Just weeks after announcing the first details of its upcoming Ryzen 7000 series desktop CPUs, AMD has disclosed an outline of its consumer and server products roadmap for the next two generations. The news comes from AMD’s Financial Analyst Day, during which CEO Dr. Lisa Su spoke about the company’s plans to capture a part of the projected $300 billion market for high-performance computing solutions. While the event is geared towards financial analysts and shareholders, disclosures about AMD’s competitive positions in various markets include information about upcoming products. The company also now has a new slogan: “Together we advance”.
In the consumer market, AMD’s upcoming ‘Zen 4’ CPU core architecture will be the basis of the ‘Raphael’ generation of desktop CPUs, which will come to market as the Ryzen 7000 series before the end of this year. This will be a high-end product line using a 5nm manufacturing process, and AMD promises an 8 – 10 percent uplift in IPC (instructions per clock) performance as well as at least a 25 percent improvement in terms of performance per Watt. 4nm versions are also planned, though it isn’t clear what the segmentation will be.
Zen 4 will also power the upcoming ‘Genoa’ line of Epyc server CPUs as well as a new line of ‘Bergamo’ based on a density-optimised variant called Zen 4c for cloud-native computing applications. ‘Genoa-X’ variants will feature AMD’s integrated 3D V-cache, a vertically stacked layer of high-speed memory on top of the CPU die. Another product line codenamed ‘Siena’ will target a new market in intelligent edge and communications equipment.
Following that, ‘Zen 5’ is a ground-up redesign planned for release in 2024 and should further improve performance and efficiency as well as introduce new optimisations for AI and machine learning for consumer Ryzen CPUs codenamed ‘Granite Ridge’. These CPUs will use 4nm and 3nm processes. The next Epyc generation will be codenamed ‘Turin’ and should be released in late 2024.
AMD also confirmed that its next GPU microarchitecture, codenamed RDNA 3, will allow for a chiplet-based modular GPU design and will use a 5nm manufacturing process. It’s said to deliver a 50 percent improvement in performance per Watt. A 4th-gen Infinity Architecture interconnect standard will allow AMD to integrate third-party chiplets allowing for heterogenous platforms and increased customisability for clients.
RDNA 3 will be seen in next-generation ‘Navi 3’ discrete GPUs, launching as the next Radeon RX series of products later this year. This architecture will build on RDNA 2, which has been used by multiple game console manufacturers this generation, hinting that the next generation will continue to leverage AMD’s hardware. RDNA 3 will also be integrated into upcoming ‘Phoenix Point’ mobile CPUs based on the Zen 4 architecture in 2023, with ‘Strix Point’ following in 2024.
For professional markets, the CDNA 3 architecture and XDNA, which is the result of AMD’s acquisition of Xilinx, will help drive significant performance gains for new products in the AI and high-performance compute space. New Instinct MI300 accelerators for AI training and Alevo smart NICs for confidential computing are also expected to add to the company’s data centre portfolio.
AMD sees growth in PC, console and cloud gaming as well as interactive metaverse applications and 3D content creation. Xilinx IP will be integrated across product lines to drive performance in AI inference and training workloads for consumers and enterprises. The company is also working on a unified AI software roadmap to make development more cohesive across various products.