Advanced Micro Devices Inc. used its annual presentation for Wall Street analysts Thursday afternoon to share early details about its next generation of chips, as well as the technologies that will power them.

Data center computing took center stage during the event, headlined by Chief Executive Officer Lisa Su (pictured). AMD revealed plans to deliver 5-nanometer central processing units for servers, a new specialized graphics card architecture dubbed CDNA and, eventually, three-dimensional chips that combine multiple types of silicon. 

First up, the CPUs. AMD will launch a new line of seven-nanometer server CPUs late this year under the Milan brand and intends to follow up the product family in 2022 with Genoa, a planned series of five-nanometer processors. The Milan chips slated to hit the market this year will use cores based on architecture that AMD is calling Zen 3 architecture, which it said will also form the basis of its 2020 consumer CPUs.

On the graphics card front, AMD is going down a different path. The company’s current enterprise graphic processing units are based on the same architecture as its consumer cards, mirroring the shared design of the Milan processors and AMD’s planned 2020 consumer CPUs. But for its next batch of GPUs, the chipmaker is preparing a new, separate architecture designed specifically with data centers in mind.

The company is calling it CDNA. Chips based on the architecture will be optimized for high-performance computing and artificial intelligence applications, AMD said, with the first generation of chips set to use a seven-nanometer process. The second generation, scheduled for release in 2022, will be based on an unspecified “advanced process.”

“AMD hadn’t done it until today because it couldn’t afford to have two architectures,” analyst Patrick Moorhead of Moor Insights & Strategy commented on CDNA. “I believe AMD can field a high-performance data center GPU, but it will need to invest at least as much in software to complete the solution.”

AMD’s enterprise chip roadmaps are underpinned by a couple of notable low-level innovations. One is Infinity Fabric 3.0, a planned technology for linking the company’s CPUs with its GPUs that will allow data to move between chips faster and thereby speed up processing. AMD also revealed it’s working on X3D, a technology that will allow it to stack multiple chips atop one another in one big 3D processor. 

“I was pleased to see AMD finally talk about this as Intel has owned the airwaves and has been showing off products with 3D packaging on stages for a very long time,” Moorhead said, referencing Intel Corp.’s Foveros chip stacking technology. AMD didn’t go into much detail about X3D, saying only that it’s expected to enable a tenfold increase in chip bandwidth density.  

Lastly, AMD previewed a number of upcoming updates to its consumer chip lineup. In the works is a new desktop GPU architecture with 50% better performance per watt than the company’s current Radeon cards and a series of seven-nanometer CPUs dubbed Ryzen 4000.

AMD is aiming to use these new products as a springboard to higher revenue growth. Executives told analysts that the company is looking to achieve a 20% compound annual growth rate by 2023 with gross margins of more than 50%. 

Images: AMD