Counting with fingers,IntelIt's time to squeeze the toothpaste again.
On September 19, the 2023 Intel ON Technology Innovation Conference opened in San Jose, California, USA. The most obvious feature is that the entire conference revolves around AI.Pat KissingerIn other words, Intel wants to make AI ubiquitous.
Intel actually knows what everyone wants to see, so this time they announced a series of new technologies and painted a lot of "pie", but almost all of them are related to AI.
AIchipRoadmap released for the first time
Now everyone is rolling up big models, Intel certainly does not want to fall behind, but also does not want to blindly follow the trend. So Kissinger put forward the concept of "AI PC", intending to be the leader and open up a new AI track.
To realize the vision of AI PC, it is undoubtedly necessary to rely on powerful chip computing power. Intel also disclosed their AI chip roadmap for the first time. Kissinger believes that the roadmap is very important for Intel at present. Delivering products according to the roadmap time can give customers and developers great confidence and realize AI strategy.
The roadmap mainly shows the development plan for three generations of AI chips, including the Gaudi 2 with 7nm process technology that has already been launched, the Gaudi 3 with 5nm process that will be launched next year, and the next generation of AI chip code-named Falcon Shores.
It is understood that the computing power of Gaudi 3 will be twice that of Gaudi 2, and the network bandwidth and HBM capacity will be 1.5 times that of Gaudi 2.
At the same time, Kissinger did not forget to brag about the performance of Gaudi 2. He said that the already launched Gaudi 2 performed well in the latest MLPerf AI reasoning performance test results, and the reasoning performance of some models exceeded the industry benchmark NVIDIA A100.
In addition, Kissinger also announced that a large AI supercomputer will be built entirely with Intel Xeon processors and 4,000 Intel Gaudi2 accelerators, with Stability AI as its main customer. This supercomputer is expected to enter the top 15 supercomputers in the world.
Four years and five milestones
In order to proceed smoothly according to the roadmap, the process technology must be mentioned.TSMCAfter starting mass production of chips using the 3nm process technology, Samsung also announced that it would start improving the yield and mass production of the 3nm process. Intel naturally refused to be outdone. Kissinger said that Intel's "four-year five-process node" plan is progressing smoothly. Intel 7 has achieved large-scale mass production, Intel 4 is ready for production, and Intel 3 is also progressing as planned, with the goal of being by the end of 2023.
Everyone saysMoore's LawIntel doesn’t think Moore’s Law is dead. They still want to continue to promote it, but they want to take a different path.
Another way Intel is advancing Moore's Law is by using new materials and new packaging technologies, such as glass substrates. This is a breakthrough that Intel just announced. Glass substrates will be introduced in the late 2020s and continue to increase the number of chips in a single package.transistorThe number of CPUs will help meet the needs of data-intensive, high-performance workloads such as AI and continue to advance Moore's Law after 2030.
New generation of Xeon and mobile processors
As mentioned earlier, Kissinger immediately brought up the concept of "AI PC", which is inseparable from Intel's signature product - CPU. This time, Intel released a new Meteor Lake architecture processor built with Intel 4 process technology, which they themselves called the revolutionary architecture change of Intel client SoC in 40 years!
However, the new Meteor Lake architecture is only used for mobile processors, that is, laptop CPUs, and desktops are not available for the time being. The Meteor Lake architecture will be first used in Intel's upcoming new generationCoreUltra mobile processor.
Intel is also very confident about the new Core Ultra mobile processor. They call it: "a turning point in the PC processor roadmap" - the highlight of this processor is that Intel has installed a neural network processor (NPU) on it, allowing traditional PCs to achieve a stronger AI acceleration experience locally.
In addition, Intel also previewed the next generation of Intel Xeon processors and revealed that the fifth generation Intel® Xeon® processor will be released on December 14, when the next generation Core Ultra will also be released.
Fully embrace AI
Since AI is used throughout, a large model is indispensable. At the Intel ON Technology Innovation Conference, Intel demonstrated the use of generative AI to generate songs. When disconnected from the Internet, the large language model is completely run locally on the PC driven by Intel OpenVINO, and real-time questions and answers are conducted with the AI chatbot.
In addition, the OpenVINO toolkit has been updated, and the Intel Developer Cloud Platform has been announced to be fully online. Intel also stated that the Strata project and the development of the edge native software platform will be launched in 2024, and will provide modular components, high-quality services and product support.
It can be seen that Intel is paving the way for AI, from processors to development kits. Whether it can "extend" Moore's Law and stand out in the AI competition depends on whether it has the opportunity and ability.