Four key takeaways from Intel Investor webinar

Four key takeaways from Intel Investor webinar

Intel focused on the company’s Data Center and Artificial Intelligence (DCAI) division, announcing a new roadmap and a series of new developments.

All eyes lately—especially those of investors—have been on Intel’s data center business.

Recently, Intel's Sandra Rivera, Greg Lavender and Lisa Spelman hosted a webinar focusing on the company's data center and artificial intelligence business units, detailing Intel's latest market forecasts, hardware plans and how to empower developers through software.

The webinar provided an update on Intel’s data center business. It included information on future generations of Intel® Xeon® chips, the latest progress on the 4th Generation Intel® Xeon® Scalable processors (codenamed Sapphire Rapids), and demonstrations of Intel hardware comparing to competitive products and running heavy AI workloads.

1. Xeon Roadmap Product List

Intel's server CPU business has made a number of advances in the Sapphire Rapids, Emerald Rapids, Sierra Forest and Granite Rapids processor products. Here is the latest Xeon roadmap, arranged in order of release:

Currently: The latest on 4th Generation Intel Xeon Scalable Processors (Sapphire Rapids)

More than 50 mainstream OEMs and ODMs are shipping systems based on the 4th Gen Intel Xeon Scalable processors, and the top ten cloud service providers are also deploying services based on the 4th Gen Xeon. As Intel's highest quality data center CPU to date, Sapphire Rapids is shipping rapidly, and has received 450+ processor product designs, setting a new record for the Intel Xeon series, and more than 200 designs have been shipped.

Q4 2023: 5th Generation Intel Xeon Scalable Processors (codenamed Emerald Rapids)

Emerald Rapids is Intel's next performance core (P-core) product. Starting today, it is officially called the 5th Generation Intel® Xeon® Scalable Processor. In preparation for the official release, it has switched from a code name to a formal brand name, is currently providing samples to customers, and is scheduled to be delivered in the fourth quarter of 2023, and mass production verification is underway.

Sandra Rivera, Intel's executive vice president and general manager of the Data Center and Artificial Intelligence Group, told investors that the product's chips are of very high quality and optimized for performance. It will have extremely high processor core performance, achieve higher performance per watt in the same power envelope, and be optimized for specific workloads through built-in accelerators.

(5th Generation Intel Xeon Scalable Processor codenamed “Emerald Rapids”)

First half of 2024: Intel Xeon processor (codenamed Sierra Forest)

Sierra Forest is the first Energy Efficient Core (E-core) processor, scheduled to be available in the first half of 2024. Today, Intel announced that Sierra Forest will have 144 cores per socket. Sierra Forest is the first CPU to use the upcoming Intel 3 process technology, and after powering up earlier this quarter, multiple operating systems were able to boot in record time (less than a day). It is progressing as planned, and the first samples have been delivered. In the Sierra Forest demonstration, the scene of 144 cores busy processing the demonstration workload through Task Manager was impressive. Designed for cloud-optimized workloads, Sierra Forest will be optimized for energy efficiency through optimized performance per watt, high core density, and high throughput performance.

(Intel Xeon processor codenamed "Sierra Forest")

Fast-forward: Intel Xeon processor (codenamed Granite Rapids)

Granite Rapids will be released in 2024, following Sierra Forest. Although the specific release date has not been disclosed, sharing a platform with Sierra Forest will facilitate its quick time to market. Shared IP and technology minimize development and design time. Granite Rapids is achieving all major milestones and has made smooth factory production progress. Its samples are also being provided to customers with good feedback. Spelman showed the innovations of Granite Rapids for the first time in a demonstration, including higher core density, memory and I/O innovations.

“We are building the world’s fastest memory interface for Granite Rapids,” Spelman said as he fired up a command prompt to show off its memory configuration. “Intel invented and led the ecosystem in developing a new type of DIMM called Multi-Way Combined Array (MCR) that enables us to achieve 8,800 teraflops on top of DDR5.”

MCR DIMM innovations increase peak bandwidth by 80 percent compared to current generation server memory technology. Lisa's demonstration showed that Granite Rapids is very stable when a healthy memory subsystem is fully loaded with read/write operations.

future

Intel has publicly stated for the first time that it will continue its energy-efficient core roadmap by developing a follow-up product, Clearwater Forest, after Sierra Forest. Clearwater Forest will be available in 2025 and will be manufactured using Intel's 18A process technology. Intel plans to achieve process technology leadership at this node - this will also be the culmination of the company's strategy to advance five process nodes within four years.

Additional Information

• The Habana® Gaudi® 3 AI accelerator has completed design certification (taped in). Although performance details have not yet been announced, the Habana® Gaudi® 2 is already in use and has excellent throughput per watt when running mainstream computer vision workloads.

• This year, 15 new FPGA products are planned to go through the Production Release Qualification (PRQ) process, which is Intel's largest FPGA new product launch to date.

2. The CPU market size exceeds $110 billion

The compact roadmap targets a larger market than expected. Rivera believes that in five years, the data center chip market will reach $110 billion, about twice the market size expected at last year's Intel Investor Conference. The reason is that the changing data center market landscape has expanded the current demand for CPUs alone.

“When we talk about computing demand, we often measure the market size by the number of CPUs shipped,” Rivera explained. “However, the number of sockets does not fully reflect the value that chip innovation brings to the market. Today, innovation involves multiple dimensions, including increasing the density of CPU cores, using built-in accelerators in silicon, and using standalone accelerators.”

Based on the above situation, Rivera listed several specific reasons for the increase in market size: First, as the data center business integrates accelerator computing and leading GPU products, Intel can better serve a wider customer base. In addition, the excellent performance of Xeon Scalable Processors in running large workloads such as AI, data analysis, security, networking and HPC has driven the demand for mainstream computing and independent accelerators "tailored" for workloads.

3. Intel-based AI is everywhere and available to everyone

In addition to the latest on chips, AI was also a major theme of the webinar.

Intel technology is the cornerstone of AI hardware, from data preparation and management on Xeon processors to small and medium-sized training and reasoning. At the same time, more and more large model training and reasoning are using Intel GPUs and accelerators.

As the market demand for general computing (mainstream is CPU) and accelerated computing (currently mainstream is GPU and dedicated accelerators) continues to grow, by 2027, the market size of logic chips will exceed US$40 billion, and there will be great potential in the field of AI chips/accelerators.

Intel is committed to driving true AI adoption, from the cloud to the network to the edge. Through an open ecosystem, Intel is making solutions more accessible and enabling more economical AI deployment. Today, Rivera demonstrated Intel’s commitment.

“Customers want their AI workloads to be portable. They want to build once and deploy AI everywhere,” Rivera said. “As we continue to deliver heterogeneous architectures for AI workloads, their deployment at scale will require developer-friendly software and a vibrant, open, and secure ecosystem.”

Greg Lavender, Intel’s chief technology officer and head of the Software and Technology Group, spoke on the spot about Intel’s advanced software and its investment in a holistic, end-to-end, system-level approach to AI software development, including standardization of programming languages ​​to achieve the “portability” Rivera mentioned.

“We believe the industry would benefit from standardized programming languages ​​that everyone can contribute to and collaborate on, rather than being tied to a specific vendor, while also allowing for organic growth based on member and common needs,” Lavender said.

Intel has made many contributions to SYCL, an open programming model based on C++, and acquired Codeplay Software, the leader of the SYCL language and community. SYCL is included in oneAPI, so customers can program and compile across CPUs, GPUs, and accelerators from multiple vendors. Other work on upstream software optimization includes optimizations for PyTorch 2.0 and TensorFlow 2.9, as well as a collaboration with Hugging Face, all of which can be trained, tuned, and predicted with the help of Intel Xeon processors and Gaudi 2.

4. Helping the development of generative AI

Intel’s open philosophy extends to the field of generative AI, where powerful tools such as ChatGPT and DALL•E 2, a model for generating images from text, are leveraging AI algorithms and large data sets to generate content from complex artificial prompts.

Generative AI is advancing at a rapid pace, and Intel is actively supporting the open generative AI ecosystem and meeting the growing performance demands.