Last week, Intel held its annual Intel Innovation event, IntelON, where I was fortunate enough to attend in person and meet up with Pat Gelsinger, CEO of Intel, for a conversation with industry analysts. Net-net, my conversations with each of the GMs and Gelsinger’s keynote have increased my confidence in the company and its future prospects.
IntelON is a conference for developers. Last year’s IntelON event focused heavily on the value of developers as the center of value creation. IntelON 2022 reinforces this focus on developers and extends it to the open-source community. Intel has many welcomed announcements in the data center, and for gamers and content creators with new GPUs and Intel’s 13th gen processors. Without spoiling anything, let’s jump into the announcements.
What I won’t be covering yet as I am still researching it is Intel’s dive into monetized software and services.
“Moore’s Law is alive and well”
Gelsinger and Intel has always been a big proponent of Moore’s law, the idea that the number of transistors will double with every semiconductor generation. Gelsinger predicted that the industry will have a trillion transistors by the end of the decade, up from a 100 billion today. Gelsinger doubled down on its “Five nodes in four years” and after personally talking with Gelsinger, my confidence is up. Lest we forget Intel’s end-to-end foundry play with its desired acquisition of Tower Semi. And massive fab investments, the most recent announcement being the first shovels in the ground in Columbus with enough room for 20 fabs. This end-to-end foundry moves Intel into Intel’s systems foundry (IFS) era from a system on a chip (SoC) to a system in package (SiP) with what Intel claims as the broadest portfolio of differentiated technologies.
I believe Intel’s differentiated end-to-end foundry has incredible capabilities and I am glad the company didn’t spin out the capability. IFS has four components—wafers, packaging, software and an open chiplet ecosystem. Gelsinger highlighted how novel packaging is giving designers the tools to increase the number of transistors per device, stewarding Moore’s Law. All makes sense to me.
Intel announced the expansion of the Universal Chiplet Interconnect Express (UCIe) consortium for creating an open chiplet ecosystem. It has more than eighty companies participating in the UCIe consortium, including Samsung and TSMC. What Intel envisions with the UCIe are components from different companies within the consortium that can be assembled together. I believe this could be great for accelerating innovation within a chip because of the competitiveness that comes with it. I am also hoping it improves time to market as it should be easier to package disparate components than assemble and test a monolithic design. I am interested to see how this impacts the software side of things because software inherently becomes less proprietary as chip designers pursue the open chiplet ecosystem that the consortium offers. I could imagine it either slowing down the process of software development for chipsets for a short time or speeding up software development as more UCIe designs progress.
Software is one of the highest cost drivers for chips and one of the most time-consuming. As chip designs become more complex, the software for the chiplets becomes just as complex. If UCIe can somehow speed up software development while increasing competitiveness, I believe chip design could be accelerated in the coming years. I will keep dreaming.
New data center and gaming GPUs
Intel announced Data Center GPUs earlier the prior week, the Flex Series, that are the first to support AV1 encoding in the data center and support Tensorflow, OpenVINO, and PyTorch. As AI becomes more popular in the data center, support for these AI and deep learning frameworks becomes critical. Intel says it should give customers a single GPU solution for a wide range of visual cloud workloads.
Intel also announced it was shipping Ponte Vecchio, the “highest performance data center GPU”. Ponte Vecchio uses Intel’s Xe HPC architecture that targets high-performance computing (HPC) and artificial intelligence (AI). The Xe HPC is a scalable architecture with over 100 billion transistors, and Intel claims 45 TFLOPS. I do not doubt that Intel is targeting NVIDIA’s A-H100 HPC GPU and with how well Intel is marketing the Ponte Vecchio alongside its 4th Gen Xeon. Intel needs some big wins here. Like AMD, I think the challenge will be more software than hardware.
The last GPU Intel announced is the Intel Arc A770 GPU for gamers. Many did not believe Intel would follow its promise to release the Arc A770, and I believe Intel hit it out of the park with the A770. The A770 won’t be the fastest consumer GPU on the market, but Intel is targeting the best price to performance. Intel undercut the average price to performance with a price tag of $329, almost $100 less than the average selling price of a similarly performing GPU, the NVIDIA GeForce 3060. Intel also claimed that the Arc A770 has 65% better peak performance than the ray tracing competition. I believe 65% is a considerable difference, but I am curious to see how much of that is sustained.
Only recently did Intel re-enter the discrete graphics market, and so far, its mobile release of Arc has been a slight success with many designs wins. Although the Arc 5 and Arc 7 did not come to the notebook front, I am pleased to see Intel release it on desktop.
I will also take a bit of a victory lap. Three years ago, I predicted Intel’s peak consumer GPU would land in the middle of the pack versus challenging a 4090 or a new RDNA 3 card. In over 30 years, I have never seen a first-out product come out on top. It’s a really good start, though and as the software stack improve, I think Intel will improve competitively. Intel doesn’t have to make money in discrete graphic immediately, but it has a few years to prove it out.
Intel Raptor Lake
Gelsinger also announced its 13th Generation Core Processors, codenamed Raptor Lake, with up to 24 performance cores, 15% better single-threaded performance, and 41% better multi-threaded performance. Intel says it has improved the P-core cache architecture and doubled the E cores.
This generation is Intel’s second generation with an efficiency and performance core configuration. Intel has increased the number of performance cores from 16 to 24 cores. I believe that is why Intel is claiming 41% higher multi-threaded performance. It will also have a 6.0Ghz clock speed out of the box. Gelsinger claimed that the 13th Gen Intel Core i9 13900K is the world’s fastest desktop processor while also touting “the Intel platform” it runs on.
The 13th Gen Core Processors also have support for DDR5-5600 and DDR5-5200 alongside DRR4 and have support for Thunderbolt 4 and Intel Killer WiFi 6E. Intel’s Raptor Lake came out earlier than I anticipated (although Gelsinger said we all should have guessed the date), and while I believe it should give the competition a run for its money, the competition is good. I am interested to see how well Intel’s 13th Gen processors stack up to AMD’s Ryzen processors. I believe Intel’s 13th Gen Core processors could have better multi-threaded performance because of the jump in performance cores. If this were to be true, it seems to be switching of roles for AMD and Intel, considering AMD was mostly known for its multi-threaded performance and Intel its single-threaded performance. While clock speed is not everything, the 6.0GHz right out of the box reveals that a lot of the performance gains are raw performance.
Gelsinger continues to keep Intel a steward of Moore’s Law, and Intel’s differentiated end-to-end foundry has incredible, future capabilities. I am also interested to see how UCIe changes the market and how quickly chiplets are assembled across a wider swath of players.
Intel is also taking huge steps with its discrete GPUs, announcing three new GPUs, two being for the data center and the Arc A770, which many people doubted would even come this year. Intel also announced its 13 Gen Core processors with 6.0 GHz clock speed out of the box and better single and multi-threaded performance. I will have to wait until more benchmarks, and community testing comes out to give comparisons, but I believe Intel is establishing an overall fortified platform with its GPU and CPU plays. I am excited to get my hands on these products and test them out.
Net-net I think the company and Gelsinger scored wins for team blue at the event.
Note: Moor Insights & Strategy co-op Jacob Freyman contributed to this article.