@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20251013

Episode Date: October 13, 2025

- AMD OpenAI - AMD Oracle - Intel Clearwater Forest (Xeon 6+) - Intel Fab52 18A looks like is catching up with TSMC - SC25 conference looks like another big one [audio mp3="https://orionx.net/wp-content/uploads/2025/10/HPCNB_20251013.mp3"][/audio] The post HPC News Bytes – 20251013 appeared first on OrionX.net.

Welcome to HPC Newsbytes, a weekly show about important news in the world of supercomputing, AI, and other advanced technologies. Hi, everyone. Welcome to HBC Newsbytes. I'm Doug Black of InsideHPC, and with me is Shaheen Khan of OrionX.net. There were big announcements from two of the big chip companies this week, none bigger than last Monday's bombshell, that OpenAI will build massive data centers with roughly $95 billion worth of AMD GPUs. AMD's stock made a huge move upward, as did AMD's standing in the AI compute market, which of course is dominated by NVIDIA, but now has an increasingly legitimate course also in the race. Shaheen, much ink has been spilt over the past

seven days on this news. I'll be interested in your take. but I believe it was only last week that you and I commented offline that AMD has seemed somewhat quiet of late. Now, it's true that earlier this year, Oracle announced two significant commitments for AMD GPUs, a deal in June to deploy 130,000 on Oracle Cloud. And in March, Oracle during an earnings call, said it planned to deploy 30,000 AMD processors. But the OpenAI deal puts AMD and its MI 450 GPU due out next year on a different plane. Just as Intel couldn't get a better endorsement than Nvidia, when they announced joint activities and Intel stock shot up,

AMD could not get a better endorsement than OpenAI. AMD has been doing well anyway, but this was a welcome recognition of their moves recently in their roadmaps of CPUs, GPUs, and APUs, and crucially, also their rack design capability as a result of their acquisition of ZT systems, which was announced last year and completed in March of this year. AMD's progress and transformation is the stuff of legends already

as it has fought wars on two fronts, with Intel on CPUs and with NVIDIA on GPUs. NVIDIA is growing more and has a lock on capacity up the stack, but it's still early days in AI, and everyone is growing. As the market matures, differentiation becomes harder, and this announcement is a big step in leveling the playing field. Because when you have a de facto standard, it takes a while to get an alternative technology going.

But once you do, it is a lot easier to get additional alternatives. We're seeing that in the CPU world too, with the emergence of arm for laptops and servers and the path that it has paved for Risk 5 chips. All of these guys are competitive because speed is the name of the game. But everyone who brags about performance is also a TSM. or to a lesser extent, Samsung's chip manufacturing customer. That includes AMD, NVIDIA, Qualcomm, Apple, Broadcom, MediaTech, and on and on.

So there's a collective reliance on TSM and Samsung. And later last week, Intel revealed its new Clearwater Forest, Zeon 6 Plus server processor. This 2 nanometer CPU is aimed at the server market and is expected to be shipped the first half of next year. The chip is manufactured at Intel's new FAB 52 using the 18A process node. This is in Arizona near Phoenix. I took a tour of the facility during a pre-announcement press event earlier this month. And for sheer technical complexity, you'd be hard pressed to find anything like it. Fab 52 is a keystone to Intel's plans to build up its U.S.-based chip manufacturing capabilities in competition with TSMC.

In fact, TSM also is building new fabs in the Arizona desert. You'd have to say, Shaheen, that between Fab 52 coming online and the U.S. government investing $10 billion in Intel, new CEO Lipbu Tam, sometimes referred to as LB at Intel, has gotten off to a positive start, even if some of the recent achievements were put in place before TAM got to Intel. As for Clearwater Forest itself, Intel said it has up to 288 e-cores, 17% instructions per cycle uplift over prior generation, and gains in density, throughput, and efficiency. No doubt more performance details will be forthcoming. Yeah, when Pat Galsinger rejoined Intel as CEO in early 2021, he set out his super ambitious plan to catch up with TSM. The tagline then

was five nodes in four years, or five N4Y. Many chip experts considered that to be just impossible, that once you fell behind, it would take decades, if ever, to catch up. And if you ever did, it'd probably be more a result of others making mistakes than you leaping forward. So yes, 5N4Y or even 4N5Y was a tall order. You could say Intel started falling behind in a visible way in 2015, when it delayed its 10 nanometer fab. It did so again in 2020 when it delayed its 7 nanometer fab. So you could also say, the last time Intel had parity with TSA,

was around 2019. It then rapidly fell behind as TSMC and then Samsung continued to execute aggressively. Since then, the very high end in chip manufacturing has been in Taiwan with TSM. Thanks to the Chips Act, TSM has been building factories in the U.S., as you mentioned, but those are a couple of generations behind TSM's most advanced fabs in Taiwan. Right now, two nanometer class chips are where the very high end is, and it looks like Intel, has in fact caught up. TSM's volume manufacturing in two nanometers is later this year in Taiwan, more or less like Intel's is in the US, and with more or less similar technology characteristics, though there will be some variations that the vendors will no doubt highlight.

The next stop is 1.4 nanometers, or 14 Ungstroms. Intel calls that 14A and TSM calls it A14. could be a point of confusion since Apple has chips with a similar name. Chips for iOS devices are named A, and chips for macOS systems are named M. They had an A14 chip a few years ago, but the latest one is the A-19, and thankfully the numbers there keep going up. The waters are stirring in advance of the SC-25 conference. Last year's show in Atlanta had by far the biggest attendance in its 30-year history, as the world has come to appreciate that HPC class technologies are driving the AI revolution. And we expect more of the same this year. We're hearing rumors that the conference venue in St. Louis is getting oversubscribed. We're also starting to hear from vendors

lining up pre-conference announcements. So we'll look forward to another great event and seeing familiar faces in St. Louis next month. Yes, it's conference season. So we should also mention the 2025 OCP Global Summit put up by Open Compute Project in San Jose this week. Now, OCP has been more about cloud-scale systems than HPC per se, but as those worlds converge, it will become more and more relevant. Then we have the Risk Five Summit, October 22nd and 23rd in Santa Clara. And following SC25, December 9 to 11, we get the Q-2B, that's letter Q, digit 2, letter B, Quantum Technology Conference, also in Santa Clara.

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.