Intel Innovation 2021 Event: The 10 Biggest Announcements To Know

Intel CEO Pat Gelsinger and other executives laid out the chipmaker’s new ‘software-first’ strategy with a series of announcements and disclosures spanning much of the company’s portfolio of silicon products at the inaugural Intel Innovation event in late October.

ARTICLE TITLE HERE

‘No Gimmicks. No Marketing Fluff. Just Geeky Goodness.’

Intel CEO Pat Gelsinger said he is renewing the chipmaker’s commitment to developers as part of his new “software-first” strategy that was first laid out in CRN’s October cover story.

Gelsinger and other Intel executives laid out this “software-first” vision with a series of announcements and disclosures spanning much of the company’s portfolio of silicon products during the inaugural Intel Innovation event this week, which is a reimagining of the defunct Intel Developer Forum.

id
unit-1659132512259
type
Sponsored post

[Related: Pat Gelsinger: Intel Will Be ‘More Ecosystem-Friendly’ Than Nvidia]

“Through the developer community, we are entering a renewed era of innovation. It‘s all about you, the developers, software and hardware,” Gelsinger said during the opening keynote on Wednesday. “Our recent actions in the ecosystem must be built on core values that are essential to you. But we haven’t done a great job recently. We need to stay connected with you through the evolution of that ecosystem. Today, this changes. We can and we will do better.”

To make Intel the silicon platform of choice in the face of increasing competition, Gelsinger said the chipmaker’s “software-first’ strategy will be built on three tenets: openness, choice and trust. This means “unlocking the garden gates for developers,” “designing for multi-vendor, multi-cloud silicon platform software solutions” and having the best of intentions for developers.

“We have your best interest as a developer at heart, and we will deliver solutions with world-leading security features, anchored to product truth,” he said. “No gimmicks. No marketing fluff. Just geeky goodness.”

What follows are the 10 biggest announcements made by the semiconductor giant at the Intel Innovation 2021 event, including the launch of 12th-generation Intel Core CPUs, new disclosures around the Arc Alchemist discrete GPU, a new development partnership with Google Cloud, the launch of new data science solutions for AI development and big new ambitions for the years to come.

12th-Gen Intel Core ‘Alder Lake’ Desktop CPUs Launch

Intel used the Intel Innovation event to mark the launch of its 12th-generation Core CPUs for desktop PCs, the first to use the chipmaker’s new Alder Lake hybrid architecture. The six unlocked models in the lineup make use of two core types, a performance core and an efficiency core to deliver major advancements in both areas. This allows Intel’s new Core i9-12900K, which has 8 performance cores and 8 efficiency cores, to deliver nearly 50 percent multi-threaded performance at a peak power of 241 watts compared to the previous-generation Core i9-11900K at a peak power of 250 watts. Conversely, the Core i9-12900K can provide the same level of multi-threaded performance at roughly one-fourth the power of its predecessor. Intel said the Core i9-12900K can deliver double-digit performance gains over AMD in select games, but it acknowledged that the tests were done before patches were issued to fix performance issues in Windows 11 for AMD CPUs.

Intel Offers New Details for Arc Alchemist Discrete GPU

Intel teased a few more details about its first Arc discrete GPU for gaming and content creation PCs, code-named Alchemist, that is due out in the first quarter of 2022. The chipmaker said the “flagship configuration” of Alchemist will feature 32 Xe cores, with each core consisting of 16 Vector Engines and 16 Matrix engines, bringing the GPU to a total of 512 execution units. In addition, the GPU will feature hardware support for ray tracing, mesh shading, variable rate shading and dedicated matrix engines called XMX for AI acceleration, the latter of which will enable XeSS, Intel’s answer to Nvidia’s DLSS super-sampling technology. The company also confirmed that Alchemist will support Deep Link technology, which allows applications to take advantage of the discrete GPU and the CPU’s integrated graphics at the same time for better performance. The first advertised capability for Deep Link in Alchemist is Hyper Encode, which will speed up the process of transcoding a single video file.

Intel Partners With Google Cloud On ‘Mount Evans’ IPU

Intel announced Google Cloud as a co-development partner for its Mount Evans ASIC-based infrastructure processing unit. The infrastructure processing unit, or IPU, is a new class of data center chip under from Intel that is meant to offload infrastructure tasks from server CPUs. Intel said the Mount Evans IPU includes an infrastructure programmer development kit that is made to simplify developer access in Google Cloud data centers. Intel debuted IPUs a few months ago, saying their offloading capabilities can allow cloud service providers to “maximize data center revenue” by letting them “rent 100 percent of their server CPUs to customers.” The company previously disclosed it is working on two IPUs, the ASIC-based Mount Evans and Oak Springs Canyon, an IPU that is based on the Intel Agilex FPGA. An Intel executive told CRN that while the IPUs are initially being developed for cloud customers, they will eventually become applicable to a wider swath of data center customers.

AWS Launches Instances With Habana Gaudi Chips From Intel

Amazon Web Services announced the general availability of its new EC2 DL1 instances that are powered by the Gaudi AI accelerators from Intel’s Habana Labs subsidiary. The launch happened on the eve of the Intel Innovation event and marked a major milestone for Habana Labs, which Intel acquired for $2 billion in December 2019 before deciding to end development on the Nervana AI chips that stemmed from a previous acquisition. AWS said the Gaudi-powered EC2 DL1 is the first AI training instance to not use GPUs and said it will offer up to 40 percent better price-performance compared to instances using Nvidia GPUs. The EC2 DL1 instances are arriving a few months later than originally promised by Intel, which said last December that they would launch in the first half of 2021.

Intel Announces New Data Science Solution With OEMs

Intel announced a new set of data science solutions for “out of the box” AI development that will combine Linux-based workstation PCs from Dell, HP and Lenovo with the Intel oneAPI AI Analytics toolkit. The chipmaker said Lenovo will launch the first solution, which will come with the oneAPI AI toolkit pre-installed alongside a custom Linux operating system built on top of Ubuntu. Intel also said it has entered a new collaboration with Microsoft that will enable “the best developer experience” when using Microsoft Visual Studio, WSL2 and Intel’s oneAPI toolkit for machine learning and AI development.

Intel Boosts Developers With Developer Zone, GameDev Program

Intel is pledging more resources for developers of all kinds with the new Intel Developer Zone and Intel GameDev Program as part of the chipmaker’s new “software-first” strategy. Intel Developer Zone is a new online portal consolidates access for previously disparate resources for developers, including code samples, reference designs, Intel Dev Cloud and the Intel oneAPI and OpenVINO toolkits. The Intel GameDev Program, on the other hand, is meant to provide an “improved set of best-in-class services” for game developers of all levels. This includes 24/7 technical support, Intel Graphics Performance Analyzers tools and an early access program for Intel hardware and other things.

Intel Partners With Chipmaker SiPearl on European Supercomputers

Intel announced that European chipmaker SiPearl has chosen the semiconductor giant’s upcoming Ponte Vecchio GPU as the high-performance computing accelerator for future exascale supercomputing projects in the continent. SiPearl plans to use Intel’s oneAPI toolkits and unified programming model to tie together the computing environments of Ponte Vecchio with the high-performance, low-power Rhea microprocessor SiPearl is developing for exascale supercomputers, according to Intel.

Intel Now Expects Not 1 But 2 Exaflops With Aurora Supercomputer

Intel said it now has higher expectations for the computational output of Aurora, the forthcoming supercomputer from the U.S. Department of Energy that will use next-generation Xeon Scalable CPUs and Intel’s Ponte Vecchio GPUs. The company said Aurora will exceed two exaflops of peak double precision compute performance when the supercomputer powers on. That’s double what Intel initially promised when it announced in 2019 that its next-generation CPUs and GPUs would go inside Aurora.

Intel Targets 30X AI Performance Gain With Next-Gen Xeon Scalable CPUs

Intel said it’s aiming to provide 30 times greater AI performance with its next-generation Xeon Scalable server CPUs than the previous generation. The next generation of Xeon Scalable, code-named Sapphire Rapids, is due out next year, and Intel said it will achieve this major leap in AI performance through a combination of software and hardware optimizations. This includes the new Advanced Matrix Extensions engine that will debut in Sapphire Rapids alongside the Intel Neural Compressor, the latter of which “automatically optimizes trained neural networks with negligible accuracy. The company also said Sapphire Rapids can perform faster than Nvidia’s A30 GPU on the ResNet50 image recognition benchmark, achieving 24,000 images per second versus Nvidia’s 16,000 images per second.

Intel Pledges It Will Achieve Zettascale Computing by 2027

Intel CEO Pat Gelsinger made a bold pledge in the opening keynote of Intel Innovation: the chipmaker will achieve zettascale computing by 2027. That means the company believes it will be capable of powering a system by that year that can perform one zettaflop, which is equal to one thousand exaflops or one sextillion floating point operations per second. It’s a tall order, given that the first two U.S. supercomputers capable of performing at least an exaflop are expected to go online in the next few months. But Gelsinger said Intel is working on a variety of new innovations, including new transistor technologies like RibbonFET and new lithography techniques like EUV. “Today we are predicting that we will maintain or even go faster than Moore‘s law for the next decade,” he said.