@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20241028

Episode Date: October 28, 2024

- TSMC delivers good news for chip manufacturing in the US - Denmark builds AI supercomputer with Eviden and Nvidia - White House tasks NNSA/DOE with assessing AI risk for National Security [audio mp3="https://orionx.net/wp-content/uploads/2024/10/HPCNB_20241028.mp3"][/audio] The post HPC News Bytes – 20241028 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 of InsideHPC, and with me is Shaheen Khan of OrionX.net. For those concerned about American chip manufacturing, good news came out of Arizona last week. A story from Bloomberg reported that TSMC's PhoenixFab delivered 4% better yield on advanced chips than fab sites in Taiwan. And this is a pretty impressive achievement. In chip

industry parlance, yield is a quality measurement referring to the percent of non-defective products on a semiconductor. If this is the start of a trend in which TSMC builds high-production advanced chip fabs in the U.S., this would deliver on the strategy behind the Chips Act passed in 2022, whose intent is to expand chip manufacturing capacity in this country and reduce dependence on chips exported from Taiwan, which of course is under threat from mainland China. When the project started a couple of years ago, there was concern by TSMC management that they could not operate with the same efficiency and quality as they do back home in Taiwan. This was despite the fact that the Arizona

factory was not going to be at the leading edge, but a couple of generations behind, so-called N-1 and N-2, with wrinkles ironed out and processes honed. Then we heard of construction and labor challenges, which further reduced expectations. So this news debunks the concerns and confirms the view that U.S. manufacturing very much remains world-class. As for the leading edge, TSMC is working on bringing those to the U.S. manufacturing very much remains world-class. As for the leading edge, TSMC is working on bringing those to the U.S. too, thanks to its participation in the CHIPS Act. A U.S. Department of Commerce announcement in April said there is a non-binding memorandum of understanding for $6.6 billion

in direct funding under the CHIPS Act and that the proposed funding would support TSMC's investment of over $65 billion in three leading-edge Greenfield fabs in Phoenix to manufacture the most advanced semiconductors. This $6.6 billion is comparable to proposed funding for Intel of $8.5 billion, also from the CHIPS Act, announced in March. There are also tax credits for the companies that add to the incentives. Technology is core to policy, and AI is the main course in the technology feast right now. So as AI turns the knob on the information revolution, it's pushing and pulling governments into figuring out what to do or not to do. This week, the White

House signaled a significant policy move with the issuance of a national security memorandum on AI. It calls on the National Nuclear Security Administration, NNSA, to take on safety testing and systematic evaluation of frontier AI models on behalf of the government to assess nuclear and radiation risk and coordinate across the U.S. government on assessments of chemical, biological, and other threats. NNSA is an agency within the DOE, and this is aligned with its existing charter that includes maintaining safety and security of the U.S. nuclear stockpile and helping reduce the global threat from weapons of mass destruction.

The OE has vast knowledge of AI, its underlying science, and potential uses in discovery and national security, so the assignment is well placed. The phrase frontier AI is mentioned several times in the White House announcement. This is to focus the efforts on likely sources of concern, namely advanced AI models that are more capable than existing ones. NNSA has been charged with providing expertise for models and systems, harnessing frontier AI for scientific research and intelligence analysis while evaluating the performance of AI and data sources for training and inference, and to identify critical nodes in the AI supply

chain for potential risks. Finally, the policy assigns an NSA and DOE with streamlining construction of AI infrastructure along with strategies for hiring and retention. Frankly, Shaheen, it all sounds kind of general but also significant, indicating a major push from DOE on AI. It underscores AI's immense importance for national security and global competitiveness. And it recalls Russian President Putin's statement from 2017 when he said that whoever controls AI will control the world. If NVIDIA CEO Jensen Wang can be considered technology industry royalty, and by the way, his company did briefly become the richest in the world last week. Then it's appropriate that he appeared with the king

of Denmark, King Frederick X, to unveil that country's largest AI supercomputer, a system built by Eviden and powered by NVIDIA processors. The system, called Gefeon, is named after the Norse goddess of agriculture and is operated by the Danish Center for AI Innovation that was partially funded by Novo Nordisk. Jensen brought his usual flourish to the occasion when he said, standing alongside King Frederick, that the system is, quote, going to be a factory of intelligence. This is a new industry that never existed before. It sits on top of the IT industry. We're inventing something fundamentally new. It's a nice big supercomputer with 191 previous generation DGX systems that add up to 1,528 H100 GPUs, 382 Xeon Platinum CPUs, and InfiniBand Interconnect, plus evidence cooling technology and system infrastructure software. It will be used for weather forecasting and quantum algorithms,

among other things. Denmark has a very distinguished history in science, with names like Hans Christian Ørsted, who discovered the link between electricity and magnetism, Niels Bohr, of course, who was a Nobel Prize winning quantum physicist and showed up in the Oppenheimer movie. Tico Brahe and his sister, Sophia Brahe, who really founded modern observatory astronomy. And Peter Naur, Turing Award winner, whose name is in the famous Bacchus Naur form, which is used to describe the syntax of programming languages. All right, that's it for this episode. Thank you all for being with us.