@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20240701

Episode Date: July 1, 2024

- Intel's Silicon Photonics Milestone - Intel's 144-core-now, 288-core-next-year Xeon Sierra Forrest CPU - Quantum Advantage: Time vs. Space - Microsoft Concludes Undersea Datacenter Project [audio mp3="https://orionx.net/wp-content/uploads/2024/07/HPCNB_20240701.mp3"][/audio] The post HPC News Bytes – 20240701 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. Hi, Shaheen. We start this week with what has been called an important milestone in silicon photonics. We're talking about Intel's integrated optical IO

chiplet with a CPU. The prototype chip has bandwidth of four terabits per second and consumes about a third of the energy of traditional copper-based interconnects. Shaheen, one of the most popular episodes of the At HPC podcast is our conversation with Professor Karen Bergman of Columbia, an expert and entrepreneur in the field. She told me Intel's news is, quote, an important milestone. Yes, that was episode 54, and you really should go listen to it. The integration of optical and electronic technologies is making great progress. To put this in a chiplet makes it part of the menu of capabilities for large chips. They said it's compatible with PCIe 5, which avoids creating a new standard. Distance is a big limitation for high-end interconnects, and this can go up to 100 meters,

which is plenty for many racks. And at 4 terabits per second, that's 500 gigabytes per second, and a third of energy, it is looking very good. Lots going on in this area with lasers and LEDs and at various layers of a system, but we're still a year or two away from commercial availability, an important milestone indeed. AMD's success has put to rest claims that x86 performance was limited by its architecture. x86 vanquished its competition in the last decade, only to be challenged by ARM and increasingly RISC-V. But it continues to be the mainstream of computing. So where is Intel in all of this, you might ask?

Well, in early 2000s, in the transition from 32-bit to 64-bit, AMD got the upper hand with their Opteron product, but ended up losing the plot as Intel caught up and led with Xeon. AMD got its innovation mojo back and is leading again, while Intel found itself in comeback mode again, and its manufacturing fell behind TSMC. Earlier in June, Intel showed more progress, with announcements that point to a stronger competitive position. Yes, Intel reiterated its plans for their Gaudi 2 and Gaudi 3 GPUs for AI, but the new news was about a 288 E-Core Xeon Sierra Forest CPU early next year, enabling them to tout more hardware threads than the latest AMD Epics.

There is a bit of a core count race going on between those two companies, along with Ampere, and this could be a big deal for Intel. But for now, Sierra Forest has 144 cores, which of course is nothing to sneeze at, but 288 cores remain six to nine months away. With that many cores, the line between CPUs and low-end accelerators starts to blur a bit. So we'll see if the chips have provision to use all those cores in concert. Now, Sheena, I'll be interested in your views on news coming out of a research project by Sandia Labs and Boston University. This is all about quantum versus classical computing, and the research say their work shows quantum delivering superior results

solving a particular type of advanced math problem. But in this case, the quantum advantage isn't faster processing, but the use of far less memory. Possible advantages of quantum computing are speed, accuracy, and energy. This is excellent work that splits the speed advantage into time and space, just processing faster versus requiring less resources. They show an exponential quantum space advantage for a so-called natural streaming problem that includes discrete optimization, implying that the method can be used for many applications. Quantum advantage can be polynomial or exponential. And as it sounds like, exponential advantage is what would make quantum computers leapfrog existing technologies.

You might remember experiments with underwater data centers, containers of computers on the seafloor having no problem staying cool, and maybe even use waves to generate energy. And of course, no worries about real estate. But hold that thought. Microsoft was one of the experimenters with its project Natick. The project started in 2013, deployed a test system off the coast of Scotland in 2018, and has just concluded. It doesn't look like it will be continued, presumably because issues like physical security and cybersecurity, including networking, challenges with upgrading

or maintaining systems, and connectivity costs are prominent. It's notable that just as this project is ending, China began a similar project last year. According to the publication Data Center Dynamics, Microsoft submerged 855 servers, this is the Scotland work that they did, and left them to run for 25 months and only six of the servers broke down. Microsoft stopped using them despite having one-eighth of the failure rate compared to on-land data centers. Another factor is the maturing of liquid cooling that was not available in 2013 when the notion of servers cooled with water or submerged data centers was radically new. Microsoft said it will continue to use Project Matic as a research platform to test new concepts around data center reliability and sustainability. All right, that's it for this episode. Thanks so

much for being with 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.