@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20250224

Episode Date: February 24, 2025

- New in-house-manufactured Microsoft QPU: Majorana 1 - DARPA US2QC program - (Funding) and harnessing idle GPUs anywhere on the internet - France's CEA advances in fusion energy research [audio mp3="https://orionx.net/wp-content/uploads/2025/02/HPCNB_20250224.mp3"][/audio] The post HPC News Bytes – 20250224 appeared first on OrionX.net.

Welcome to HPC News Bites, a weekly show about important news in the world of supercomputing, AI, and other advanced technologies. Hi everyone, welcome to HPC News Bites. I'm Doug Black of InsideHPC and with me is Shaheen Khan of OrionX.net. Quantum made the news last week with two nearly simultaneous announcements from Microsoft and also from two research universities in China. Microsoft said its Myorana 1 quantum chip is the first to be powered by a new topological core architecture that, quote, will realize quantum computers capable of solving meaningful

industrial-scale problems in years, not decades. Microsoft said that top of conductor is a material that can control my Iran particles to produce more reliable and scalable qubits. She and I should note that Microsoft calls the top of conductor, quote, a new state of matter that isn't a solid liquid or gas. And that to me is quintessentially quantum. You've told us what it isn't, so what is it? Yeah, exactly.

Microsoft's work is technologically excellent, conducted by an excellent team who have stuck to their mission for many years. It is nice to see this milestone. It is a bit reminiscent of the quantum error correction advance we saw from Quera in 2023 and the Willow QPU by Google in 2024. Some of the claims by Microsoft have yet to be validated. You might wonder where the MyAirana chip was fabricated. The answer is very interesting.

It's all in-house with proprietary tools and fab process at Microsoft Labs. The chip is too non-standard. It has to manage both super and semiconductors, uses different materials, it needs new simulation software, and all that adds up to a unique process. We might add another important factor, ready access to HPC systems in Azure. The achievement is significant enough to get noticed

by DARPA, Defense Advanced Research Project Agency in the US, which has a long history of cultivating promising technologies. For example, in the 2002 to 2010 timeframe, DARPA's high productivity computing systems, HBCS, advanced industries work for multi-pedaflop systems for industry and national security. Microsoft pursuing topological and PsiQuantum pursuing photonics have been selected by DARPA

for its Underexplored Systems for Utility Scale Quantum Computing, US2QC program. We've covered that in this program before. That itself is part of DARPA's larger quantum benchmarking initiative, QBI, which aims to verify and validate paths to utility scale quantum systems by 2033. Other approaches include superconducting, trapped ion, neutral atom, silicon spin, and there's interest in other new approaches. We are still far from quantum computing's transistor moment or even VHS Betamax moment, when one

technology wins and drives rapid progress. A useful quantum computer is still about 10 years out based on current roadmaps. AI investment is barreling ahead despite the DeepSeq phenomenon, suggesting possibly that expensive compute power may not be necessary for training large language models. But the focus is shifting to post-LLM model AI, that being reasoning models, which require even more compute than LLMs do. Last week, the Wall Street Journal ran a story on a possible AI data center workaround. It's the notion of distributed networks sharing the compute power of idle GPUs, including those in gaming PCs,

workstations, and servers. There are a number of startups working on setting up these networks. One of them is putting forward the idea of investing in GPUs, or at least a part of a GPU, the way you can invest in partial stock shares and renting them out to end users. City at home and folding at home led the way here well before there was even grid computing. So the idea of harnessing decentralized resources is interesting but not new. And there are, as you mentioned, several initiatives for providing it, including for wireless access, storage, and compute.

And some of them use cryptocurrencies to help with governance and to compensate participants. The ideal workload for these is task parallel, meaning an unending stream of independent tasks where results are aggregated as they come in and you don't care where each task is performed. More recently, we saw websites that used your browser to do crypto mining on your system. That's malware, and malware is the ultimate in ease of use because you don't even know you're using it. So if you make it that easy while also getting permission and making it secure and paying for it, then maybe.

But this also illustrates the inventiveness that springs up whenever there is short supply for an expensive commodity, in this case, GPU compute. Sticking with the AI data center theme and how to power them, there's news about fusion energy from the French Alternative Energies and Atomic Energy Commission, CEA, a significant research center and HPC site, which reported that it maintained a nuclear fusion

plasma for more than 22 minutes. That's a 25% improvement over the previous record. CEA said this demonstrates how control of plasma offers hope that fusion plasmas can be stabilized for greater amounts of time to produce fusion energy. The end goal is to control the plasma which is naturally unstable while ensuring that all plasma-facing components withstand its radiation without malfunctioning. The West Tokamak is run from a CEA facility in the south of France.

The previous record was held by the East Tokamak in China a few weeks before. There are several fusion projects in the world, in the US both in government labs and private companies and startups, in Europe, South Korea, and in China. These are not the same machines and the same approach, but have many similarities. They all struggle to generate more energy than they take in and do it fast enough, big enough, and safe enough to be viable.

Like many other worthy but distant goals, every advance is welcome, it's significant, and will bring with it desirable side effects, even if they do not get us much closer. So we should always remember that these are very difficult grand challenges, and we need time and resources to methodically chip away at them.

Small advances are important, even if we are not anywhere near the destination, and fusion remains many years out. All right, that's it for this episode. Thank you all for being with us. HPC News Bytes is a production of OrionX in association with Inside HPC. Shaheen Khan and Doug Black host the show.

Every episode is featured on insidehpc.com and posted on orionx.net. Thank you for listening.