Shulou(Shulou.com)11/24 Report--

On the eve of Intel on Technology Innovation Conference 2023, Pacific time, I had the opportunity to interview Intel CEO Pat Kissinger to learn about Intel's progress over the past year and to gain a full understanding of Intel's development direction in many cutting-edge areas such as client computing, data center, edge computing and cloud computing. The premise for talking about these topics is that Intel not only focuses on accelerating the AI ecosystem through hardware, but also enables developers to deploy AI applications on millions of client devices on a large scale through the Intel open source AI development kit in the OpenVINO toolkit.

Execution, execution, or execution?

Kissinger told me from the beginning that he was pleased that the company had regained the innovative spirit of the Intel Information Technology Summit (Intel Developer Forum, IDF). Through Intel on Technology Innovation Conference (Intel Innovation) and Intel on Industry Innovation Conference (Intel Vision), Intel brings the entire ecology together in a practical way to promote creative acceleration and growth. This is also one of the extensive efforts made within Intel to improve its execution. Before Kissinger took office, Intel had encountered obstacles in terms of execution. This year is the third year of Intel on Technology Innovation Conference, and Intel is renewing systematic communication with developers. Kissinger hopes Intel's annual conference for developers will go a step further than the previous year. Intel has received valuable feedback from these activities, which will eventually help drive the company's future work and meet the needs of the industry.

Kissinger said that the company will continue to promote "five process nodes in four years" as planned, and the Intel 18A process nodes will be delivered as planned. Intel's long-standing strategy is to rely on process leadership to provide leading performance and power consumption, so process nodes have become the cornerstone of Intel's revitalization strategy. Intel was two years ahead of TSMC, but that advantage has since disappeared. As a company, Intel seemed to have lost its way in the past, like a ship without a rudder. Almost all designs need to return to the older process, which means that the value proposition and timing of the product have changed greatly. Therefore, to build a competitive process node and implement it according to plan is very important for the company to turn the tide. This is the most important indicator that I am concerned about. Kissinger announced at Deutsche Bank's 2023 technology conference that Intel is not only speeding up the construction of a contract factory in Arizona, but also that some customers have already paid a deposit on Intel 18A production capacity, which is an important vote of confidence in Intel's manufacturing process. Intel has also partnered with industry-renowned companies such as Synopsys to expand partnerships based on Intel's advanced process nodes and promote a leading IP portfolio based on Intel 3 and Intel 18A. With Synopsys's leading position in the industry, this is a key partnership deepening for any contract manufacturer, especially a fast-growing one like Intel. Judging from the recent news, I am more confident in Intel's ability to deliver on its promises and strengthen its execution, but I probably wouldn't have said that a year ago.

As Kissinger said, the key to Intel's success is to find world-class execution, not only to meet expectations on time, but also to exceed expectations.

Data Center and AI

In view of the popularity of the generative AI market at the end of last year, many people focus on the generative AI training solution based on GPU. The reality is that while this is the current hot spot, it is unlikely to be the only beneficiary of the recent boom. As we have seen in machine learning and deep learning, the initial surge began with AI training, but it also led to an outbreak of demand for AI reasoning. Amazon has made it clear that 80 per cent of Alexa's machine learning budget is spent on reasoning.

I would like to reiterate my previous prediction that Intel is in a strong position in the future generative AI field. Although we haven't seen much of what we can call "critical evidence", there is evidence that Intel can offer highly competitive and viable alternatives to Nvidia and AMD products in AI reasoning. For the latest MLPerf testing, Intel recently submitted test results for three products-- the Gaudi2, the fourth-generation Xeon processors, and the Xeon CPU Max series. I have always disagreed with some manufacturers and competitors in the industry that do not submit product results to MLPerf for testing, but Intel did.

MLPerf test results show that Gaudi2 has a cost-effective advantage on the same workload. For specific workloads, the fourth-generation Xeon CPU is well suited to use popular open source AI frameworks and libraries to build and deploy generic AI workloads. What few people want to talk about is that most AI workloads currently run on CPU around the world. Intel remains the only server CPU provider to submit MLPerf test results. This is the first time that Intel has submitted the MLPerf test results of the Intel Xeon CPU Max series (high-bandwidth memory up to 64GB). In terms of GPT-J, it is the only CPU that can achieve 99.9% accuracy.

These results show that Intel is very competitive in the AI reasoning performance of specific workloads. I believe this also reinforces Intel's commitment to fully deploy AI in hardware and software. But we have to start somewhere, and Intel has a lot to do from winning "specific workloads" to winning "most workloads". I also look forward to learning more about the progress at Intel on Technology Innovation Conference.

AI accelerated PC

Nowadays, everyone is talking about AI, and Intel is no exception. Intel is working to implement AI in everything from cloud computing to PC. Intel has integrated AI into every platform, from Meteor Lake on the PC side to Sapphire Rapids on the server / data center side. I expect different forms of AI applications to emerge at the edge of the data center (such as operators) and vertical industries (such as retail and manufacturing). Intel has been preparing for this moment for many years, and with the launch of Meteor Lake, Intel will have a significant impact on how businesses and consumers use AI applications and their coverage. There is still a lot of work to be done at the software level to implement these functions, so I think Intel must build AI functionality into it a few years before the hardware is put into use.

Before starting to take advantage of the hardware and build AI applications, developers need to know the functionality and performance of AI. Generative AI makes developers' demand for AI performance grow rapidly. We gradually see that everyone in the industry is talking about how to use flexible models with high performance and high energy efficiency to implement generative AI applications on devices. You still use the cloud for model training and retraining, but it is well known that cloud reasoning is too expensive to scale, especially for applications such as generative AI.

In my conversation with Kissinger, I sensed that he believed that AI PC was a turning point in history similar to Wi-Fi. Intel's Centrino processor may not be the starting point for Wi-Fi, but it does lead to new usage scenarios and new applications, because developers can assume that the device will have networking capabilities. He even mentioned that when Centrino went public, the 802.11 IEEE standard on which Wi-Fi was based was the fourth iteration of 802.11g.

Intel said that hundreds of independent software providers will support Meteor Lake, which means that more developers will soon announce support for Meteor Lake. Intel's OpenVINO is at the core of Intel's open source software features. Intel says it is gaining widespread recognition of edge computing across the industry. We haven't seen this on the PC side yet, but I believe that Meteor Lake may be the beginning. We may see OpenVINO make Intel a big success on the PC side next year. I'm seeing a lot of interesting developments in the industry right now, which makes me think next year will be the year of AI PC.

Summary

I do believe that Intel is surpassing the current artificial intelligence bubble and not just considering high-end cloud applications. Intel is the only company that can make AI ubiquitous in tens of millions of client, cloud connectivity, and edge smart use cases. When Intel goes all out to implement it, it has always been the creator of the market. Intel's ability to lead and shape the industry is the foundation of its success, and coupled with its execution and leadership in manufacturing, it is the secret of success. Packaging technologies such as Foveros are driving Intel to launch some interesting new products (such as Meteor Lake), while the adoption of core architecture has put Intel back on the path of market makers. Foveros will also benefit other chipmakers because it has an advantage over TSMC's CoWoS and seems to be more available.

I look forward to the product and technology developments announced at Intel on Technology Innovation Conference in 2023.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.