@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20240212

Episode Date: February 12, 2024

- Honda taps Cadence Design System digital twin supercomputer for CFD, air taxi R&D - Chip news: Nvidia, TSMC, SKHynix, IBM AIU, OpenAI - Kathy Yelick to Deliver ISC 2024 Keynote on Post-Exascale Computing - Google settles with Singular Computing over claims of stolen AI chip tech [audio mp3="https://orionx.net/wp-content/uploads/2024/02/HPCNB_20240212.mp3"][/audio] The post HPC News Bytes – 20240212 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. Hi, Shaheen. Let's start with a story that you found in the European publication EE News, that Honda is acquiring a digital twin supercomputer from Cadence for multi-physics CFD simulations. It's a turnkey system. Honda has worked with Cadence

for 10 years to accelerate design processes. And get this, Honda will use it to design an air taxi, which is a four-seater self-operating flyer that looks like a combination helicopter and airplane that lands and takes off vertically. Shaheen, if there were any remaining doubts, I think it's safe to say we do now live in modern times. That's right. Jetsons is coming. And it needs multi-physics simulations, obviously. Now, we covered a few weeks ago the acquisition of ANSYS by Synopsys, where all sorts of scientific engineering analysis merges together. Now we have Cadence, another software giant starting out with electronic design automation, this time working

on computational fluid dynamics. Their system has a new GPU resident solver, so it doesn't have to go back and forth as often. The solver is for large EDDI simulations, LES. Their simulation times went from weeks to hours, and it's 20 times better energy efficient compared to its CPU equivalent. Now, an EDDI, by the way, is the twisting, churning, swirling pocket of fluids that wreaks havoc on an otherwise smooth laminar flow, and it can be the start of real turbulence. So needless to say, simulating eddies is complicated, but also essential for complex flow. Their CFD supercomputer is an appliance that is GPU agnostic in principle, and can currently use either AMD or NVIDIA GPUs alongside up to a thousand CPU cores.

While we're on GPUs, this week was an active news cycle for chips. Demand for AI chips is expected to continue strong for years, and that means more and more companies think they have time to enter the fray. The big opportunity is better power efficiency, embedding AI in more things, but really any unique angle that serves a big enough market would do, like better performance by architectural improvements like processor in memory or specializing the chip for a big enough use case like automotive. forming a business unit to build optimized chips for hyperscalers. AI workstations are coming. SK Hynix, the Korean memory chip company, and TSMC are forming an alliance for next-generation packaging. And OpenAI CEO Sam Altman remains on a mission to raise multiple trillions of dollars for AI chips and their availability, though his strategy seemed to have a welcome shift, focusing now on helping TSMC and perhaps others versus actually competing with them head on. Right. This is the chip industry's version of nature hates a vacuum. The vacuum is the

tremendous unmet demand for AI processors. We should also say IBM announced this week that its AI chip, the AIU, which stands for Artificial Intelligence Unit, is powering a cluster at the University of Albany. It's a prototype system on a chip architecture for AI inferencing. Developed over the course of five years, the AIU has 32 AI processor cores and contains 23 billion transistors. And you know, it's worth mentioning IBM's slow, steady return to advanced computing. AI has helped fuel IBM revenue growth and its share price jumped

last month after better than expected earnings. The stock is now at a 10-year high. The International Supercomputing Conference, ISC, announced that Kathy Yellick will be keynoting the event in Hamburg. The conference, coming up in May, is a great snapshot of the global state of HPC, AI, and quantum technologies. Listeners of this podcast will recall the episode in October 2022 when Kathy was our special guest. She is actively involved in creating a strategic vision for the next generation of leadership class supercomputers. Yeah, that was an excellent episode. May 12 is also when the latest top 500 list comes out. And I look forward to that as I do to catching up on the European, Latin American, and Asian supercomputing. There are specific sessions for those.

They also have a lecture series named after Jack Dungara, who was also our podcast guest just after he was awarded the Turing Prize, which is really the Nobel Prize of computer science. So for our listeners, we have two super excellent episodes to look up. Okay, and we should close the loop on a story we mentioned two weeks ago, that a patent infringement lawsuit brought by Singular Computing against Google about low-precision computing has been settled. Yes, as it's all too common, it was settled on the proverbial steps of the court in Boston just before the actual trial.

Also, as expected, they did not disclose terms. But I see it as a win for the plaintiff. Armed with this, they might just be able to pursue other similar cases. All right, that's it for this episode. Thank you all for being with us. HPC News Bites 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.