Intel ships delayed Sapphire Rapids Xeons to Argonne lab • The Register





Intel’s ever-delayed Sapphire Rapids Xeon Scalable processors are now available … in the chipmaker’s Dev Cloud. Anyone looking to actually pick up Intel’s next-gen datacenter silicon is, however, still out of luck, as it looks increasingly likely the chips won’t begin volume shipments until early next year.

During his keynote today at Intel’s Innovation event, CEO Pat Gelsinger announced both the Ponte Vecchio datacenter GPUs and high-bandwidth memory-equipped Sapphire Rapids CPUs had begun shipping to at least one customer: Argonne National Laboratory. This US government lab hopes to build its Aurora A21 exascale supercomputer out of Ponte Vecchio as well as Sapphire Rapids components.

“It’s the computing brains behind the Aurora supercomputer, and together we’re delivering these blades right now to Argonne National Labs to build the world’s highest performance supercomputer,” Gelsinger boasted, referring to Sapphire Rapids.

The lab machine was supposed to come online last year, and has been repeatedly delayed since 2017 by Intel’s inability to ship chips on time, including those Sapphire Rapids Xeons. It’s not clear how many have actually shipped to Argonne.

For everyone else still waiting on those processors, they may not have to wait much longer to get a taste of the chips, though that won’t happen in their own datacenters. Sapphire Rapids is one of several products Intel is making available in its Dev Cloud. The service is available in beta to select customers to explore Intel’s chips and software features.

“We’re pretty excited for developers to go kick those tires on it,” Lisa Spelman, corporate VP of Intel’s Xeon product division, said during a press briefing ahead of Innovation. “And this is a unique and rare opportunity to do so before you’ve even hit launch or had the opportunity to buy it and bring it into your own data center.”

If you want to actually install these processors in your own racks, or you’re not allowed into the Dev Cloud, you’ll probably have to wait to 2023.

Sapphire Rapids has become something of a sore spot for Intel, as the chip’s launch date has steadily slipped backwards. It was supposed to land in 2021, then that was revised to shipping to everyone in the first half of this year, then by the end of this year, and now likely next year.

If you’re feeling generous to Intel, you might want to consider that Sapphire Rapids is the chipmaker’s most ambitious to date. The Xeons take a cue from AMD and adopt a multi-die, aka chiplet, packaging technique. Presumably this was done to overcome yield issues and push core counts higher, another sore spot for Intel, which has trailed its x86 and Arm rivals in core count for several generations. Sapphire Rapids is also Intel’s first datacenter chip to support PCIe Gen 5.0, Compute Express Link, DDR5 memory, and HBM.

Intel declined to confirm the availability of Sapphire Rapids during the keynote. The Register asked for clarification on the time-frame, and was told Intel is not releasing any launch timing yet. That’ll fuel industry speculation of a first-quarter 2023 launch.

In an interview with Bloomberg, Sandra Rivera, EVP and GM of Intel’s Datacenter and AI Group, blamed overly ambitious design choices for the delays.

When we find issues, sometimes it will push out the schedule, but we think that’s the right trade off for our customers

And in response to questions this month, Ronak Singhal, senior fellow at Intel, told The Register Chipzilla is prioritizing quality with this generation.

“We won’t compromise that quality, even if it means delaying the product,” he said. “We’re going through a rigorous validation cycle, and when we find issues, sometimes it will push out the schedule, but we think that’s the right trade off for our customers, and what they’re expecting from us.”

Intel has confirmed at least one SKU won’t be making an appearance until next year. At Innovation this week, Intel shared details of a telco-optimized Xeon designed to accelerate radio access network (RAN) workloads running at cell sites.

The chipmaker’s vRAN Boost Xeon Scalable processors incorporate dedicated accelerators for virtual RAN tasks. vRAN allows a litany of specialized hardware used to operate cell towers to be virtualized, much in the way appliances like firewalls, routers, and switches have been virtualized in the datacenter.

By building the accelerators necessary to virtualize RAN functions into members of its Sapphire Rapids family, Intel said it can reduce complexity and curb power consumption up to 20 percent compared to traditional accelerator cards.

In the meantime, the datacenter CPU market is growing more crowded. AMD is expected to launch its fourth-gen Epyc Genoa this fall, Arm is aggressively pushing its Neoverse V2 architecture, which will power Nivdia’s Grace CPU, and Ampere is already working on a follow up to its Altra Max CPUs based a custom core design.

And if Sapphire Rapids does end up launching in 2023, it will not only face competition from other chipmakers, but Intel’s upcoming Emerald Rapids processors, which at least as of February are on track for a 2023 launch. ®




Leave a Comment