The SC #116
Issue #116 of your weekly supercomputing news but freshly rebranded. ISC, ARM CPUs, AMD faces off to Nvida and Weka proves the bottleneck is still not compute.
Welcome to your rebranded and updated weekly roundup of news on all things supercomputing with a side of my own opinions. The old name bore little resemblance to what I actually talked about so it’s time for an update!
As a quick reminder, I tend to cover the bits of supercomputing and HPC that are important to people working in the field. The focus is generally more on the tech rather than what it can do. Less drug research and weather simulation and more CPU architectures and schedulers if you see what I mean. Though there might sometimes be a twist of finance thrown in. Sorry, can’t help myself!
Strictly no AI hype! Frankly AI news is so well covered everywhere else, I often gloss over even the supercomputing relevant bits on the assumption you’ll have seen that elsewhere.
First up, its ISC next week though you’d never know it! AI has so completely taken over any form of news cycle, social media feed and even in person conversation that it barely registers. Is anyone even going? 😁 I note the hyperscalers are all notably absent from the sponsor list again.
Last week’s news was quite ARM flavoured. ARM has teamed up with Supermicro as resellers/ system builders of their first CPU. If you’re confused, see my previous take on ARM creating their own CPU [2]. Short version, though, I think it puts them in a much more complicated position. That doesn’t seem to have bothered AWS, who released version 5 of Graviton [3], their ARM based CPUs last week, currently only available in the M flavour (memory optimised) VMs but I’m sure we’ll see the C and R versions pop up shortly.
AMD isn’t taking this lying down though, with claims that its latest “Venice” generation of CPUs outperforms Nvidia Vera ARM based CPUs at rack scale [4]. I am so looking forward to getting my hands and on and benchmarking some of these myself.
Weka (a storage vendor) says they have a way to boost inference performance without buying more GPUs [5]. Not a great surprise really, I can’t remember the last time I had a supercomputing problem that was actually compute and not IO bound!
We still have a few places left for HPC Club so if you’ve not booked your place yet, I wouldn’t hang about [6] or go here for the booking link[7] for those of you behind corporate firewalls that hate anything that doesn’t look like a boring vendor site (and some that do).
All the News in Depth
Cloud and vendor releases for Supercomputing & HPC (and AI too if you change the filters).
[1] ARM teams up with Supermicro for its new CPUs
[2] and my own previous thoughts on ARM making its own CPU
[3] AWS gives us new Graviton 5 ARM CPUs
[4] AMD pushes back on the performance claims of Nvidia’s ARM based Vera CPUs
[5] Weka boosts inference performance
HMx Labs Updates
[6] Details and tickets for HPC Club are now available
[7] Direct link to book if your corporate firewall hates fun
We’re expanding FLOPx to be more useful for more of you, but to do that we need to know which benchmarks you care about
Still hiring for a HPC Engineer (but we’ve filled the marketing / bus dev role)
Off Topic
This bit is new. Not supercomputing related but probably adjacent in some form or another, just things I’ve come across that I think are worth your time.
Has AI solved software? Not so much. I certainly still can’t get it to solve the more complex HPC adjacent problems I throw at it (at the systems as opposed to node level stuff).
Creator of OpenCode on Vibe Coding. He has some really interesting take and I found myself agreeing with a lot more than I expected for someone working inside the AI machine.
Creator of Zig. I’ve never used Zig, I don’t currently have a use for Zig. I still found this very interesting and was surprised how much I had in common with him in terms of his philosophy.
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
