IntelON 2022: Moore’s Law is not dead, on the contrary, it is still alive and well

Intel recently held its annual innovation event IntelON. IntelON is a conference for developers, and last year's IntelON event focused on the value of developers as the center of value creation. IntelON 2022 strengthens its focus on developers and extends it to the open source community. Intel has made a number of big announcements in the data center, including new GPUs and 13th Gen Core processors.

Moore's Law still holds true

Gelsinger (Intel CEO) has been a strong supporter of Moore's Law, which states that the number of transistors will double with each generation of semiconductor advancements. Kissinger predicts that chips will have one trillion transistors in 2030, up from 100 billion today. Kissinger doubled down on his “four-year, five-node plan.”

I believe Intel's differentiated end-to-end foundry technology has incredible capabilities, and I'm glad the company didn't spin off that capability. IFS has four parts - wafer, packaging, software and open chiplet ecosystem. Gelsinger highlighted how novel packaging can give designers the tools to increase the number of transistors per device, thereby following Moore's Law.

Intel announced the expansion of the UCIe Alliance to create an open chiplet ecosystem. There are currently more than 80 companies participating in the UCIe Alliance, including Samsung and TSMC. Intel's vision for UCIe is that components from different companies in the alliance can be packaged together, reducing time to market because it will be easier to package different chiplets than to package and test a monolithic design. I'm interested to see how this impacts the software side, as software becomes less proprietary in nature as chip designers pursue the open ecosystem of chiplets offered by the alliance. I can imagine it either slowing down the software development process for the chipset in the short term, or speeding up software development as more UCIe designs are progressed.

Software is the highest cost and most time-consuming component of a chip. As chip designs become more complex, the software for the chiplets becomes equally complex. If UCIe can somehow speed up software development while improving competitiveness, I believe chip design may accelerate in the coming years.

New data centers and gaming GPUs

Intel released the Flex series of data center GPUs earlier the previous week, which is the first product to support AV1 encoding in the data center and supports Tensorflow, OpenVINO and PyTorch. As AI becomes more popular in the data center, support for these AI and deep learning frameworks becomes critical. Intel said it should offer customers a single GPU solution for a variety of vision cloud workloads.

Intel also announced the launch of Ponte Vecchio, its "highest performance data center GPU." Ponte Vecchio uses the Intel Xe HPC architecture for high-performance computing (HPC) and artificial intelligence (AI). Xe HPC is a scalable architecture with more than 100 billion transistors and Intel claims a computing power of 45 TFLOPS. The benchmark is NVIDIA's A-H100 HPC GPU

Another GPU announced by Intel is the Intel Arc A770 GPU for gamers. Many people don't believe Intel will keep its promise to release the Arc A770. Although the A770 is not the fastest consumer GPU on the market, Intel is aiming for the best price/performance ratio. Pricing is $329, which is nearly $100 less than the average selling price of the NVIDIA GeForce 3060 with similar performance. Intel also claims that the Arc A770's peak ray tracing performance is 65% higher than NVIDIA's.

Intel only recently re-entered the discrete graphics market, and so far its mobile Arc has had some success, winning a number of designs. While Arc 5 and Arc 7 aren't coming to laptops, I'm glad to see Intel releasing it on desktops.

Raptor Lake

Gelsinger also announced its 13th generation core processor, codenamed Raptor Lake, which has up to 24 performance cores, a 15% improvement in single-thread performance and a 41% improvement in multi-thread performance. Intel said it improved the P-core cache architecture and doubled the number of E-cores.

This generation is Intel's second generation of multi-configuration core products, taking into account both efficiency and performance core configurations. Intel has increased the number of performance cores from 16 to 24, which is why Intel claims a 41% improvement in multi-threaded performance. It will also have a 6.0Ghz clock speed out of the box. Gelsinger claims that the 13th generation Intel Core i9 13900K is the fastest desktop processor in the world.

The 13th generation Core processors also support DDR5-5600 and DDR5-5200 as well as DRR4, and support Thunderbolt 4 and Intel Killer WiFi 6E. I believe Intel's 13th generation Core processors may have better multi-threaded performance due to improved performance cores. If true, this appears to be a role reversal for AMD and Intel, considering AMD was once known for its multi-threaded performance, while Intel was known for its single-threaded performance. While clock speed isn't everything, 6.0GHz out of the box shows there's a lot of performance potential.

Summarize

Gelsinger continues to make Intel a steward of Moore's Law, and Intel's differentiated end-to-end foundry has an incredible future. I'm also interested in seeing how UCIe changes the market and how quickly chiplets are developed among a wider range of players.

Intel is also making huge strides with its discrete GPUs, announcing three new GPUs, two for data centers and an Arc A770, which many even doubted would launch this year. Intel also announced its 13th generation Core processors with 6.0 GHz clock speeds and better single- and multi-threaded performance.