Intel Innovation Event And Meteor Lake Chip Signal A Turnaround Has Begun
From its focus on state-side manufacturing to new cutting-edge CPU designs, software enablement and AI, Intel appears to be on the path to recovery sooner rather than later.
It’s no secret that a few years back Intel lost its edge on chip design and manufacturing, especially the latter, under previous leadership. Even Intel’s current CEO, Pat Gelsinger has gone on record acknowledging this in more recent years. However, at its annual Innovation event in San Jose last week, Intel demonstrated to me that not only are its new IDM 2.0 manufacturing and chip foundry plans beginning to take shape for meaningful impact, but its relentless pursuit of returning to PC chip design dominance is also starting to bear fruit. In addition, it’s now clear to me that the culmination of this two-pronged approach is showing signs of a true turnaround in company dynamics and perhaps execution as well, the likes of which we probably haven’t seen in several years. Gelsinger has a favorite slogan that “the geek is back” at Intel and, after last week’s meetings and presentations, indeed I would have to agree. As such, I’ll step through a few key takeaways from last week’s Intel Innovation 2023 event, so you can get a better flavor for why I left San Jose with this renewed level of optimism for the company.
AI Everywhere All At Once But With A Purpose, Not Just Party Tricks
It’s obvious that AI is a hot topic and could very likely be the most impactful new field of technology in decades. That said, when it comes to client computing on desktops and laptops, I had tough questions going into the conference on what real-world end user applications and experiences that AI and machine learning could bring to the modern PC. Fortunately, Intel came locked and loaded with a ton of demos that went well beyond what we’ve seen so far for AI-enhancements on the PC, which have typically be relegated to disjointed generative AI demos and simple webcam scene manipulation.
One demo that sort of got my hackles up was when an executive showed generative AI on stage and asked it to write and email to a colleague discussing a certain topic and ideas for further discussion. It was almost like the sender was saying, “I don’t have time to write you a proper email (or you’re not important enough), so my AI is reaching out.” I don’t think that was the intent of the demo obviously, and the AI admittedly did the job well, but taking the human out of interpersonal communication was the first thing that crossed my mind. So just be sure to check what your AI is writing for you in the future, and add the you back in, if need be. On the contrary, however, an AI-powered app called Rewind was very impressive as a possible key, enabling tool of the future.
The founder of this company has suffered from hearing loss over the years and thus the app’s technology struck a chord with Intel’s CEO Gelsinger, who has also suffered from this type of sensory loss as well. In short, the small team at Rewind developed a listening and watching AI of sorts that you can enable on your PC, and it will run in the background capturing anything you’ve seen on your display, spoke into your mic, heard in the room or on the meeting calls you’ve been on. You can then query the AI to summarize your meetings or research, after taking copious notes for you obviously, and much more. In fact, it summarized Gelsinger’s keynote on stage, at that point in time, quite well. The technology is also fully encrypted and stores its data on your machine locally, rather than going out to the cloud with your personal or professional data. There’s still a bit of a privacy concern here for me, but you can also disable and re-enable the app with a couple of clicks. You can think of this app as Windows Co-Pilot on steroids, and it’s actually planned to be running on Intel Meteor Lake Core Ultra laptop systems when they ship later this year.
Intel executives and Gelsinger also demoed a number of apps and use cases on stage, from hearing aid or earbud integration into the PC, such that it would auto-detect when you needed to be in noise cancelling or pass-through mode, to sports and retail clothing assistance apps, AI assistance in Adobe Suite apps and more. It’s safe to say that Intel laid the foundation and a vision for what you will be able to do with AI on the PC, such that attendees at its Innovation event could see the benefits of AI-PC and data center integration with a purpose, not just gimmicky tech demos.
Software Is Key To Enabling All Hardware, Especially When It Comes To AI
Another level of detail that I was hoping we’d get a lot more insight from Intel on, was software. You might be surprised to know that software engineers often out-number hardware engineers in a chip company like Intel by a factor or 2 or 3 to 1. The reason for this is simple — without optimized tools and software for these chips to run on and developers to code with, adoption will undoubtedly become a challenge.
Intel was quick to point out the deep resources the company is bringing to bear in this new AI-infused reality, from its well-known OpenVino platform, to new initiatives like Project Strata for edge AI and the IoT. You can think of OpenVino as a light-weight wrapper that allows developers to work in their native frameworks for machine learning. Gelsinger noted that “the newest release, OpenVINO 2023.1, “brings us closer to the vision of any model, on any hardware, anywhere.” Incidentally, OpenVino support now includes Arm as well, and the company has been very vocal about supporting RISC-V technologies for Intel Foundry Services. The company was also quick to point out support of other mainstream AI standards like DirectML, WinML and ONNX RT.
Meteor Lake Core Ultra Is The Most Innovative Intel CPU Design Approach In Many Years
Beyond the company’s focus on delivering AI at scale, and the software to support it, Intel also unveiled details of its Meteor Lake-based Core Ultra mobile processor architecture. Core Ultra is both a bold bet and a very innovative design approach that is likely to pay off in spades in both performance and efficiency. The key to this highly chipletized (I’m inventing a word here) architecture is what’s known as a disaggregated topology. That is to say resources on the chip, from its CPU cores, GPU and media engines, as well as its new NPU (Neural Processing Unit) for AI acceleration, all reside on discrete tiles with Intel Foveros technology stitching them together over a high speed IO and network-on-chip fabric.
Meteor Lake will be built partially on Intel 4 fab process technology, with new P-Core and E-Core CPU architectures, Intel’s latest Arc graphics core with even a disaggregated media engine, again for power efficiency. All in, Intel is claiming a big lift in performance-per-watt metrics in Meteor Lake and a 2X lift in graphics performance and efficiency.
My colleagues Marco, Chris and I had a great interview with Dan Rogers and Robert Hallock of Intel, with a deep-dive on Meteor Lake, which I of course highly recommend you check out...
IDM 2.0 - Huge Risk, Equally Huge Upside, Great Timing And The Wrap-Up
With Intel’s 10s of billions in investment in state-side chip manufacturing for internal consumption and its Foundry Services division, Intel CEO Pat Gelsinger has put a lot on the line for the company, in hopes of a big payoff and a return to the manufacturing giant status that was once affectionately referred to as “Chipzilla” in the industry. With billions in investments in Chandler Arizona and Ohio as well, these new facilities will come on-line starting in 2025. Again, it’s a big bet, but if you stop to think about it, if the company were to chip off even 10% of the demand that is now oversubscribing TSMC in some cases, it could mean big business and several billion in new revenue for the company annually.
If you couple this opportunity with the fact that the US government is striving hard to bring chip manufacturing back state-side with the CHIPS act, government contracts being awarded, etc. and you start to see the moon and stars aligning a bit for Intel here. The fact is much of the semiconductor industry is too reliant on TSMC, and Intel is poised to take advantage of that dynamic.
The final piece of the puzzle I think the company needed was the new culture of execution that Gelsinger brings to the mix. In a meeting with Jim Johnson, Corporate Vice President General Manager, Client Engineering at Intel, he spoke of the build versus outsource decisions that come to every design meeting now. In other words, if Intel currently can’t build it itself as cost-efficiently or performant as an outside vendor, design teams are tasked with doing what’s best for the company, regardless. This healthy competition for internal demand, should help keep the company focused on the leading-edge processes it needs as well. As a side bar, in his keynote, CEO Gelsinger held up an Intel 18A (18 angstrom) wafer declaring it “the Picasso” of wafers and noting that its next, next gen (after Arrow Lake) Panther Lake processor architecture will be built on it, and in fabrication starting Q1 2024.
Wrapping things up, I’d say Gelsinger and the team at Intel are showing clear signs of being on the right track again. There was a day when fabless semiconductor companies had a clean shot at out-innovating Intel, but no chance of out-manufacturing it. With Gelsinger truly “bringing the geek back,” in chip design, software and manufacturing, Intel seems well-targeted to reclaiming at least some of that dominance. The key is going to be in the execution, and that is always easier said than done. However, these early indications signal that the execution is coming back now too, along with the geek.
