@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20250901

Episode Date: September 1, 2025

- MIT Quantum Index report 2025 - AMD+IBM for supers+quantum - Here comes tent-as-a-datacenter - Nvidia Earnings - Hot Chips conference recap [audio mp3="https://orionx.net/wp-content/uploads/2025/08/HPCNB_20250901.mp3"][/audio] The post HPC News Bytes – 20250901 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. Shahin, just as last week we posted a podcast on the Stanford AI Index report. Now MIT has issued their quantum index report, and the introduction touches on two universally acknowledged quantum truths, that the, quote, exact timing of when quantum will become commercially viable is unknown, and that, quote, there are a lot of folks who are interested in what's going on in quantum, but the field is impenetrable to them, according to the report's editor-in-chief. Yeah, the report covers 10 areas and provides written analysis and an interactive. site, really quite a nice collection of data. It covers patents, academic research, venture funding, quantum and corporate communications, policy, workforce, education, public opinion. Number nine,

if you're keeping track, is quantum networking. And number 10 is quantum processor benchmarking. It notes the substantial market momentum in quantum processor performance patents around the world, which have risen five-fold from 2014 to 2024, with China holding 60% of quantum patents as of 2024, followed by the U.S. and Japan, continued substantial investments into the area, continued industry and public buzz around the technology and its intriguing nature and promise,

and a skill set gap that really applies to all these major technology trends. Doug, we've said many times, it's hard to believe that all this activity is going, nowhere. But the reality remains the same. A lot of progress, a lot more left to do, and no killer app yet to drive adoption. Staying with quantum, IBM certainly has established a respected place in the quantum community for its quantum R&D work. And last week, they announced plans with AMD to develop computing architectures combining quantum computers and HPC, what they're calling quantum-centric supercomputing. AMD and IBM are collaborating.

to develop scalable open-source platforms that integrate AMD, CPUs, GPUs, and FPGAs with IBM Quantum to accelerate a new class of algorithms outside the current reach of either paradigm. There's definitely merit to a hybrid supercomputer that integrates CPUs, GPUs, and QPUs, and as you mentioned, IBM and AMD are a good match here and are both doing really well independently. They plan a demo this year to showcase. hybrid workflows. AMD's FPGAs and Fast CPUs have also been used for real-time error correction of QPUs by others. So that is expected to be part of this collaboration as well. There was an interesting story in Fast Company about a new way to build data centers,

massive ones to keep pace with demand for AI infrastructure. Mark Zuckerberg said meta will build five gigawatt data centers out of hurricane-proof tents that can be constructed faster than conventional data center buildings. Shaheen, maybe we should call them disposable AI factories. Wow. These things have a habit of becoming permanent fast, though. It's an interesting development. We know any major effort to build AI capabilities these days

needs to start with securing energy, transmitting that energy, getting and installing hardware, extracting heat from the systems, and transmitting that heat. we should now add the physical construction of a data center to that list. And this method accelerates that, trading off speed for long-term durability, which maybe can be added on the side as they go.

It shows the ends companies will go to to accelerate their AI capabilities. It indicates a stampede, a race against the clock, a higher appetite for risk, and it's all good news for GPU providers. The biggest story last week in HPCAI, and really in tech, and actually in the entire business world at large, was Nvidia earnings, which were blockbuster by any measure, except possibly when measured against Nvidia itself. The company reported record, revenue, and operating profit with sales growth of 56%. It's the ninth straight quarter of more than 50% year-over-year growth, but the quarter had the slowest growth rate in more than two years.

One issue is that sales of AI chips to China have been impeded, and that issue, appears far from resolution. The bottom line is the stock is down about three or four percent from its high earlier last week. Let's end with a mention of the Always Good Annual Hot Chips Conference held on the Stanford campus right across from the Hoover Tower. The show has been growing over time with a broader interest in chips, but the conference remains very much focused on one day of tutorials, two days of technical talks, about a dozen selected research posters,

and a dozen simple vendor exhibits on the lawn outside the lecture hall. No official numbers, but physical attendance seemed about the same, four to 500, and it sounded like another 1,500 to 1,500 people were online, so massive interest in the show. Meta had a very interesting talk about their augmented and mixed reality, a specialized system on a chip for low power, what they called World Lock Rendering, really cool stuff. Google covered ironwood, a chip archery,

architecture and its latest TPU rack scale design as they work to optimize performance per dollar and performance per watt. D-Matrix talked about their in-memory compute cores and fast die-to-dye connectivity, chiplet-based architecture. Other talks covered IBM's Power 11, which mostly follows on from the very nice IBM Power 10, Intel Zeon processor's efficiency cores, e-cores, which looks like a beast, NVIDIA RTX-50-90 for graphics, including, quote, neural rendering, and perhaps taking it beyond graphics. AMD Radion RX-9,000, also for improved ray tracing and AI accelerations for gaming. Condor Computing, the US-based subsidiary of Taiwanese Andes Technology, and its high-performance Risk Five-compatible CPU cores. There were lots of mentions of optical I.O. and interconnects, including by IR Labs, UCIE Optical IO

re-timer chiplet, light matters passage M1,000, a 3D photonic interposer, and Invidio's co-packaged silicon photonics switches, as AI data centers look for multi-building and multi-campus networking. This is reminiscent of the old grid computing that started within HPC. Then we had high-performance Ethernet with Broadcom's Tomahawk Ultra Switch Chip. The Sunday tutorial was on rack scale design, including advancements in liquid cooling and power systems, and featured NVIDIA, AMD, meta, Google, and Microsoft. One big highlight was an excellent keynote by Dr. Atsuyoshi Kweki, CEO of Rapidis, about how they have made so much progress in such a short period of time,

as Rapidis helps to propel Japan in cutting-edge technologies and semiconductor manufacturing. There are sure to be lessons for other countries there. 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. Thank you.