah.edu/2025/program... #HPC #AI #AIinfra #cloud #supercomputers