The SC #117
Issue #117 of The SC. ISC26 kicks off in Hamburg. My barber gives me quantum computing stock tips. LLNL picks Cornelis over Nvidia for networking. 80.000 cores in a rack? Yes please. FP64 and PF8 square off and I ask about the future of open source.
ISC26 kicks off in Hamburg today! The AI news has even been kind enough to have a small break in its insane pace so that everyone can talk about supercomputing instead 🎉 My worries about the lack of interest are also completely unfounded it seems with early numbers showing that the conference already has more registered participants than last year. HPC Wire has a few themes to look out for, the last of which is quantum computing…
In what felt like a scripted TikTok meme, but I can assure you wasn’t, as I sat in the barber’s chair yesterday, his opening conversation was about how much money he’s made on D-Wave and IonQ stock. I kid you not. Knowing what my day job is, he then proceeded to ask for advice on which quantum computing company to invest in next and which company would win. Have we hit peak hype for quantum already? Or is it genuinely poised to be the next useful thing? I don’t know but I think a lot of the plumbing that quantum computing uses may end up being more valuable (short term) than the quantum computing itself if the plans to use superconducting conventional computing play out.
In less funny but more important news, LLNL & co have selected Intel’s spin off Cornelis Networks for the Lynx supercomputer over the more traditional Infiniband in an announcement no doubt perfectly timed to allow Cornelis to talk about this at ISC. Nicely played. I guess this shouldn’t be too surprising in many regards as they were already using AMD APUs for El-Capitan and don’t seem averse to bucking the AI trend. Good on them!
Fancy over 80,000 CPU cores in a rack? Liquid cooled of course. Now you can with HPE Cray’s new offerings based on AMD’s Venice CPUs. With SSDs on coolers too it looks like a genuinely interesting design. I guess you better start planning for higher power consumption compute dense racks regardless of if you’re adopting GPUs or not. Even with Turin our testing showed 2KW per 1U sled theoretically giving you100KW fully loaded racks. Without a GPU in sight! eeeek.
The FP64 vs FP8 debate is getting lively post the “FP8 is All You Need” paper (see, we can’t even name research papers without a nod to the one that kicked off the current AI boom!) and its subsequent presentation at TPC26. My gut feel here is that FP64 will remain in demand and relevant though I’ll admit I have little to back that up beyond the overall tone of the conversations I’ve had in this space.
There’s another important conversation I think we should be having (maybe at ISC too?) which is around the future of open source software. I think this is even more important for the HPC and supercomputing community than for software in general and I suspect most of us are burying our heads in the sand around it.
See you in Hamburg!
All the News in Depth
Cloud and vendor releases for Supercomputing & HPC (and AI too if you change the filters)
ISC is go with more attendees than last year already signed up.
HPCWire has a few topics to watch for
LLNL Picks Cornelis Networks for its Lynx supercomputer.
80k + cores in a rack with AMD Venice
FP8 or FP64? Which side of the argument do you fall on?
Shrinking Data Centres and using quantum computing plumbing for conventional compute
HMx Labs Updates
I think open source software is more important now than it has ever been
A little out of date now and is it turn out, rather unfounded!
Off Topic
Modern day Luddites. This one is worth a watch, and I do mean watch as opposed to just listen as I first did via Apple Podcasts.
Linus Torvalds compares compilers to AI generation
Will vibe coding be the worst mistake we ever make as an industry?
Ed Zitron said something really interesting in this one. Like Ed, the praise heaped upon me by code generating genies rather rolls off my back. I don’t place any weight on it. I’d go so far as to say I tend to completely ignore it. I assumed everyone did this. Apparently not? Et tu brut? Also, the overall point by Cal Newport around both doom trolling and the backlash to it is very on point.
Know someone else who might like to read this newsletter? Forward this on to them or even better, ask them to sign up here: https://cloudhpc.news
