@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20251117

Episode Date: November 17, 2025

- Chip Restrictions in China - Quantum Computing: IBM, Quantinuum, D-Wave, US DOE, UK NMI-Q, Julich - SDSC 40th Anniversary, the original 5 NSF centers - SC25 glance ahead [audio mp3="https://orionx.net/wp-content/uploads/2025/11/HPCNB_20251117.mp3"][/audio] The post HPC News Bytes – 20251117 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 HPC Newsbytes. I'm Doug Black of Inside HPC, and with me is Sheen Khan of OrionX.net. The Wall Street Journal reported that U.S. export controls on advanced semiconductors are severely constraining China's aviation. AI progress and forcing it to dedicate scarce resources from its indigenous chip manufacturing company, SMIC, to Huawei's AI work. We've reported here in the past that lack of access to high-end GPUs has forced Chinese

vendors to use their own slower GPUs, which they've done in large numbers to reach in aggregate the performance they need. This is less efficient in terms of power and cooling, but can work for some applications. But in another report, the journal's investigation showed how a Chinese AI firm accessed NVIDIA's top Blackwell chips via an Indonesian data center despite U.S. export restrictions. The report said NVIDIA sold servers through Averis, a U.S.-based subsidiary of partly blacklisted InSper. Indonesian Telecom Indosat bought 32 GB200 racks, about 2,300 chips. after securing Shanghai AI startup INF Tech as a customer. INF will use the chips for finance and health AI.

The arrangement appears legal because the chips aren't used for military work, but it highlights loopholes in U.S. controls and how Chinese firms increasingly rely on offshore compute to bypass restrictions. We discussed a lot of this in our recent podcast with Dan Naistead, who expanded on this issue, in some detail. It was a fantastic discussion and I encourage you to listen to it. It really shows the complexity of regulating the flow of bits and atoms in our times, gaining access to high-end US chips from within the West and sending the results back or via remote access to Western clouds

via international subsidiaries or friendly partners or outright smuggling and on and on. But the ability to manufacture high-end chips remains an objective for China, as it does for several other governments. So they are reportedly also working on getting more from older technologies by adding more steps to the manufacturing process. This makes it possible to build higher-end chips to a degree, but at a low yield.

The inefficiency you mentioned is real, and it includes reliability issues because of pushing chips outside of their design envelope. The process would require more aggressive bin sorting, which is the process of testing chips, after they are manufactured to see if any of them pass the higher threshold and then put them in their own bin, so to speak. SMIC may have to discard a large fraction of chips and wafers that way.

Up to 95% of wafers was reported, for example. Bottom line, without extreme ultraviolet EUV lithography equipment, China's fabrication limits will persist, and that will widen the gap between them and the West. the volume of high-end chips manufactured in Taiwan, the U.S., South Korea, and Japan is expected to remain a huge advantage over production in China. The quantum sector has been putting out news of late. Here's a rundown of the latest developments.

IBM, which of course has been in the Quantum Vanguard, released a big set of announcements, including their Nighthawk quantum processor, which will have 120 qubits linked with 218 tunable couplers. IBM said this increased cubit connectivity will let it execute circuits with 30% more complexity than their previous processor. IBM also announced Loon, an experimental processor, the company said, has the components needed for fault-tolerant quantum computing. Quantinium announced Helios, a system that the company said features 98 connected physical cubits with single-cubit gate fidelity of just under 100%, and the same for two-cubate gate fidelity across all cubit pairs,

making it, they said, quote, the most accurate commercial quantum computer in the world. Quantinium, which is the product of a merger between Honeywell and Cambridge Quantum, also recently raised $600 million in venture capital. From a policy perspective, Washington is eyeing the quantum competition with China, very closely. China is investing heavily in quantum development. A few years ago, in an interview with a DOE official, the issue was framed as a race to see, quote, who shuts down who first. Earlier this month, the U.S. Department of Energy announced $625 million in funding to renew its

five National Quantum Information Science Research Centers. And the UK two weeks ago announced the National Metrology Institute, Quantum, NMIQ, an international collaboration of the G7 Nations and Australia, to advance global standards and measurement science for quantum technologies. Yes, there's been a steady stream of quantum news. Following the rumors that the U.S. government might want to take an equity stake in quantum technology companies, D-WAVE said they believe there are already use cases for quantum computers within the government, and therefore they prefer that instead of just investing in them for equity, the government would buy their products and an equity stake as part of that deal could be

worked out. D-Wave is the original quantum computer company, having been founded in 1999 in Canada, and then it became a U.S. company after going public in a merger in 2022. Another bit of news was a benchmark published by the ULIC Supercomputing Center and NVIDIA that was run on Europe's X-Scale class supercomputer Jupiter. This one was to simulate a 50-kubit universal quantum computer. Classical simulation of quantum computers demands large memory and processing power. So 50 universal qubits is a new record. It also shows how common GPUs are getting in the quantum world.

Someday we will have CPUs, GPUs, and QPUs all tightly integrated in one system. For now, QPUs are still. still getting ready. We should note that the San Diego Supercomputing Center is celebrating its 40th anniversary this year. It's one of the original five supercomputer centers funded by the National Science Foundation and opened in 1986. The five include SDSC, the National Center for Supercomputing Applications, NCSA at the University of Illinois, CNSF at Cornell, PSC, which is supported by the University Pittsburgh and Carnegie Mellon, and the Von Neumann Center at Princeton. As I was reminded last week, SDSC started with a Kray XMP-48, which, if I remember right, means four vector

processors at just over 800 megaflops of peak performance and eight megawatts of memory. The system was ward addressable, and each word was 64 bits. Well, fast forward 40 years, and their latest system is also a cray. This one, I believe, with AMD APUs, which Cray can configure 96 of per rack. Shaheen and I are both in St. Louis this week for the Super Computing Conference. Always a highlight of the year. Last year, the conference really expanded with a big increase in attendees and exhibitors. We'll see soon if that continues this year.

But as AI continues to explode and because, as many of us believe, AI is an HPC workload, this all means good things for this conference. Later today, of course, the new Top 500 list of the world's most powerful supercomputers will be released, and we will post a special podcast commentary on that later this week. Beyond that, Jean, what are some highlights at this year's SC that you're looking forward to? Well, as you mentioned, and as you know, I believe HPC is essential and foundational in our times. If only because it is necessary as the amount of data increases, it is also the superset and the first stop for many technologies, and that includes AI and quantum.

There are so many good papers about HPC enabling AI and using AI for traditional HPC applications at this conference. The always good student cluster competition, and I'm especially looking forward to technologies that can take the system and data center architecture to the next level. So that also means memory and interconnect technologies. I also expect a lot more quantum work this year. I was happy to see more quantum analytics.

analysts would attend this show. Then there are over 500 exhibitors with lots of advances on display and opportunities to learn from their experiences. So it's impossible to keep up with all of that, but we'll do our best, and as usual, we'll report on the most notable. All right, that's it for this episode. Thank you all for being with us. HPC Newsbytes is a production of Orion X in association with InsideHPC. Shaheen Khan and Doug Black host the show. Every episode is featured on Inside at HPC.com and posted on Orionx.net. Thank you for listening.