Quantum #67
Issue #67 of the weekly HPC newsletter by HMx Labs. New people at HMx Labs, GPU, ASICs and the AI market and a HPC skills diagram.

With everything that’s been going on over the last few weeks and months at HMx Labs I’ve been a little remiss and completely forgotten to introduce everyone to our newest HPC engineer Callum Ward. He’s been with us for a couple of months or so now and some of you may have met him at HPC Club.
In an attempt to make space in my brain (which she is proving very effective at) and sort out our business strategy we’ve also recently taken on Amerah Saleh.
Lastly, we also now have the world’s most experienced HPC apprentice (if you know you know 😆) helping us get Supercomputing Strategy Group sprinting along. I’m sure many of you will be pleasantly surprised to be speaking to him in due course so I won’t ruin the surprise.
Oh and HPC NErD has found a permanent home, keeping an eye on me and making sure I don’t slack off too much too 😁

Rest of the real HPC news from last week is below as usual.
In The News
Updates from the big three clouds on all things HPC. Very quiet this week!

Pretty interesting data driven take here from Next Platform on the future of hardware sales as it relates to AI (and not)

The first half of this podcast about GPUs and ASICs from the hyperscalers is quite interesting though could have done with a bit more depth for my liking.

It will be interesting to see how the hyperscaler silicon in the AI space evolves for sure. As the podcast also clarifies they aren’t really ASICs in the sense you might understand from the use of the word in cryptocurrencies and blockchain and are still closer to a GPU than an ASIC. For now. I guess if we ever see enough standardisation in the processing operations needed by models, we’ll see those operations get embedded into hardware… but if that happens what’s to stop them being embedded into CPUs too…

Interesting take on networks in HPC… one thing it doesn’t discuss, and I think we need to be talking about more, is knowing where your data is and scheduling accordingly. Data locality at a continent, region, zone, rack and node level needs to be exposed and available for use in compute scheduling decisions in the future.

From HMx Labs
Laptops running faster than 120 CPU servers? Only in HPC 😁

Tackling that broad niche problem in HPC and visualising skills

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