Intel is trying hard to reclaim its appeal to developers, gamers, and the broader tech community. This message has been pushed enthusiastically by CEO Pat Gelsinger with statements like “we are bringing the geek back” at Intel. Also, part of this re-engagement with the community is the return of the Intel Developer Forum (IDF), but under a new name: Intel InnovatiON. In addition to some new chip announcements, Intel focused on new software developments and its new foundry services (which Intel refers to as “IDM 2.0”).
The focus on software led to the redesign of the Intel software developer page and a new unified software repository. These changes were driven by Intel’s new CTO and SVP/GM of the Software and Advanced Technology Group Greg Lavender, who came from VMWare, where he had been CTO under Gelsinger, prior to Gelsinger moving back to Intel.
A key part of Intel’s unifying software strategy is its OneAPI development environment. Intel is preparing to ship the oneAPI 2022 edition of its toolkits with 900 new features. The new oneAPI release adds cross-architecture software development capabilities for CPUs, GPUs, and FPGAs through a unified C++/SYCL/Fortran compiler and Data Parallel Python. Intel is betting on SYCL from the Khronos Group as the open-source heterogeneous processing abstraction layer alternative to Nvidia’s CUDA development platform.
The second day Intel Innovation keynote also focused on Intel and third-party AI software solutions because Intel customers are telling them AI processing is required on everything that computes. Much of the session focused on tools for improved AI on Intel Xeon, where the company claimed it is the most widely used server platform used from the cloud to the edge for AI. Intel’s OpenVINO 2022.1 add more models applications from edge to clouds and auto-discovery and optimizations for system accelerators, optimizing for latency or throughput.
Intel PC Processors go to 12
Probably the biggest news was the reveal of the 12th generation Intel Core architecture, code named “Alder Lake.” It’s the first client processor to be released manufactured in Intel’s recently renamed Intel 7 process. It’s also the first mainstream Intel consumer processor with heterogeneous hybrid x86 cores. Taking a page from Arm mobile processors, Intel’s hybrid architecture places performance cores and efficiency cores on the same die. A new “Thread Director” feature in Alder Lake will coordinate with Windows 11 to direct different processing threads to the optimal core. Threads with specialized processing requirements go to the performance cores, while background tasks like virus protection run on efficiency cores. Note that this rule isn’t hard and fast. Sometimes, the proper core isn’t available, so the task will be started on an available core.
There are three versions of Alder Lake: desktop, mobile, and ultramobile. The first parts introduced are the Core i9, Core i7, and select Core i5 parts for the high-performance desktop market. The desktop parts target computers where performance is king, but the heterogeneous Alder Lake designs should really shine on laptops. The top part is the Intel Core i9-12900K, with 16 total cores (8 P cores and 8 E cores) and a maximum Turbo frequency of 5.2GHz. The Core i9-12900K performance numbers shown by Intel running under Win11 are impressive, but we need to wait and see third party results with the appropriate patches for AMD Ryzen processors before we can declare who has the PC gaming CPU performance crown for the moment.
Alder Lake also introduces other system features such as 16 PCIe 5.0 lanes and dual-channel DDR5 memory. While Alder Lake does not support the full AVX512 instruction set, it does support Intel Deep Learning Boost instructions. It also has Gaussian and Neural Accelerator 3.0 for speech and audio processing. Mobile versions of Alder Lake will ship to OEM customer in November with different CPU core numbers.
Other Intel News
During the Intel Innovation event, Gelsinger announced that the Argonne National Labs Aurora supercomputer, which uses Intel CPUs and GPUs, is now expected to exceed 2 exaflops of peak double-precision compute performance. That’s double the previous performance target. Aurora uses Intel’s Ponte Vecchio datacenter GPU. In another supercomputer design, SiPearl has also chosen Intel’s Ponte Vecchio as the compute accelerator for the European Exascale Supercomputer, but SiPearl’s design uses a CPU based on Arm’s Neoverse 1 CPU instead of an Intel Xeon CPU.
While we didn’t get much information about Intel’s consumer gaming Xe GPU, branded Intel Arc (not to be confused with Synopsys Arc CPU IP cores), the company did dig deeper in one of the key display technologies for Arc — Xe Super Sampling (XeSS) — which is an open-source alternative to Nvidia’s Deep Learning super sampling technique. Intel’s Arc GPUs have an XMX AI accelerators for XeSS, but the technique should be able to run on competitors GPUs as well. XeSS, like the AMD and Nvidia alternatives, does require game developer support in the game code.
Intel also made several partner announcements. To begin with, its previously announced Mount Evans Infrastructure Processing Unit (IPU) was co-developed with Google Cloud. Taking a page out of the Nvidia playbook, Intel released workstation hardware and an Intel oneAPI AI Analytics toolkit to enable “out of the box” AI development for data scientists. This solution is now available for Linux-based workstation PCs from Dell, HP and Lenovo and for Windows 11 on Microsoft Surface Laptop Studio.
FedEx collaborated with DEKA Research & Development Corp. to develop the FedEx delivery robot called Roxo, which appeared on the event stage with the presenters. The robot is designed for autonomous last-mile delivery to a customer’s door. Roxo runs on 11th Gen Intel Core i7 processors, Intel RealSense depth cameras and uses the OpenVINO development platform for the AI inferencing engine.
Intel also had a lab session on its second generation Loihi 2 neuromorphic computing chip and the new Lava open-sourced programming framework. Loihi 2 is up and running. It’s manufactured with a pre-production version of the Intel 4 process and will offer 2x-10x faster circuits and 8x more neurons in a smaller die than the original Loihi neuromorphic chip.
The Loihi architecture is an asynchronous spiking neural net and Loihi 2 has an enhanced architecture using learnings from the original Loihi. Applications include scene and gesture recognition, adaptive robotic arm controller, odor recognition, and mathematical problem solving (train scheduling, etc). Programming these chips take a new software framework. The Lava framework offers a hardware abstraction layer and Python interfaces. It can also be used on other hardware like CPUs and FPGAs.
What we took away from the event is that Gelsinger is an energetic, enthusiastic, optimistic, and aggressive leader. He appears to be infusing that energy into the organization and challenging it to do better in the best Andy Grove style. With a new process roadmap and new chip designs and an emphasis on software to tie disparate architectures together, Intel InnovatiON put the new Intel on display. The company needed a reboot — it could no long skate on its market dominance against a resurgent AMD, a game-changing Nvidia, and a ubiquitous Arm. Intel has the technology, the talent, extensive customer base, and leader to pull off a rebound.