@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20240304

Episode Date: March 4, 2024

- GPU Shortage, HPE, Dell Financial Results - GPU allocation: CSP, On-prem, AI PCs, Embedded AI - Chip Capacity, Intel Fabs in Germany, Japan Advanced Semiconductor Manufacturing - Singapore AI program, Mid-career Training - AI business justification, AI talent shortage [audio mp3="https://orionx.net/wp-content/uploads/2024/03/HPCNB_20240304.mp3"][/audio] The post HPC News Bytes – 20240304 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. No surprise, we're starting this week with news about chips, specifically the lack of GPUs, along with possible signs that more accessibility may be

in the offing. Related, HPE announced its earnings last week. Revenue came in below expectations, and the company is attributing that to the shortage of GPUs that would otherwise go into customers' AI infrastructures. HPE's year-over-year revenue dropped by 14% to $6.8 billion. Meanwhile, Dell's revenues were also down 11% year-over-year revenue dropped by 14% to $6.8 billion. Meanwhile, Dell's revenues were also down 11% year-over-year, but they improved their operating income by 25% year-over-year. All companies seem to highlight their backlog in AI systems and point to shortages. Yeah, no surprises indeed. There are reports that GPU shortage is starting to ease, that users can leverage cloud

systems or sell their underutilized GPUs to cloud providers who can keep them busy, but I'm not seeing signs of that really. 2024 continues to be a year in which all GPUs from all vendors will be sold one way or another, so it's about how many you can ship in the face of what looks like fixed and locked up capacity right now. Exactly how that capacity is allocated and where these GPUs will get installed is a matter of strategic decisions. Cloud, and especially tier two cloud service providers, CSPs, large customers, on-prem clusters, all the way down to AI PCs, embedded systems, and mobile devices. Sticking with chips and efforts to alleviate supply problems,

Intel and TSMC have announced new fabs. Let's start with TSMC and Japan, which of course used to be a chip manufacturing powerhouse and still commands a good portion of the chip supply chain. And manufacturing in general has always been important to Japanese businesses and the government. But as extreme ultraviolet EUV technology started working, they were disrupted like Intel was. So they're addressing it, working with TSMC. In 2021, TSMC formed a majority-owned subsidiary in Japan called Japan Advanced Semiconductor Manufacturing Inc., or JASM. Minority owners are Sony Semiconductor

Solutions, SSS, Denso, and Toyota. All in all, about $20 billion of investments with strong support from the Japanese government. The fabs would directly create over 3,400 jobs. One smart move they made is in the selection of target markets, automotive, industrial, and consumer, which makes this easier since those fabs will not have to be so leading edge. They'll be so-called N-3 and below, covering the spectrum from 6 to 40 nanometer processes. They also mentioned HPC-related applications, though in that timeframe, more advanced processes will be common for HPC. Their first fabAB is to

start by the end of 2024, and they just announced the second FAB to start by 2027. That all will add up to more than 100,000 12-inch wafers per month, significant net new capacity. Yeah, what's going on with TSMC in Japan is a very important development. It's impressive how quickly they have progressed and a bit of a contrast to TSMC's challenges in their U.S. fab in Arizona. In the case of Intel, the company is going with two new multi-billion dollar sites in Germany that have the capacity to host up to six additional modules in the future. The fabs will accommodate advanced wafer fab tools and are projected to come online in late 2027. Shaheen, you picked up on an announcement from Singapore about a five-year AI initiative aimed at making the country an AI powerhouse that includes money for scholarships, overseas AI internships, and extensive training programs amounting to nearly $750 million. This for a Singapore population of about five and a half million people. Very, very impressive. Notable in that

program is Singapore's, quote, mid-career pathways program for mature mid-career individuals, end quote. The AI initiative is in addition to what they call a ramp-up of their, quote, continuous education and training program under TESA initiative to provide pathways for mid-careerists to transit to AI jobs, end quote. Now TESA stands for Tech Skills Accelerator. I've been saying that governments need to take steps now to mitigate the risks of job losses for AI, and this initiative by Singapore is very nice to see. The skill set challenge of AI is across the board. Implementing AI needs talent, training AI to do a task needs data from humans who do the task now, and learning to perform new jobs that AI will create requires training as well. Yeah, very much along those lines, we ran a news story this week about a survey of IT

managers looking at AI implementation barriers, one of which is not funding. Companies are eager to put money into AI strategies, but they have significant problems dealing with proving the value of AI projects to senior management. That's the most commonly cited hurdle, along with the lack of trained AI talent. Maybe more countries should adopt Singapore's approach. 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.