@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20240226

Episode Date: February 26, 2024

- Intel Foundry Event, “Systems Foundry” Era - Nvidia Earnings, Valuation, AI Learning vs AI Inference, In-Memory Computing for AI - HPC in Space, Data Centers on the Moon - ISC-24 Conference [audio mp3="https://orionx.net/wp-content/uploads/2024/02/HPCNB_20240226.mp3"][/audio] The post HPC News Bytes – 20240226 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. Hey everyone, welcome to HPC News Bites. I'm Doug Black. Hi Shaheen. Let's start today with Intel's major event this past week to showcase progress with its chip business. The goal of Intel Foundry is to rival TSMC as a chip manufacturer by 2030. Intel CEO Pat Gelsinger has made the competitiveness of Intel's own chips and chip fabrication for others a cornerstone of his strategy. The Intel event featured big names in the industry, including US Secretary of Commerce Gina Raimondo, Microsoft CEO Satya Nadella, OpenAI CEO Sam Altman, and top executives from MediaTek,

UMC, Broadcom, ARM, Ansys, Synopsys, Cadence, and Siemens Software, basically the supply chain of the chip industry. Yeah, great lineup. And they also announced the $15 billion backlog for the foundry business, which included an order from Microsoft for their 18A process that will be ready by the end of this year. As we've covered on this podcast, Intel is really two companies, and Intel, the fab company, plays a crucial role in the geopolitics of technology. Three notable parts of this announcement include, one, the backlog, which shows market traction. Two, Intel's plan to be a one-stop shop for chip building for customers who want that, covering not just lithography, but also packaging, assembly, and testing.

And the third one is Intel talking about the quote, systems foundry, end quote, era. This is very interesting. When chiplets showed up, they pointed to a future where the chip and the chiplets on it would pretty much be the entire system. Apple's M1 chip already exemplified that concept, and it will become more common in the next five years. Staying on chips, another big story in tech last week was NVIDIA's earnings. It was really amazing to see how all of Wall Street waited in suspended animation for NVIDIA's quarterly announcement. Earnings came in above estimates, so not only did NVIDIA shares take off, so did the rest of the market.

And NVIDIA's valuation now is at $2 trillion. To be sure, there are those who see future chinks in NVIDIA's armor. Some say they want to as well as the AI market transitions from training AI models to AI inference. But the Wall Street Journal had a story over the weekend saying that 40% of NVIDIA's earnings are now on the inference side. So that kind of undermines those who would dearly love to short NVIDIA stock. All that said, many companies, including AWS, Google, Meta, and other hyperscalers are working on their own inference chips and Intel, AMD, Samba Nova, and Grok say they have alternatives to NVIDIA. D-Matrix, a chip design startup founded in 2019, has an interesting thought piece on InsideHPC about the advantages of DIMC, digital in-memory

computing, which they say will cut data movement in inference. And as we know, it's data movement that adds electrical consumption and heat to AI chips and servers requiring more cooling and so forth. GPUs are in high demand and short supply, and that's expected to remain the case for a while. And it disproportionately benefits NVIDIA, whose leadership of the market is secure for the next few years. But as you mentioned, there are a growing number of alternatives and Intel Foundry can inject new supply into the market, which would change the game. HPC in space is in the news again, Shaheen. From Rome, we hear from a company called Leonardo. This is the aerospace company, not the pre-exascale Bull-Saquana supercomputer at the

Chenica HPC Center in Bologna. It has sent a satellite into space with an HPC class data center on board. This is a cloud project called Milsca, M-I-L-S-C-A, for the Italian military that aims to provide the government and armed forces with HPC AI and storage capacity using a constellation of satellites. The company says this will enable faster data processing and availability for users and will make communications less vulnerable to interruption. Yeah, what a great story and further evidence that HPC will be everywhere. There was also a data center-related lunar landing last week, the first U.S. spacecraft on the moon since the Apollo 17 in 1972. This according to the Houston-based Intuitive Machines.

The company's Nova C-IM-1 is the first commercial spacecraft to get to the lunar surface. The spacecraft had tipped over on its side at its landing site near the moon's south pole, but its antenna was pointing in the right direction and its solar cells are operating, so it's functional. NovaSea's cargo has scientific and artistic payloads under NASA's Commercial Lunar Payload Services Initiative. It has an IT infrastructure of its own that's running a virtual data center. The lander is developed by Lone Star Data Holdings. It's the first step for them in a plan to build data centers on the moon. Okay, and before closing, we should add that registration is open for the ISC 2024 conference in Hamburg this coming May.

Right. It continues to be a good time to plan your attendance and trip to ISC. It promises to be another must-attend show. All right, that's it for this episode. Thank you all 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.