Intel Ships Its First Xeon 6 Chips for Data Centers

Intel has begun shipping the first of its next-generation server processors: a 144-core Intel Xeon 6 processor with Efficient cores (E-cores) that is designed for public and private clouds in situations where power efficiency and performance are critical, the company announced today (June 3).

The Intel Xeon 6700E chip, previously code-named Sierra Forest, delivers 4.2 times improved performance and 2.6 times performance-per-watt improvements compared to data centers running on 2nd generation Intel Xeon processors. The efficiency gains will enable cloud providers and other data center operators to consolidate racks by a 3-to-1 ratio, Intel executives said.

For the past year, Intel has talked about its two-product strategy within its Xeon 6 processor family: Performance core (P-core) chips designed for all-out performance for artificial intelligence and high-performance computing workloads; and E-core chips that balance energy efficiency with performance and are ideal for cloud-native applications, content delivery networks, microservices and consumer digital services.  

Intel Xeon 6 Details

At the Computex conference in Taiwan today, Intel shared more information on the Intel Xeon 6 family and Gaudi 3 AI accelerators, including architecture details, performance metrics, and product launch dates between now and the first quarter of 2025. This includes the launch of the 288-core Intel Xeon 6 E-core chip, which is expected early next year.

Analysts say sharing detailed Xeon 6 processor launch dates is significant because it shows the company is confident that they are on track to deliver the products on time. That’s a far cry from recent years when Intel suffered notable delays of 3rd and 4th-generation Intel Xeon chipsets.

Today’s launch of the Intel Xeon 6 processor family is important because it demonstrates that Intel is listening to the market and diversifying to suit the expanding requirements of its customers, said Ian Cutress, chief analyst at More than Moore.

“Xeon 6 could simply be perceived as a next-generation update, but Intel is diversifying its portfolio and aligning around two different core families,” which addresses different customers’ needs, Cutress said. “P-cores have benefits in the traditional enterprise that needs high performance. And the E-cores have benefits for cloud providers who want to sell instances per vCPU.”

Intel Xeon 6 will allow Intel to better compete against AMD’s CPU offerings, said Matt Kimball, vice president and principal analyst at Moor Insights & Strategy. Intel has long dominated the x86 server chip market, but AMD has grabbed market share in recent years.

Intel Xeon 6’s increased core counts, 3nm process node and architecture – including Intel’s built-in acceleration engines – puts the company in a strong position to compete, Kimball said. Intel’s AMX acceleration engine, for example, speeds AI inferencing because the workload is moved off the CPU.

“When you look at the performance perspective, AMD has been so far ahead of Intel for so long. But with Intel Xeon 6, Intel is back in the competitive game,” he said.

However, AMD is not standing still. AMD’s forthcoming 5th Generation AMD EPYC family of processors, code-named Turin, is expected to be available during the second half of 2024 and feature up to 192 cores and 384 threads, an AMD spokesperson said.

Intel Xeon 6 Roadmap and Launch Dates

Intel plans to launch its first P-core chip, formerly codenamed Granite Rapids, and the Gaudi 3 AI accelerator, in the third quarter of 2024.

The P-core chips will range between 86 to 128 performance cores. The 128-core 6900P P-core chip, available in the third quarter, doubles the performance of AI inferencing and general computing functions and improves HPC performance by 2.3 times when compared with previous 5th generation Intel Xeon processors.

The 6900P will “be the most powerful Xeon that we deliver,” said Justin Hotard, executive vice president and general manager of Intel’s Datacenter and AI Group, in a media briefing.

Intel Xeon 6 P-core processors will perform AI inferencing 3.7 times better than AMD EPYC processors, while Xeon 6 E-core processors will provide 1.3 times better performance per watt over AMD EPYC chips on media transcoding workloads, said Matt Langman, general manager and vice president of Intel Xeon 6 processors, during the media briefing.

Intel plans to round out its Xeon 6 offerings in the 2025 first quarter, including the launch of its 288-core 6900E E-core processor and processors designed for network and edge applications, he said.

As for the newly available Intel Xeon 6 6700E, general compute, database, and web workloads are three times faster, while performance per watt is more than two times better than 2nd generation Intel Xeon processors.

Cutress said Intel is comparing the new 6700E chip to 2nd generation Intel Xeon chips so data center operators who are deploying servers from about five years ago can see the performance gains if they upgrade.

Intel Xeon 6 processor (Source: Intel)

Besides performance-per-watt improvements, upgrading to 6700E from 2nd generation Xeon chips will enable a data center with 200 racks to consolidate to 66 racks, said Intel’s Langman. That data center can save up to 84,000 megawatt hours of energy and 34,000 metric tons of carbon emissions.

As a result, the rack space and energy savings from E-core chips will allow data center operators to reach their sustainability goals, but it also frees updata center resources for those that want to deploy new hardware for AI workloads, he said.

Intel 6 Chip Architecture

During a media briefing, Intel executives touted the Intel 6 architecture, which comes in three parts. Xeon 6 is built on a new modular-die fabric that serves like the foundation and plumbing of a house. It provides better and faster traffic flow, said Kimball the analyst.

Sitting on top of the modular-die fabric is the Xeon multi-die architecture. There are two dies, also known as chiplets, that are paired together: The compute die with cores, cache and memory controllers, and the I/O die with UPI, PCIe, CXL, which makes memory faster and more accessible, and Intel Accelerator Engines, Kimball said.  The third piece of the architecture is an embedded multi-die interconnect bridge.

The architecture is meant to provide the flexibility to mix and match the dies/chiplets to boost the performance or performance per watt of the Xeon 6 processors.

For example, the 6900 series processors could have three compute chiplets on the P-core version and two compute chiplets on the E-core version, while both have two I/O chiplets, said Intel senior fellow Sailesh Kottapalli in the media briefing.

“Xeon 6 was designed from the ground up for flexibility by adding the right mix of cores, accelerators, I/O and memory to support all of these workloads,” Kimball said. “Think of Lego blocks that can be put together in different configurations depending on the needs of the workload.”

AMD has long designed its processors using a chiplet design, so Intel has played catch up here, Cutress said. “It’s a lot more cost-effective, and it’s a lot easier to design.”

Gaudi AI Accelerator Prices are Cheaper than GPUs

Intel also announced pricing for its Gaudi AI accelerators. A standard AI kit including eight Intel Gaudi 2  AI accelerators with a universal baseboard costs $65,000, which is one-third the cost of competitive AI accelerators, the company said.

A kit featuring eight Intel Gaudi 3 AI accelerators with a universal baseboard costs $125,000, two-thirds the price of the competition, Intel said.