AMD CEO Lisa Su teased the company’s upcoming third-generation EPYC processors during the opening keynote of the CES 2021 virtual conference.

“Third-gen EYPC will reset the bar for data center computing. With up to 64 leadership Zen 3 cores, our next generation server processors extend our leadership in performance, total cost of ownership, and security,” boasted Su.

EPYC 3, code named Milan, are expected to begin shipping later this quarter, and as Su mentioned, the chips are AMD’s first to utilize the chipmaker’s new Zen 3 core design.

Introduced in November 2020, with the launch of AMD’s Ryzen 5000 series desktop chips, the Zen 3 microarchitecture is a complete redesign of the original Zen architecture, the company said. AMD claims Zen 3-based chips boast 19% higher instructions per clock, while larger pools of level 3 cache helps to reduce latency over the previous generation Zen 2 architecture.

However, like Zen 2, AMD is still using Taiwan Semiconductor Manufacturing Co.’s (TSMC) 7-nanometer manufacturing process for both Zen 3 and EYPC 3. TSMC’s new 5-nanometer process, which is already being used by chipmakers like Apple, won’t come to AMD's processors until the launch of Zen 4 sometime in 2022.

Su teased the upcoming chips by sharing some performance metrics for its EYPC chips running the Weather Research and Forecasting model (WRF) as a benchmark.

WRF is used to crunch large amounts of climate data to simulate weather conditions and generate forecasts. In this sole benchmark Su claimed the dual socket AMD server outperformed a “comparable” Intel system by a wide margin.

“The third-generation EPYC server completed the forecast 68% faster than the competition,” Su said. “The speed to complete in this case is very important because if the simulation runs faster, it means scientists can improve the accuracy of the forecast by using a larger data set for the model, or run more simulations per day.”

While the benchmark gives some indication of how EYPC 3 will perform compared to Intel’s Xeon scalable processors, it’s hardly a fair fight. The benchmark demonstrates chips performance in a single application, likely chosen by AMD to highlight the performance delta with its competition.

What's more, AMD is comparing its new 32-core EPYC 3 processors against Intel’s now year-old, second-generation Xeon scalable processors, specifically the Xeon Gold 6258R.

Even in a dual-socket configuration, the Intel server is at a notable core deficit compared to the AMD system, which features 64 cores and 128 threads compared the Intel system’s 56 cores and 112 threads.

AMD's EPYC 3 will face a far more comparable challenger with the release of Intel's 10-nanometer Ice Lake Xeon Scalable processors early this year.

Gregory Bryant, executive VP and GM of client computing at Intel, earlier this week said the new chips — which are the first to use the company's long-delayed 10-nanometer manufacturing process — have entered volume production with shipments expected to begin later this quarter.

While Intel hasn’t disclosed core counts for its upcoming server processors, Bryant hinted that we could see Xeons with more than 28 cores with the upcoming release.

“Ice Lake delivers significant increases in core count, performance, integrated AI, and security features across a wide variety of workloads,” he said.