@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20231016

Episode Date: October 16, 2023

- China Exascale Investment - Samsung 3nm, Intel, TSMC - Exascale Day - Women in HPC, Why Women Stopped Coding - GPU Shortage and Competition, Nvidia H100, AMD MI300, Intel Gaudi-2 [audio mp3="https://orionx.net/wp-content/uploads/2023/10/HPCNB_20231016.mp3"][/audio] The post HPC News Bytes – 20231016 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 with Shaheen Khan. Shaheen, a rare bit of HPC news came out of China last week with a PRC announcement that the country, now standing at about 200 exaflops of aggregate compute power, intends to reach 300 exaflops in 2025. Among other things that we can say about this, it's interesting how the Chinese government can quantify how much

compute power the country has. In the West and the US, no one knows how much total computing resources we have, right? I think it's indicative of the US, no one knows how much total computing resources we have, right? I think it's indicative of the size, maturity, and control of government-driven versus market-driven investments. They're both necessary and common. The Chinese government is reportedly investing heavily, but market maturity is in different league in the US and then in Europe. It's just different approaches and phases of growth. Tracking the aggregate number is useful, but too hard when there are thousands

of private organizations with significant capacity, not even counting many regional and several global cloud providers. What is notable though is supercomputing is becoming critical for global competitiveness for companies and countries. Those with supercomputers have an advantage over those who don't. Related news out of Asia, TSMC and Intel get a lot of airtime for their chip manufacturing, but Samsung cannot be overlooked. It's top three,

not top two. And Samsung reminded everyone. Their three nanometer fab is in production and it has new unnamed but HPC related customer. Getting customers is not surprising, though, when there are many tens of chip companies and projects out there and reserving capacity in a fab is the hard part right now. Yeah, this was a story originally published by a Korean news publication. So let me ask Shaheen, do you have a guess who that HPC customer is? Just kidding. It should be noted that both TSMC and Intel have announced they are on their way to three nanometer and TSMC's big customers are well-known, Apple, NVIDIA, AMD, et cetera. And it's expected that Intel will be the first company for their own high-end fabs until they become part of the Intel Foundry offering.

Now, Wednesday of this coming week is Exascale Day, the number Exascale being 10 to the 18th power. So that's 10-18, October 18th. It's a celebration created by HPE, which makes sense since HPE is building, along with AMD and Intel, the first three US Exascale systems. At InsideHPC, we're planning a full line of Exascale-related coverage all week. Excellent. It's been a clever and fun event that the Cray team did before the HPE acquisition. I personally look forward to Avogadro Day five days later. Just joking. We had a chance to discuss the evolution of Exascale systems with Paul Messina, who led the exascale project and has played a critical role in supercomputing for many decades. That episode should drop this week, so stay tuned for that.

Looking at the Women in HPC event at SC23, I was reminded of an episode in the radio program Planet Money back in 2014, titled When Women Stopped Coding. There was also an NPR segment. It's great journalism and work, and I highly recommend everyone to listen to it. Yeah, the story notes that many early programmers were women, and that the number grew until 1984, when it plateaued and then dropped. The reporters set out to figure out why that happened. And they attributed it to the arrival of PCs, which families tended to give to their boys rather than their girls. And early use of PCs gave boys a really significant leg up on the route to getting computer science degrees. Now, as you said, Gene, supercomputers bring competitive advantage, and they're built from GPUs these days. So the IT world is lusting after GPUs, especially for generative AI workloads.

There's a long lead time and they're expensive and securing them is a big effort. It's a big opportunity that invites competition. NVIDIA's H100 is the target and a lot of vendors are coming after it, including AMD when the MI300 and Intel with Gaudi 2, and more than a dozen other players right there in the hopper, no pun intended. Well, this shortage is really a strategic inflection point in the market. It is causing customers to rethink their software strategy. DOE labs have been ahead of this curve by promoting a diversity of systems, including a testbed at Argonne that we covered in a podcast a few weeks ago. 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.