@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20240805

Episode Date: August 5, 2024

- Intel Financial Results - Nvidia Blackwell Delays - Which programming language is most energy efficient? - Discoveries led by exascale software [audio mp3="https://orionx.net/wp-content/uploads/2024/08/HPCNB_20240805.mp3"][/audio] The post HPC News Bytes – 20240805 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 Inside HPC and with me is Shaheen Khan of OrionX.net. Let's start with the intense week for the tech center. The three major chip companies, stocks, have all taken major hits, especially Intel, which reported a second quarter loss of

more than $1.5 billion and announced upcoming layoffs of 15,000 employees. The stock is at its lowest point since 2012, and some analysts and journalists are using language indicating Intel is in steep decline. But it might be a good idea for everyone to read a major news piece in Friday's Wall Street Journal, stating that Intel's core PC and server chip businesses maintain strong customer loyalty, and that many CIOs take a long view of their choice of chip supplier, that switching chip providers can cause major headaches. It's also true that the transition Intel is working toward, the expansion of its foundry business, is extremely expensive, as are the foundries it's building for advanced data center CPUs and GPUs used for AI.

So the Intel obituaries may be a bit premature. Also, as you have said, Shaheen, Intel building fabs that will give us a domestic supply of advanced chips, in addition to those produced by TSMC in Taiwan, is a matter of national security. There was general retreat among technology companies, including AI vendors, by the way, with notable exceptions. The market is unforgiving when a company misses expectations, but the market is not focused on national security either. And right there is a challenge for critical technologies. I hope Intel is left alone and strongly supported to complete its major transformation to catch up and lead in semiconductor

manufacturing. That is a matter of national security. That path was never going to be easy. And while some people considered it impossible, the progress the company has made is substantial and impressive. In fact, a proper analysis must take into account that Intel has a merchant chip side, which has lost some strength because of the semiconductor manufacturing side, which is a matter of competitiveness for the US and the West in general. If you look globally at critical technologies, you will see that national players are often shielded from market forces so they can build strength and eventually lead. Several news articles this week said NVIDIA is delaying the

release of its new Blackwell chip by at least three months. The party line from NVIDIA is they are on track for release later this year, which is what they said when they announced it in March. But when you are part of the Magnificent 7 and your valuation hovers around $3 trillion, any indication of technical challenges is relevant news. This is also a side effect of NVIDIA moving to an annual cadence for chips. So delays become harder to absorb without becoming visible and subject to misinterpretation. And the whole cadence can get out of sync. Yes, mass shipments of Blackwell are now expected in 2025. But you're right that the annual upgrade

schedule changes how customers evaluate obsolescence. NVIDIA's H100 and H200 GPUs are still very powerful and will remain interesting for several years. But when everything is in short supply, a delay impacts overall supply of GPUs. It becomes a question of capacity allocation versus obsolescence. We've talked about infratech emerging as a big requirement for IT deployments, and especially in light of the energy requirements of GPUs. Providing sufficient energy is one aspect that leads to new sources of energy through research. And cooling the systems is another, which has led to a rapid growth of liquid cooling technologies. So we see everything from chilled doors to direct-to-chip liquid

cooling, along with immersion cooling in which servers and other gear literally sit and operate in non-conductive liquids. Touting energy efficiency is also an important marketing requirement because it in fact influences business decisions. But you were saying, Shaheen, that it also includes the choice of software. Yeah, very interesting. Energy is spent at the direction of software, so it makes sense to analyze and optimize that. You can also take a more holistic approach and consider how much energy it takes to write and package an app, and then how that app uses energy. So along these lines, a 2017 study titled, quote, Energy Efficiency Across Programming Languages, end quote, by lead authors from University of Minho in Portugal, looked at 27 languages across programming paradigms,

imperative languages like C, functional languages like Haskell, object-oriented like C++, and scripting like Python. They also considered execution type compiled in a virtual machine or interpreted. And they looked at 10 problems programmed in each and the electricity required to run them. They found C at the top and Python at the bottom. Fortran was number five, which is a bit of a surprise. A more recent study by the same team in 2021 showed changes in the middle, but C and Python continue to be the first and the last. Doug, you've been telling me the Department of Energy has been issuing a series of updates about the software side of its exascale supercomputing program, including software used to

simulate super diamonds, which are significantly harder than natural diamonds pulled from the earth, optimizations of turbine performance on massive wind farms, and three astrophysics codes used to simulate the size of stars and the size of the universe. Very cool projects and good reminder of why HPC remains at the forefront of human discovery. Yeah, it appears that with the Frontier Exascale system at Oak Ridge Lab fully installed and stabilized, we're starting to see the workload results come in, and they're impressive. We'll be talking about the development of exascale class software with our next podcast guest, Mike Heroux of Sandia National Lab. One theme we'll get into

with him is the work he and his team have done with E4S, the software stack for the first U.S. exascale systems, and an effort to provide reusable scientific software components. So we'll look forward to that discussion. Okay, that's it for this episode. Thanks so much 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.