Intel delivered its rebuttal to AMD’s EPYC 3 data center chips today with the launch of the long-awaited 10-nanometer Ice Lake Xeon family.

In addition to using Intel’s leading-edge process node, the new chips feature a revamped core architecture, improved security, and higher core counts that now top out at 40. All of this together makes for a processor that Intel claims will deliver a 20% improvement in instructions per clock (IPC) over the previous-generation Cascade Lake refresh and a 46% improvement in “popular data center workloads.”

“We’re excited about getting up to the 40 cores and offering that flexibility to customers,” said Lisa Spelman, VP and GM of Intel's Xeon products and data center marketing. “The increased core count, combined with built-in acceleration features, allows us to deliver really strong gen-on-gen performance with an average increase of 1.46 times versus previous generations.”

Ice Lake Arrives in the Data Center

The launch of Ice Lake in the data center marks a major milestone for Intel, which has recently undergone a complete restructuring and change of leadership, following repeated delays to its 10-nanometer and upcoming 7-nanometer manufacturing capabilities.

The move to a 10-nanometer process is a significant one for the company, as it has been stuck on 14-nanometers for more than six years now due to challenges associated with the extreme-ultraviolet lithography used to fabricate the new chips.

The chips announced today target single- and dual-socket servers as well as workloads ranging from telecommunications and edge to data center and cloud deployments.

“We made significant improvements to the compute portion of the chip, in the memory domain, and I/O [input/output] domain as well as how we bring the entire [system on chip] together with a balanced architecture,” said Sailesh Kottapalli, chief architect for data center processors at Intel.

Many of these improvements are rooted in Intel’s new Sunny Cove core architecture. Sunny Cove includes an improved front-end branch predictor, larger level-two (L2) caches, enhanced vector processing capabilities, and other structural changes to enhance performance in data center and cloud workloads.

Intel Leans on Vector, AI Accelerators

With Ice Lake, Intel continued to push integrated AI acceleration delivered by its Advanced Vector Extension (AVX) and Deep Learning Boost capabilities.

Intel’s AVX and Deep Learning Boost accelerators are designed to accelerate vector calculations and artificial intelligence workloads respectively, but code has to be optimized for them. Dave Hill, senior data center performance director at Intel, argued that these capabilities give Intel a significant performance advantage over AMD’s latest EPYC processors, at least in optimized workloads.

And in a heavily cherry-picked selection of benchmarks comparing the chipmaker’s new 40-core Xeon Scalable processor against AMD’s recently announced Zen 3-based 64-core EPYC 3, Intel touted a substantial performance advantage in workloads taking advantage of AVX and Deep Learning Boost technologies. This, of course, comes as no surprise as AMD’s chips aren’t optimized for either technology.

“These workload categories that we've chosen really leverage our workload acceleration instructions because they really show the benefit that you don't necessarily need more cores, you can deliver even better performance with fewer cores with software that's optimized for the workload acceleration instructions,” Hill said.

Intel declined to provide additional context regarding how the chips compared in non-AVX or-DL-Boost accelerated workloads. "There will be many workloads that get posted to SPEC and other industry-standard benchmarks by OEMs. The timing of when those are available is up to the OEM on when they submit the results to SPEC.org," a spokesperson for Intel said in an email to SDxCentral.

The chipmaker does claim a roughly 150% improvement in performance over Cascade Lake in SPECrate 2017’s integer base benchmark. Based on performance benchmarks submitted to SPEC.org and scores provided by AMD, this would likely put Intel’s 28-core Ice Lake processors somewhere between AMD’s 32-and 48-core EPYC 3 processors in performance.

Intel Adopts PCIe Gen 4.0, Expands SGX

In addition to a substantial performance boost over the company's aging 14-nanometer processors, Intel’s new chips deliver a bevy of improvements including higher I/O and memory bandwidth as well as new security capabilities.

“With the latest third-gen we’re increasing I/O bandwidth, the number of memory channels, and memory capacity with Optane persistent memory,” Spelman said.

Ice Lake marks the first generation of Xeon Scalable processors to add support for PCIe-Gen 4.0, which provides twice the bandwidth of the previous generation. What’s more, Intel has increased the number of PCIe lanes to 64, up from 48 on the previous generation.

Ice Lake also gains support for up to eight lanes of memory, maxing out at six terabytes of memory per socket.

And on the security front, Ice Lake adds support for Intel’s Software Guard Extension (SGX).

“We’ve offered SGX for several years now on the Xeon E platform and now we have the opportunity to the two-socket platform,” Spelman said. “The third-gen Xeon Scalable is Intel's first mainstream two-socket data center processor to feature SGX.”

The addition of SGX delivers larger secure enclaves for confidential computing applications, while integrated cryptography acceleration enables the CPU to process encrypted workloads without a substantial performance hit, Spelman explained.

Optane, Networking, and FPGAs

Launching alongside the revamped Xeon Scalable processors are Intel’s new Optane persistent memory modules and solid-state storage, 800-series Ethernet NICs, and AgileX FPGAs.

Intel claims its 200-series Optane memory modules now deliver up to 32% higher bandwidth by eliminating cache flushes. Persistent memory allows for much higher capacities compared to traditional memory modules and can store workloads long after the server has powered down or a workload has been terminating, thus allowing for much faster resume times the next time the workload is called on.

Intel’s Optane solid state drives are based on the same memory technology as its persistent memory modules but are designed to accelerate slower bulk storage operations, improving response times for hyperconverged infrastructure (HCI), virtual desktop infrastructure (VDI), and database workloads.

Intel claims improvements to its Optane memory technology allow for a 2,600% increase in the number of operations per second per gigabyte.

Intel also debuted its 800-series NICs, which capable of up to 200 GB/s of total throughput over standard network interfaces. The NIC is available in several form factors and supports numerous protocols, including NVMe.

Finally, Intel announced the availability of its Agilex FPGA, which is capable of 400 GB/s Ethernet and is targeted at 5G, cloud, and edge deployments.

It’s this ecosystem that Intel sees as a key differentiator as the company faces stiff competition from its competitors in the CPU arena. “We see customers continuing to choose Xeon because of this massive ecosystem that we’ve built around it and that we continue to nurture and grow,” Spelman said.