@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20231204

Episode Date: December 4, 2023

- Why would The New Yorker cover HPC technologies? - Open Benchmark Council's TOP100 lists - Intel as one of the largest customers of TSMC's high-end fab? - Digital Twins for hyropower at ORNL and PNNL [audio mp3="https://orionx.net/wp-content/uploads/2023/12/HPCNB_20231204.mp3"][/audio] The post HPC News Bytes – 20231204 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. Hey, everyone. Welcome to HPC News Bites. I'm Doug Black. Hi, Shaheen. There was a big and colorful story in The New Yorker this week on NVIDIA and AI and Jensen Wang that drew a lot of attention,

a lot of discussion in the community. Fun reading, curious in some ways though. It characterized Jensen as a, quote, patient monopolist. I'm not sure really what that means. And while it's true NVIDIA dominates the GPU market, there is the emerging presence of Intel Gaudi GPUs and new GPUs coming soon from AMD. Jensen also was quoted kind of dismissing AI as little more than an advancement in data processing. He said, quote, I know how it works, so there's nothing there. It's no different from how microwaves work. Shaheen, is that really true? Well, HPC impacts many aspects of our lives. If that means a more accurate weather forecast or a safer car, the general public experiences the benefits but doesn't need to know more really.

But when HPC enables AI with all its social and global implications, it gets more complicated. AI and by implication, HPC has become interesting to the general public, so it gets attention from non-traditional sources like the New Yorker. But that spotlight comes with serious responsibilities and also strings attached, and you may get quoted out of context in the process. A little bit, yeah. So now HPC looks poised to change many aspects of ordinary life in ways that are not well-defined. It is not just the traditional HPC arena, and it changes the role that the HPC community, its vendors, and its leaders play in society and globally. Now, if any community is qualified to do this well, the HPC community is. And if you ask me, it's just the beginning as the digital

revolution leads to a growing avalanche of data and HPC is the only way to harness it. Separate news, the Open Benchmark Council, benchcouncil.org is their website, is a non-profit focused on performance evaluation. You might like to check out a series of community efforts they have launched in recent months, Bench 100, Open 100, AI 100, and Chip 100, where they tally up the top 100 achievements in each area with an invitation to comment and contribute to these lists. It's a great way to get a perspective and a reminder of important milestones and how we got here. It sounds like a good compliment to MLPerf, Shaheen, an objective source of technology evaluation. Now, there was a story originated from EE News based on a report

from Taiwan-based chip analyst Andrew Liu that Intel will spend $14 billion on manufacturing its new three nanometer chips at TSMC over the next two years. The report said this will likely make Intel TSMC's second largest customer for three nanometer chips in 2025 behind Apple and ahead of AMD. Interesting development. Kudos to Intel for juggling a few complicated and strategic imperatives, becoming a foundry service for any customer, not just internally and at the high end, competing with TSMC and Samsung, becoming a leader in chip performance and price performance in the IT market, competing with AMD and NVIDIA, but also others, and protecting the x86 application base while being open to new capabilities coming from outside that base. Really, really hard to do. So HPC has always been about what we now call a digital twin to mathematically model nature and simulate it as realistically as possible. With new tools, digital twins can now form an ongoing parallel digital universe. Not surprisingly, Oak Ridge National Lab and Pacific Northwest National Lab are driving these concepts. They're working on an open platform framework

research effort for hydropower systems. The Oak Ridge announcement stated that the U.S. hydropower fleet has an average age of 64 years and really requires extensive modernization, which will in turn require sensors, data and control systems, analytics, simulation, cybersecurity, and advanced computing for optimizing hydropower. For all the attention given to generative AI, another killer app is also AI-based digital twins, as you've described it, on the factory floor and power plants, for two examples, where performance, anticipation of repairs and upkeep, and overall maximization of infrastructure can be worked out virtually, then executed in the real world. All right, that's it for this episode.

Thanks so much for joining us. HPC News Bites 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.