Intel announced Thursday’s earnings announcement that it would need to re-spin its fourth-generation Xeon scalable “Sapphire Rapids” processor before mass production begins. This redesign and tapeout took some time. Therefore, Intel has postponed the volume lamp of SPR CPU to the latter half of 2022. Meanwhile, Intel’s Xeon “Granite Rapids” processor computing tiles scheduled for 2024 will be powered in the second half of the third quarter. .. Intel has also signed a contract with AWS to develop a custom data center solution.
Not so great for sapphire rapids
After all, previous iterations of Intel’s next-generation Xeon scalable “Sapphire Rapids” processor had security vulnerabilities that required hardware mitigation.
As a result, Intel had to re-spin the CPU, fix the problem, tape the new steppings, and then continue with the regular testing and verification steps. This, of course, postponed the mass launch of the product. Some Intel customers (probably hyperscale data center operators) are already running early Sapphire Rapids processors (perhaps because certain features are turned off) without being affected by security issues. increase.
“We did not ship at the required quality and security level. […] Obviously, the product shouldn’t have had such a bug in the first place. ” […] So we did something else [stepping], This talks about bulk SKUs, but they will be shipped in bulk later this year. ”
Intel has hundreds of customers with different requirements for Xeon processors. For example, operators in hyperscale cloud data centers typically require some specific features and high performance, but may not need all the features that Xeon offers. As a result, Intel can ship new CPUs without any negative impact before they are generally available. But for a typical customer like a traditional server maker, Intel needs to offer a CPU model that will satisfy many types of customers with different workloads and software. Therefore, chip bugs may not affect giants such as AWS, Meta, and Microsoft Azure. However, it can affect companies with different needs and software stacks. Therefore, we recommend that you fix any hardware issues before shipping.
However, new stepping (and the resulting retest / revalidation) automatically delays a large number of boots. As a result, Intel plans to make the 4th generation Xeon scalable “Sapphire Rapids” processor generally available only later this year. Therefore, the production lamps for these CPUs will start in 2022, but will continue to work well until 2023.
“Sapphire Rapids [is] “There are some SKUs out there. That’s good, but there aren’t any major SKUs out there, so they’ll happen later this year,” Gelsinger said. “Of course, they’re out this year. It will contribute much more next year than it will. ”
Intel has acknowledged that the Sapphire Rapids processor is late for the party and will only be widely available in 2023. It’s unclear how this will affect the availability of Intel’s 5th generation “Emerald Rapids” CPUs, which are socket compatible with Sapphire Rapids. (And created using the same Intel 7, also known as 10nm Enhanced SuperFin process technology, as SPR), scheduled for its debut in 2023. However, at this point, Intel has no plans to delay Emerald Rapids, claiming that the CPU certainly looks “normal.”
“Emerald [Rapids] Enter sapphire [Rapids] Intel CEO said: [Rapids] And Sierra Forest is a product of 2024. “
Granite Rapids powering up this quarter
However, while Intel’s 4th generation Xeon scalable “Sapphire Rapids” processor has problems, the development of one of its successors, codenamed Granite Rapids CPU, seems to be on track.
“We made the tape in the first step of the Granite Rapids CPU and expect it to turn on this quarter,” said an Intel director. “Later this year, we plan to tape a number of in-house and foundry customer test chips to various process nodes such as Intel 3 and Intel 18A.”
Intel’s codename Granite Rapids is an important product for Intel. The CPU should arrive in 2024. It uses a new platform that takes over the next socket LGA4677-based platform that supports Intel’s 4th generation Xeon scalable “Sapphire Rapids” and 5th generation Xeon scalable “Emerald Rapids” processors. You can name the 6th generation Xeon scalable processor while improving performance and bringing additional features. This will make Intel more competitive in the server market.
Granite Rapids compute tiles are created using Intel 3 manufacturing technology, which promises to provide a denser, higher performance library, increased inherent drive current, and reduction through resistance. These are three features that are especially useful for data center processors. Meanwhile, Intel states that I3 nodes provide an 18% performance / watt improvement over I4 nodes. Intel’s I3 process node has never been listed in Intel’s future client PC products (Meteor Lake with I4 compute tiles will be taken over by Arrow Lake with I20A compute tiles). Heavy-duty data center applications rather than regular client products.
Custom data center solution for AWS
In addition, Intel announced during the call that it will develop a custom data center solution for Amazon Web Services.
“In the second quarter, we will expand our partnership with AWS to co-develop multi-generational data center solutions optimized for AWS infrastructure and include Intel as a strategic customer for internal workloads, including EDA. “We have agreed,” said Gelsinger.
Intel’s wording about expanding partnerships with AWS is ambiguous at best. The company doesn’t say exactly what it plans to co-develop with the cloud giant. However, given the fact that Intel’s main capability is chip development, we speculate that Intel is talking about data center-grade x86 system-on-chip (SoC) or multi-chiplet / multi-tile system-in-package (SiP). I can do it.
Intel’s intention to build highly customized data center-grade x86 processors for large clients such as AWS and Meta has always been an important part of the IDM 2.0 strategy. AWS is probably the perfect client for Intel and its foundry department. AWS consumes a lot of chips, can use almost any IP provided by Intel, can provide well-developed in-house IP explicitly designed for AWS-enabled services and platforms such as Netflix, and is hard. I am willing to pay a lot of money for the ware. needs.
Strategically, AWS’s insistence on x86 means that AWS will continue to use this architecture for the foreseeable future. Of course, AWS is a company that uses all available platforms, including AMD, Nvidia, and Arm (such as Ampere Altra / Altra Max based platforms). Still, of course, Custom Xeon shows that AWS is taking a very serious approach to customized x86-based solutions.
