Like clockwork, the major data center original equipment manufacturers (OEMs) announced a torrent of new server chassis built around Intel’s Ice Lake Xeon Scalable processors. The new processors, announced this morning, deliver a 20% uplift in instructions per clock over the previous generation, but for OEMs, Intel’s new third-gen Xeons deliver more than a performance boost.

In many ways, Ice Lake, which is Intel's first generation of chips based on its 10-nanometer process node, represents the most compelling upgrade to Intel’s Xeon lineup since the launch of its Skylake-based 14-nanometer processors in 2015.

According to Vik Malyala, SVP at Supermicro, Intel’s Ice Lake refresh is compelling thanks to three core promises: higher performance per core, expanded memory bandwidth, and PCIe-Gen 4.0 connectivity.

And Malyala isn’t alone. Lenovo's Kamran Amini, VP and GM of Lenovo's Infrastructure Solutions Group, highlighted several of these improvements as key enablers of new workloads tailored to customer demand.

This is because Intel’s new Xeons are its first to support PCIe-Gen 4.0, bringing the chipmaker inline with rival AMD, which has supported the technology for several generations.

The upgrade from PCIe-Gen 3.0 to Gen 4.0 might not sound that significant until you realize that each new generation of the PCIe standard delivers a two-fold increase in bandwidth over the previous generation. What’s more, Intel has bolstered the number of PCIe lanes from 48 to 64 in Ice Lake. All told, Intel’s latest third-gen Xeon Scalable processors offer more than twice the input/output (I/O) bandwidth of it is older chips.

For Malyala, the move to PCIe-Gen 4.0 opens the door to higher density PCIe-based storage servers based on Intel’s processor cores.

“Now, we are able to have all PCIe-Gen 4.0 U.2 drives for storage, so, the local storage performance is improving,” he said. "There's an increased interest in [virtual RAID on CPU], for people who are trying to do the software-based RAID on NVMe drives."

He adds that the additional PCIe lanes also mean that servers can support larger drive arrays without needing to rely on a PCIe switch, which dramatically improves performance and eliminates overhead.

And it's not just storage. Both Supermicro and Lenovo see the move to PCIe-Gen 4.0 as critical to enabling support for larger banks of GPU accelerators.

“We’re seeing broader use cases [for GPUs]. I think people think GPUs are there for high-performance computing (HPC) or artificial intelligence (AI). We think it’s much broader than that,” Amini said. “I think what Intel is doing with [PCIe] Gen 4.0 will help them accelerate something that they had been missing today from an I/O capability perspective.”

Another key change with Intel’s Ice Lake refresh was the boost in core counts from 28 cores to 40. This in conjunction with the improved PCIe bandwidth enables customers to move workloads to single-socket servers without compromising on the number of add-in cards, Malyala explained.

“People tend to keep the blast radius small. If something were to go down, they don’t want 200 [servers] to go down,” he said, adding that because of this, customers have shied away from the “top binned” Intel processors, instead favoring 12-and 16-core processors.

“If you look at that sweet spot, with Ice Lake, now all of a sudden I can replace a dual-socket [server] with a single socket and I’m able to reduce my cost,” Malyala said.

The new chips also feature expanded memory densities up to eight channels and six terabytes of memory when used in conjunction with Intel’s Optane persistent memory.

Malyala explains that this capability enables Supermicro to offer much higher memory densities in its servers, including the company’s Twin servers, which are physically constrained to 16 memory slots. By using a combination of 128-gigabyte memory dims and Intel’s Optane memory, it’s now possible to achieve four terabytes of memory per node into those systems.

“This is one thing that you know we couldn't imagine before, and we are able to make it happen,” he said.

These capabilities are also opening the door to new server platforms from Lenovo designed for memory-intensive workloads.

“Intel really announced a broad set of solutions last generation around persistent memory with Optane technology," Amini said. "We're taking that same technology and now expanding it to other applications and workloads that make sense.”

He said Optane has been ideal for applications like SAP HANA, but with improved memory speeds, Lenovo is looking to extend Optane to new workload-specific server platforms.

Finally, in the lead up to the Ice Lake refresh, Intel talked extensively about developing workload-specific SKUs for things like data center, cloud, edge, and 5G. And this extends beyond the chips to the OEMs building systems around them. According to Malyala, Intel has built an ecosystem around its processor families that make it easy for customers to change from one generation to another and focus on what they care about.

With Ice Lake, “we are focusing more on what kind of markets can take advantage of it, not just in the data center or HPC, which have been the core for us, but we’re talking about the edge … and a subset of that is going to be the 5G infrastructure,” Malyala said, explaining that customers want to focus their energy on their business, not validating multiple platforms.

And one of the ways that Intel has helped to build this ecosystem is by developing custom accelerators for specific vector or machine learning workloads in the form of the Advanced Vector Extension (AVX) and Deep Learning Boost. While these accelerators have to be specifically coded for, Malyala argues that the adoption of these technologies is growing.

“If you take a look at the previous generation versus the current generation, the AVX-512 adoption has increased quite a bit,” he said. “People have actually developed their software to run better on that, which means the investment has gone in. … It’s not about speeds and feeds anymore.”

Amini echoed these sentiments, noting that Ice Lake better positions Lenovo to serve customers no matter what workloads they want to run.

“We're trying to ensure that workloads customers want to run Lenovo can server them with either part. I think it comes down to economics, use cases, applications, a variety of things for customers to decide which architecture, he said.

Supermicro and Lenovo are far from the only OEM's rolling out new servers based on Intel's third-generation Xeon Scalable processors. Last month Dell announced the availability of 17-new servers running on Intel's Ice Lake and AMD's EPYC 3 processors ahead of the official launch. Today, Hewlett Packard Enterprise, Cisco, and Gigabyte were among the many announcing support for the chip.