@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20250217

Episode Date: February 17, 2025

- Arm to sell its own CPU? Meta as first customer? - Big AI, Big Power, Big Chill... Big Pollution? - Intel's fate puts national security and market forces into focus - Europe falls in-line to avid falling behind in AI - GE Aerospace gains access to exascale power via DOE INCITE program [audio mp3="https://orionx.net/wp-content/uploads/2025/02/HPCNB_20250217.mp3"][/audio] The post HPC News Bytes – 20250217 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 Inside HPC, and with me is Shaheen Khan of OrionX.net. So the Odyssey of Arm Holdings took a new twist last week. The softbank-owned British company has had a chip licensing business model since its founding in 1990 until now, according to press reports that Arm is not commenting on. Arm has been increasingly accepted

in the HPC AI industry and is now reportedly planning to manufacture its own server CPUs and the first such customer will be Meta. ARM made a big splash in HPC in 2020 when Fujitsu unveiled the ARM-powered Fugaku supercomputer, which debuted as the world's most powerful system. Shaheen, I've heard you say many times you're a fan of the Fugaku system design. I sure am. Fugaku and its predecessor, the K supercomputer, are just beautiful, elegant, very well designed. The last 10 years have seen a consolidation of CPU architectures. Many of them left center stage, alpha, MIPS, and then Spark, and to a large extent Power,

while remaining quite interesting technologically. What's left is x86, which is going strong thanks to AMD and TSMC, and recently also Intel again. Then we have ARM, which built on its strengths in mobile devices and moved up the chain. For servers, a dozen companies have pursued it, starting with Calzada, Cavium, and Qualcomm. Today we can count Fujitsu, Amazon, Microsoft, Google, Nvidia, Ampere, as well as Cyperil, Baidu, Huawei, and many more. And then there is Risk 5, which shows up under the hood in a lot of server chips, but also as the main character for companies like Andes, Alibaba, and many more. So in this crowded competitive scene,

ARM is entering. I see this as a continuation of a strategy to enable customers to build their own chips because that's possible now, even at relatively low volumes. You can segment the market into those who have the wherewithal to just license the IP and build their own chips, a market they've probably exhausted by now, and those who would like to build their own chips but need help with design or with manufacturing and packaging or whatnot. It expands the total available market for ARM, and it's also a good move to help block risk five.

We've talked many times about Intel as having two or three major components, fabricating and packaging semiconductors, designing and selling chips, and software to enable chips for devices and systems. We've also talked many times about how hard it would be to break up Intel into two or three parts

and keep them successful. But only the manufacturing part is considered a matter of national security. The Chips Act was a major indicator of that view. So the idea of a breakup remains in the push-pull between national security and market forces. Meanwhile, many countries provide very significant subsidies for local manufacturing of chips. Intel made great progress to regain global leadership and fabrication under Pat Gelsinger, who was CEO for about three years, but abruptly

retired last December. And maybe now we're seeing the reason why. It was reported last week that Intel's board has been pursuing a breakup of the company, including overtures to TSMC, to take an active and possibly controlling role in the spin out of Intel manufacturing operations. While the news helped Intel stock show strong weekly gain, it remains to be seen

whether such a strategy is feasible. A New York Times story reports that the Trump administration has encouraged TSMC to do a deal. The main issue will be how effective the business logic is and how the White House will view the deal if it comes into final form. With big AI comes big power, provisioning and transmitting electricity. It also comes big chill, I guess, extracting and transmitting heat, and now a brand new wrinkle, big pollution. Not exactly smokestacks, but AI factories seem to have more in common with industrial factories than we thought.

A story is emerging from Memphis, Tennessee, where the so-called Colossus AI system is installed for XAI, which aspires to compete with OpenAI, Google, Anthropic, Baidu, and other big players. Seems like the system relies on 15 gas turbines that produce more pollutants, like formaldehyde, than allowed by health standards. The turbines have reportedly been in operation since last summer without public notice or oversight, according to the Southern Environmental Law Center.

The turbines are 20 years old, so maybe new ones would do better, and maybe data centers should be built well out of town. But the story illustrates the responsibility that local regulators and local officials have for the oversight of power plants used by large AI class data centers. There are pronouncements from Washington on energy for big AI and also from state governments, but local authorities play a critical role as well. Notwithstanding all the complexity, the AI race is a full-on race, and Europe is the target of viral narratives as overly regulated and missing out. So instead of falling behind, the EU is falling in line. It announced this week a massive investment

in AI infrastructure, the likes of which we are seeing in the US, China, India, and other countries. At a summit in Paris called AI Action, the EU launched Invest AI, which they build as quote, an initiative to mobilize 200 billion euros for investment in AI, including a new European fund of 20 billion euros for AI gigafactories. Yes, this is in line with the effort pushed by the White House to amass a trillion dollars of private money for new American AI infrastructure. Part of the European announcement is a pledge by France to dedicate to AI a gigawatt of power

from the country's abundant supply of nuclear energy. And that's equivalent to more than 290 utility scale wind turbines. In French President Macron's announcement, he riffed on President Trump's often heard phrase, drill baby drill, only for Macron it was plug baby plug. And Shaheen, that's in the early running for my favorite quote of 2025. Let's end with news from GE Aerospace's announcement that it has successfully

competed for more than 3 million hours on U.S. Department of Energy supercomputers, including the Frontier and Aurora Exascale systems. They're using them to develop their new commercial jet engine architecture to make it more aerodynamic and energy efficient. This has been done under DOE's InSight program, which grants access to HPC systems at DOE National Labs. One aspect of the program that we think is admirable is that research results must be shared with the public. All right, that's it for this episode. research results must be shared with public.

Alright, that's it for this episode. Thanks so much 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.