@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20231113

Episode Date: November 13, 2023

- SC23 Starts, TOP500 later today - Storage at Scale, DAOS, Aurora, Linux Foundation - Sandia National Lab and DDN - TSMC 1.4nm, SMIC financials [audio mp3="https://orionx.net/wp-content/uploads/2023/11/HPCNB_20231113.mp3"][/audio] The post HPC News Bytes – 20231113 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. Hi, Shaheen. SC 23 starts today. It's, of course, a highlight of the year.

Last year in Dallas, the first SC to really be out from under COVID, that one drew more than 11,000 people. So possibly this year, it might be even bigger. You know, as you prepare for the conference, I encourage everyone to check out a video panel discussion on the Inside HPC site. We spoke with four industry figures for their ideas of sessions to go to, key topics that will be the focus of discussions, and some of the latest market trends. Also, the new Top 500 list will be released later today. Shane, what are some of your thoughts about this year's conference?

Definitely the Top 500 list, as usual. Every year, I pick one topic and try to go to every talk that covers it to the extent possible. And this year it's CXL and storage again. The configuration of these high-end systems is going to be very interesting as it relates to not just computing, but also storage. Concepts like object storage, computational storage, erasure coding, processing to persistent topology, on and on. And all that at scale is a great challenge. And then all of that designed and optimized for modern media, variously called storage class memory, non-volatile memory express and VME, persistent memory, flash memory, or solid state drives or solid state disks, which both translate to SSDs. So all of that will be interesting.

Along those lines, Aurora Storage is going to be very interesting because it is using DAOs, Distributed Asynchronous Object Storage, at scale. This is an open-source storage software that Intel drove since 2012 and is now moving it to the Linux Foundation, which means the software and the governance are donated to the foundation and Intel remains a significant contributor. Yeah, DAOs is getting support from Argonne Lab and Akta Labs, Google Cloud, and HP, and of course, along with Intel. In their November 9th announcement, Argonne raved about DAOs, saying it has performance 20x greater than our current file systems and will provide better performance across a broad range of I.O. workloads, allowing application science codes to achieve high performance with less effort. Sandia National Laboratories and DDN announced they have partnered over the last three and a

half years on a parallel storage system for scientific workloads. Infineon, they have designed extensions for their new product that targets HPC storage workloads with, they say, simplicity and data management features. Yeah, that's a great example of public-private partnership and co-design. I'm quoting from the Sandia announcement, quote, other aspects of HPC have the leeway to fail from time to time, while data storage must be reliable 100% of the time, end quote. That is so true, and it adds another dimension. You can't drop a packet.

You can't be down. It has to be completely rock solid. Now, there are so many file systems out there and so many storage models out there. Sometimes I joke that you're running out of letters. Pick a letter, add FS to it, and there's a file system with that moniker. And that's because everything is data and its usage can be unique and optimizing it has to fit that uniqueness. Let's move to semiconductor news. It emerged this week that TSMC's next generation 1.4 nanometer fabrication plant will remain in Taiwan. Now, that was the assumption all along, but TSMC is courted

by all manner of countries, and there are complex training, staffing, supply chain issues, so these announcements become significant. We already covered TSMC's financial performance a couple of episodes ago, which was weaker than usual, despite the strength of high-end fabs that are pretty hot. This week, China's largest and leading-edge fab company, SMIC, standing for Semiconductor Manufacturing International Corporation, showed an 80% drop in year-over-year profits in Q3. On TSMC, the company began production of chips using the 3nm process in Taiwan at the end of last year. They were the first company to do so. And TSMC has also scheduled commercial production of the 2nm process for 2025 with the 1.4 nanometer process to follow.

We haven't heard much news of late about TSMC's efforts to build advanced fabs in the US. That's something that given the geopolitical situation, we all have our eyes on. All right, that's it for this episode. Thanks so much for being with us. HPC News Bytes 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.