@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20231030

Episode Date: October 30, 2023

- SC23 Conference "all in" with Streaming - Intel Financials, On-track with Fab, Progress in AI - Quantum Computing Round-up, Still Early Days - Oxide Computer All-Custom Cloud Computer [audio mp3="https://orionx.net/wp-content/uploads/2023/10/HPCNB_20231030.mp3"][/audio] The post HPC News Bytes – 20231030 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. Hi, Shaheen. Let's start off with SC23. They tweeted out, or should I say X'd out, that the conference is all in on streaming.

They said the entire program will be recorded. This is very important for inclusivity because not everyone who wants to be part of the conference can get to Denver. So we say this is all to the good. That has been a long time coming. With so many sessions, it's a complex, expensive task to physically equip it, manage all the agreements and paperwork, and have the bandwidth to transmit it all.

So here's also a shout out to Synet, the SC network that becomes the fastest network in the world for a couple of weeks. Intel announced their quarterly financials. The reason we're covering it here is to check on chip manufacturing and AI. Intel said it's on track to catch up and lead in fab technology and capacity. There are three phrases you'll hear often, EUV lithography, glass substrates, and advanced packaging. They'll be first to install high NA EUV lithography. That's high numerical aperture extreme ultraviolet with equipment from Dutch company ASML, which uses mirrors from the German

company Zeiss. These mirrors are smaller than 30 centimeters usually, and the flat ones are like 25 millimeters. Why am I talking about mirrors? Because ASML said if you scale their newest EUV mirror to the size of Germany, the biggest bump would measure less than 0.1 millimeters. That shows the precision of these things, man. Anyway, Intel fabs represent over $100 billion in investment, primarily in the US and Europe. They have secured undisclosed, but reportedly quote, major customers for Intel 18A and Intel 3, and partner with Tower Semiconductor for Foundry Services. That business is growing. In AI, Mobileye is growing well, and G with Tower Semiconductor for Foundry Services. That business is growing. In AI, Mobileye is growing well, and Gaudi 2 is starting to get some good marks for deep learning.

We'll cover AI chips in more detail in a full episode of the AdHPC podcast week after next. So we know Pat Gelsinger is an enthusiast who sometimes can anticipate the truth, as they say, but here's a quote that conveys his confidence in Intel's direction. Quote, we are on track with five advanced process nodes in four years. We are hitting or beating all our product roadmap milestones. We are establishing ourselves as a global scale systems foundry for both wafer processing and advanced packaging, and we are driving financial discipline and operational efficiency. Now, Sheen, in quantum news, Atom Computing, whose modality uses arrays of optically trapped neutral atoms, said it will be the first to exceed 1,000 qubits in its quantum

platform to be released next year. IonQ said it's achieved 29 algorithmic cubits on a barium platform, which they say is a quantum first. And we also saw that quantum company D-Wave for the second time may be delisted from the New York Stock Exchange because its stock price was less than a dollar during a 30-day trading period. Yeah, I don't think that D-Wave news is too important, and I expect they'll be fine. They were probably the last quantum company to raise money through a SPAC and became public in the process. 2023 in general is what I call the year of working under the hood. Lots of advances, and it will pay off, but it won't be this year, and it won't be general

purpose. There's no transistor moment yet, not even a VHS versus Betamax moment, but several, as you mentioned, modalities or approaches are making great progress. It's mostly research, and when there is a product, it fuels additional research. You need physical qubits, fidelity, error correction, so logical qubits too, connectivity, so you get a large circuit and better usability, and then the ability to map the algorithm to quantum physics, integrate the system into data centers as an accelerator, which means IO, so you can get data into and out of it, the software stack, including the usual development, administration, middleware, etc., etc.

It's a lot. So as a vendor or practitioner, you need a balance of passion and faith on one end and credibility and progress on the other. And that's really the challenge. In other news, Oxide Computer, which was founded in 2019 and has raised $78 million so far, launched what they call the first commercial cloud computer. By that, they mean a rack-scale design and a programmable interface to the infrastructure, a fully integrated hardware-software approach to go beyond OpenStack, OpenNebula, or CloudStack, or the like. What's interesting to me, Shaheen,

is they have done their own firmware, hypervisor, and control panel, which they say lets them integrate and control every aspect of the system for developers to code and deploy quickly. All right, that's it for this episode. Thanks so much for being with us. HPC Newsbytes is a production of OrionX in association with InsideHPC. Shaheen Khan and Doug Black host the show. Every episode is featured on InsideHPC.com and posted on OrionX.net. Thank you for listening.