@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20231218

Episode Date: December 18, 2023

- Intel Gen-5 Emerald Rapids launch - New York State's own "Chips Act": $10B for NanoTech Complex with IBM, Micron, others - Quantum Computing market size from Hyperion Research - AI "Benchmarketing": Nvidia takes AMD's bait [audio mp3="https://orionx.net/wp-content/uploads/2023/12/HPCNB_20231218.mp3"][/audio] The post HPC News Bytes – 20231218 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. Intel announced its new Gen 5 Xeon data center chip this week, codenamed Emerald Rapids, and they say the new CPU delivers a 21% performance increase and 36% higher average performance per watt over previous Xeons,

along with built-in AI acceleration in its 64 cores for 40% higher inference on models of up to 20 billion parameters. I went to Intel's media pre-launch event in Oregon two weeks ago that included a surprise appearance from CEO Pat Gelsinger. Shaheen, if Intel's future can be read into the upbeat energy and confidence that he exuded, you'd conclude Intel's prospects look pretty bright. He spoke mostly about Intel's future, which includes a crowded and rapid product roadmap, particularly on the CPU side. He also talked about the Gaudi 3 GPU coming out next year.

And he said Intel is determined to leave its history of late and lacking products behind and regain leadership in the chip industry, a tall task against the formidable NVIDIA and AMD. Yeah, definitely. And they seem to be executing well. Keeping up with all the code names across a dozen chip makers is a task. But if you've been keeping up, the progression here is from Sapphire Rapids, SPR, to Emerald Rapids. Next will come Granite Rapids. That's in 2024. And this chip, Emerald Rapids, comes in a dozen varieties with 32 to 64 cores and 3.9 to 4

gigahertz. Notably, it also supports type 3 memory transactions under CXL 1.1 specifications, gaining access to remote memory in a cache-coherent way. The chip is based on the Intel 7 process. New York State announced partnership with IBM, Micron, also Applied Materials, Tokyo Electron, and other players to invest $10 billion into expanding its nanotech complex in the state capital, Albany. That's also home to SUNY Albany and RPI, two prominent universities. We've talked about it before, leadership in advanced technologies like AI, semiconductors, and quantum computing is correlated and even necessary for economic strength. That means communities, cities, states will not leave it just to the national government, but will invest directly to achieve or retain leadership in these fields. This is another very good example of that.

Yeah, this is almost like New York State's CHIips Act. This is a new high NA EUV center that will drive what the organization said will be the next decade of semiconductor technology innovations. The fab will produce sub-two nanometer chips, and you knew ASML had to be involved here, and indeed, state funds will be used to buy a billion dollar ASML chip making machine from the Dutch company, whose equipment is key to making most advanced chips. Hyperion Research released the results of their new quantum computing market study at the Q2B Silicon Valley conference that I know you attended and spoke at, Shaheen. Hyperion estimates the global quantum market was worth $848 million in 2023, with anticipated growth to $1.5 billion in 2026 for a CAGR of 22%.

Pretty nice growth. Now, you might ask, where is all that $1.5 billion going to be spent? Well, hardware, on-prem as well as cloud, will be 30%, going to $469 million in 2026. All software system and applications is 22% or $338 million. Cloud-based fees by end users will be 15% or $225 million. And professional services will do the remaining 13% or $195 million. Again, this is all by the end of 2026, which is three years from now. By the way, we are ending 2023 with some significant advances in quantum computing, especially in error correction technologies. So 2024 is looking promising. This week also saw another indication that competition in GPUs is heating up and AI is taking HPC into really the fabric of society. AMD and NVIDIA are engaged in some benchmarking, as we used to say, with claims and counterclaims that are getting

attention in social media. The ping pong is really unnecessary since all of these guys will likely sell whatever they can build. So the real battle is getting Wayfair allocations from TSMC. But it does show that GPU competition is not real, and that's an important industry milestone. Yeah, when AMD announced the availability of their MI300 GPU last week, their benchmark showed throughput and latency numbers that were better than NVIDIA's latest H100. NVIDIA took the bait and cried foul in a blog this week, arguing that AMD's results did not use optimized software. And the H100, if benchmarked properly, is two times faster. And AMD replied to that.

All right, that's it for this episode. Thanks so much 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.