@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20240930

Episode Date: September 30, 2024

- Intel Xeon 6, Gaudi 3, MRDIMM, AMX - Intel 3 Fab, TSMC, Samsung, Chip subsidies - AMD Turin, AMD UDNA - Quantum update [audio mp3="https://orionx.net/wp-content/uploads/2024/09/HPCNB_20240930.mp3"][/audio] The post HPC News Bytes – 20240930 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. Intel remains very much at the top of the news on both the financial, corporate, and product fronts. Last week, Intel announced their performance-optimized Xeon 6th generation CPU, codenamed Granite Rapids, and the AI-focused Gaudi 3 GPU. A version of the new Xeon chip,

codenamed Sierra Forest Forest and optimized for energy efficiency versus performance was already announced a couple of months ago. While the advancements of Intel's corporate redirection are expected in 2025 and 2026, these new products are getting favorable notices from early testers and analysts. The technology review site Foronix reported that they tested Granite Rapids, aka Xeon 6980P, with 128 cores against AMD's EPYC 9684X, and it appears Xeon has leapfrogged its x86 competitor on a composite benchmark by roughly 20%. As for Gaudi 3, Intel said it beats NVIDIA's H100s by about 1.8x for performance per dollar on Meta's Lama 8B LLM, according to the company. Intel did not have comparison data versus Blackwell, NVIDIA's flagship GPU, scheduled for shipment next year.

Intel also declined to share expected Gaudi shipment volume, although they did say they feel comfortable with expected numbers. The Xeon 6 launch is significant. All three products received good marks. GAUDEE 3 for heavy-duty AI, Sierra Forest with so-called efficient cores, and Granite Rapids with performance cores and on-chip matrix instructions and associated two-dimensional registers. This is part of Intel's Advanced Matrix Extensions, AMX, which comes on top of Advanced Vector Extensions, AVX,

and Advanced Performance Extensions, APX. If you can use them, they boost performance on the chip you already have and may even remove the need for an accelerator. Also notable is the use of the new multiplexer combined ranks DIMM, or MR-DIMM for short, which allows improvement in memory bandwidth without having to make room on the motherboard for additional DIMM slots, something that is increasingly difficult as chips themselves grow in size. The announcement also bodes well for the Intel 3 FAB process. The chips show good clock speed and power consumption metrics. On the corporate mergers and acquisition front, there continue to be no real news, just talk of who may have wanted to put money into Intel on their undisclosed terms. There were reports that Intel is on track to receive $8.5 billion in grants from the CHIPS Act. To put this in context, however,

TSMC will also receive a subsidy of some $6.6 billion from the U.S. government. And unofficially, Samsung is also mentioned as a recipient of another $6.6 billion. By contrast, China's official numbers for chip subsidies were $19 billion in 2014, $28 billion in 2019, probably as part of a five-year plan that they have there, and another $47.5 billion in 2024, with a stated goal of achieving world-class status by 2030. Korea has a massive $450 billion program through 2047 for what they call Semiconductor Supercluster Construction Plan, and recently allocated $19 billion for chips. And the European Chips Act set aside $46 billion in subsidies, and Japan has put in $4 billion in one chip company, Rapidus, and $1.5 billion in another. We already covered AMD's announcement at the Computex event in June.

Codenamed Turin, it is the fifth generation of AMD EPYC series based on their Zen microarchitecture. It follows Naples, Rome, Milan, and the trio Genoa, Bergamo, Siena that represented the fourth generation. Torin is coming this year and uses the TSMC N3 fab to provide 192 cores. It will probably equalize and exceed on performance, but Intel is starting to be a real competitor. Yes, AMD is holding an event in a couple of weeks in San Francisco, and it is expected they will have a lot more to say about Turin. Meanwhile, in Berlin at the IFA 2024 Consumer Electronics Show, AMD's GM SVP for Computing and Graphics Business Group said the company will unify its consumer-focused RDNA and data

center-focused cDNA architectures into a unified microarchitecture called uDNA, designed for the company to tackle NVIDIA's entrenched CUDA ecosystem. The announcement comes as AMD has decided to deprioritize high-end gaming graphics cards, as we've mentioned in previous episodes of News Bites. On the quantum front, trapped ion quantum company IonQ announced a $55 million contract with the United States Air Force Research Lab in a partnership for quantum scaling, networking, and deployability, along with compatibility with existing telecommunications infrastructure. The contract will be delivered over four years and follows a $20 million plus contract between IonQ and the Air Force Lab last year. The Air Force Research Lab, AFRL,

is really at the pinnacle of all segments of quantum tech, quantum sensing and measurement, quantum communication, and quantum computing. They are just excellent, and they are a major funder of R&D and enabler of this industry, as well as being a leading center of excellence themselves. Quantum tech continues to attract funding and researchers and global and geopolitical attention. So in similar news, IQM, based in Finland and expanding, won a 5 million euro contract to deliver what was billed as the first quantum computer in the Czech Republic. AQT, an Austrian company, installed a trapped ion quantum computer at Leibniz Research Center, LRZ, the well-known supercomputing site in Germany. Germany also has a 35 million euro project that they have split up into three sub-projects,

all focused on miniaturized quantum technology for defense, security, and civilian applications. They awarded one of the sub-projects to Quantum Brilliance and Parity QC to develop the world's first mobile quantum computer by 2027. And this does not include a steady stream of quantum funding rounds and governments recognizing that they need to open their checkbooks and participate. All right, that's it for this episode. Thank you all for being with us. HPC Newsbytes 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.