The SC #138

Issue #138 of The SC. Weekly supercomputing news. Pre SC26 announcements and everything in supercomputing this week that has nothing to do with Jensen Huang.

The SC #138

Buckle up and lock in. Its SC26 this week and there’s already a lot to digest before its even started!

AMD continues to show off its latest 6th gen (codename Venice) CPUs and we have some additional analysis from both Next Platform and STH. In point of fact, we have managed to get our hands on the EPYC 9996 and have results for COREx on it too. Will see if I can get all the ts crossed and is dotted on being able to publish them later this week.

Something that we were asked to do is ensure we are using the latest libraries to take advantage of the potential performance and it seems that those are change are now publicly available too.

AMD have also managed to get their Helios double width racks into Azure. Who are also opening the new HDv2 HPC VM SKU for public preview. If you can get their webform to sign up to work. I couldn’t it. Remnants from last week’s outage perhaps?

Speaking of Microsoft and faux pas, they also published the fantastic headline that their VMs are “in line with industry benchmarks” for LLM inference. Talk about underwhelming 😆

Getting into the slightly more SC26 specific news items next, the team from IsambardAI have published a paper to be presented at the conference comparing the latest Nvidia Vera CPU to AMD’s 5th gen (Turin) and Intel’s Granite Rapids. Again, a bit of a shame that it doesn’t include the latest AMD 6th gen chips but I guess that’s release timings for you.

More SC26 news will be filling your feeds pretty imminently and I shall refrain from repeating the endless press releases that will no doubt have already found their way to your inbox already. I find HPC Wire is pretty good to stay on top of things till Glenn Lockwood gives us his post conference rundown.

I know it’s always down the bottom of the newsletter but this week I do want to call out the Off Topic and From HMx Labs sections. They both have a single article each, and both are a long read. But both are worth your time. They cover related ground that anyone working in tech really needs to be thinking about. Off topic from supercomputing in the strictest sense but honestly I’d read them both above the rest of the news this week. Even if I am a little biased about one of them since I wrote it!

Have a good week all.


All the News in Depth

Cloud and vendor releases for Supercomputing & HPC (and AI too if you change the filters)

More details on AMD's 6th gen CPUs from Next Platform and Serve The Home

You might need to upgrade your compilers and libraries to get the best from them too

The supercomputing team at Isambard have released a paper to be presented at SC26 this week comparing AMD 5th Gen (Turin), Intel Granite Rapids and Nvidia Vera. 

OpenMP Release v6.1

Los Alamos moves to OpenCHAMI (Bristol/ Isambard showing the way!)

Azure releases its new HDv2 VMs. In theory open for public preview but the form to sign up won’t work for me; and this also.

Azure will also be serving up AMD Helios

Also Microsoft makes the staggering claim their VMs are able to perform inference “in line with industry benchmarks”.


HMx Labs Updates

Ok I'll admit this is a heavy read. Its long, it builds on a lot of prior work that it assumes at least a passing familiarity with on the part of the reader (see end notes) but I truly believe that if you're not at least thinking about the impact of this on your own work you risk finding yourself on the back foot and sooner than you'd think.

It's titled Does Innovation Die in the Intelligence Era but it covers the mechanisms of extracting value from your work.

Does Innovation Die in the Intelligence Era?
Copyright, patents and trade secrets in the age of intelligence. A look at how to appropriate value from innovation in an AI world.

Off Topic

Rather than summarise this myself I'll just leave a quote from the article, Human Pilled by Tim O'Reilly, itself:

The game I want us to play is one in which all of us build the future together, so that we control this wonderful new wave of technology for ourselves. I’m starting to see the shape of an economy where AI belongs to people rather than to the big labs, where we have personal AI agents working for us and with us, and we build an economy around that. I think it’s going to happen not because of the big labs, but in spite of them.

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