@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20240722

Episode Date: July 22, 2024

- ORNL’s RFP for post-exascale "Discovery" system - Forced mate in 12 moves? New chess puzzle challenges engines - Solid oxide fuel cell (SOFC) to power GPU Datacenter - Keep your data for 5,000 years, anyone? [audio mp3="https://orionx.net/wp-content/uploads/2024/07/HPCNB_20240722.mp3"][/audio] The post HPC News Bytes – 20240722 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 with Shaheen Khan. We start this week with DOE's RFP, released Friday Friday for its next-generation supercomputer. This will succeed Frontier, which is the first American exascale class system that's been the top-ranked since 2022. The new supercomputer, which will be called Discovery, is expected to deliver three to

five times the speed of Frontier, and delivery is scheduled within three to four years. The deadline for vendors to respond to the RFP is Friday, August 30th. Well, as DOE goes, so goes the rest of the market. So this is very exciting. And to get a better idea of how the DOE is going about it, let me point you to a couple of full-length HPC podcast episodes that we've done. In March, we caught up with Matt Seeger, Project Director for Discovery, which, by the way, is the sixth iteration of the

Oak Ridge Leadership Computing Facility, OLCF-6. That's episode 81. And in episode 30, back in July of 2022, our special guest, Dr. Horst Simon, took us through the RFI that preceded this RFP. A new chess puzzle emerged this week that shows the complicated geometry of a problem space, problems that do not cleanly belong to one class or require a suboptimal path towards the optimal solution. That makes it relevant to AI, implying that the ability to solve some problems comes at the expense of difficulty solving other problems. The chess puzzle was reported by the popular YouTube channel Agadmator, and reportedly computational analysis to the depth of 70 on the website lichess., would not find the solution. It's a very nice puzzle, by the way,

especially if you're interested in chess. Yeah, subsequent work showed that going to a deeper level would find the solution. As you said, that's an example of a problem where a longer context window is the way out. Also note that the video was posted by David Deutsch, a well-known name in HBC, and a professor at Oxford whose seminal work showed how a quantum Turing machine and a quantum algorithm would work, and that was a key development in the modern quantum computing industry. Bloom Energy, which is a solid oxide fuel cell energy company, and CoreWeave, the GPU cloud company, have formed a partnership to deploy Bloom's fuel cells for CoreWeave at a data center in Illinois next year. Now, Shaheen, I'll admit I'm unfamiliar with solid oxide fuel cell energy, referred to in the trade as SOFCs. It is considered one of the most efficient technologies for generating power from hydrogen, natural gas, and other fuels. The cells operate at 800 degrees Celsius

and reduce carbon emissions by two-thirds compared with coal power plants. Well, no energy, no data center. And this is further evidence that securing energy is a prerequisite in building a data center, along with enough network connectivity and the right location for staffing and governance, et cetera. So the typical questions are about commercial viability, scale, transmission, and location, reliability, roadmap, cost, availability, can you even get it, etc. Not unlike buying a new AI system. If you are in the business of owning data centers, you'll likely end up with a mix of energy types, and you'll get the energy wherever you can find it. Taking a very long-term view of things has been a topic for, well, a very long time. I am reminded of the 10,000-year

clock project that Danny Hillis, co-founder of Thinking Machines back in 1983, was driving, and data formats for super long-term storage like RDF or super large data like HDF. Well, Cerabyte, a German company, is introducing technology that etches data on a ceramic-coated glass tablet by rapid-fire laser beams, creating dents that are 50 to 100 atoms thick. It's a clever solution for large enough bodies of data, but you'd still need to store and preserve the tablets and have the means to read them so many years later. For small data, like recovery phrases for cryptographic keys to your digital wallet, they make small fireproof metal devices, for example. Cerebyte was founded in 2022 and stores data for thousands of years while keeping it safe from lost by fire, radiation, and other effects without needing a constant supply of electricity. Current storage methods retain data for around 100 years and need to be updated every 10 years or less.

So I would say, along with solid oxide fuel cells, we should keep our eye on ceramic on glass data storage. Okay, that's it for this week's episode. Thanks so much for being with us. HPC Newsbytes 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.