@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20241111

Episode Date: November 11, 2024

- TSMC revenue jump - TSMC barred from making leading-edge chips abroad - AMD-Fujitsu team up for a new superchip? - AMD, Intel, Nvidia datacenter revenue - Green energy momentum in the US [audio mp3="https://orionx.net/wp-content/uploads/2024/11/HPCNB_20241111.mp3"][/audio] The post HPC News Bytes – 20241111 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. We start with chips and TSMC with news that that company has returned to a period of high revenue growth. The company reported that year over year, October revenue grew by 29.2% and the first three quarters grew by more than 31%. This is due in part to TSMC's increasing

its fab capacity for high-end chips and really across the board. Lead time for GPU shipments seem to be weeks now instead of months. The supply chain has loosened up. The digitization of human life will continue and it will need chips, so the bigger long-term trend is there. It was also reported last week that TSMC is barred by Taiwan's technology protection laws from producing its leading-it chips outside of Taiwan. Right now, that would be the 2-nanometer technology chips. Of course, there are other technical and human resource reasons why it is more efficient,

if not necessary, to build the most advanced chips at home first, and then replicate it outside the country after you have the recipe down. Chip architectures are becoming more heterogeneous as we move to tiles and as chips become an integration of capabilities not so concerned with the underlying instruction set. In that vein, AMD might be planning to expand its CPU portfolio beyond x86. Earlier this month, the company announced a partnership with Fujitsu, and at least one tech publication, TechRadar, speculates that AMD may have a, quote, ARM superchip in the works. In the HPC community, we think of Fujitsu for its ARM-powered Fugaku

supercomputer in Japan that was the number one system in the world from 2020 to 2022. The collaboration with AMD would combine Fujitsu's ARM with AMD's GPU technology, aiming, according to TechRider, quote, to build energy efficient and open source solutions. As for Intel and ARM, we know that as of a year ago, Intel CEO Pat Gelsinger was not a fan of Arm. I was at an Intel event last December in which he said, as a foundry, I'll produce whatever you want, but as a chip company, he recommended against Arm. We're seven years into AMD's revival. It was seven years ago AMD sat me down at ISC 2017 in Frankfurt and told me they had an epic CPU coming out that was an Intel killer. Did I fully believe them?

No. At that time, it was hard for industry observers to trust AMD after a string of boom and bust cycles, most recently an extended period of bust. Although, Shaheen, I do recall saying to myself, if they've got a chip that delivers that kind of performance at a lower price than Intel, then they really have something. Well, within two years of that, AMD was not only winning on price performance versus Intel, they were getting repeat orders and expanding their portfolio, boosted by TSMC's superior FAB technology and a solid chip architecture. AMD turnaround hit a new milestone last week when AMD data center revenue for the first time exceeded Intel's in Q3. AMD was at $3.5 billion compared

to Intel's $3.3 billion. How did NVIDIA do, you might ask? NVIDIA's data center revenue in Q3 was $14.5 billion, a record high, a 41% increase quarter over quarter, and a whopping 279% growth year over year. Now, what gets counted under the label of data center is not always apples to apples, but in terms of magnitude, it's pretty indicative. Europe has a good record of pushing for sustainable green energy. Some think it is ahead of the U.S. in general adoption and consistent membership of the Paris Accord, but energy usage is increasingly a function of data center infrastructure, so-called infratech, and in the data center industry, the U.S. sets the agenda. Smarter and more efficient use of

energy and from renewable sources is good for the environment and enables the age of computation to advance. But importantly, it also saves money. We certainly know American hyperscalers, including the cloud service providers, are committed to green energy, and perhaps we're seeing results. Yes, according to a report from S&P Global Commodity Insights, the rising energy demands of global data centers are driving shifts toward renewable power sourcing, with U.S. data centers accounting for some 50% of total U.S. corporate clean energy procurement this year. The number in Europe is only around 20%. They report that U.S. data centers have contracted

at least 50 gigawatts of clean energy through the end of Q3 this year, with solar power being the major source of supply, coming in at 29 gigawatts, followed by wind at 13, and a growing emphasis on hybrid and nuclear solutions. And as we've discussed, Microsoft is getting behind a plan to revive part of the old Three Mile Island nuclear facility. In Europe, data centers and large tech companies have increased their clean energy procurement in 2024 to 5.3 gigawatts compared to 3.7 for all of 2023. So score one for the U.S. data center industry.

Okay, that's it for this week. Thanks for joining 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.