Just weeks after asserting the primary particulars of its upcoming Ryzen 7000 collection desktop CPUs, AMD has disclosed a top level view of its client and server merchandise roadmap for the following two generations. The information comes from AMD’s Financial Analyst Day, throughout which CEO Dr. Lisa Su spoke in regards to the firm’s plans to seize part of the projected $300 billion marketplace for high-performance computing options. While the occasion is geared in the direction of monetary analysts and shareholders, disclosures about AMD’s aggressive positions in numerous markets embody details about upcoming merchandise. The firm additionally now has a brand new slogan: “Together we advance”.
In the buyer market, AMD’s upcoming ‘Zen 4’ CPU core structure would be the foundation of the ‘Raphael’ technology of desktop CPUs, which can come to market because the Ryzen 7000 collection earlier than the tip of this yr. This can be a high-end product line utilizing a 5nm manufacturing course of, and AMD guarantees an 8 – 10 % uplift in IPC (directions per clock) efficiency in addition to at the very least a 25 % enchancment in phrases of efficiency per Watt. 4nm variations are additionally deliberate, although it is not clear what the segmentation can be.
Zen 4 may also energy the upcoming ‘Genoa’ line of Epyc server CPUs in addition to a brand new line of ‘Bergamo’ based mostly on a density-optimized variant referred to as Zen 4c for cloud-native computing purposes. ‘Genoa-X’ variants will function AMD’s built-in 3D V-cache, a vertically stacked layer of high-speed reminiscence on prime of the CPU die. Another product line codenamed ‘Siena’ will goal a brand new market in clever edge and communications gear.
Following that, ‘Zen 5’ is a ground-up redesign deliberate for launch in 2024 and will additional enhance efficiency and effectivity in addition to introduce new optimizations for AI and machine studying for client Ryzen CPUs codenamed ‘Granite Ridge’. These CPUs will use 4nm and 3nm processes. The subsequent Epyc technology can be codenamed ‘Turin’ and must be launched in late 2024.
AMD additionally confirmed that its subsequent GPU microarchitecture, codenamed RDNA 3, will permit for a chiplet-based modular GPU design and can use a 5nm manufacturing course of. It’s stated to ship a 50 % enchancment in efficiency per Watt. A 4th-gen Infinity Architecture interconnect customary will permit AMD to combine third-party chiplets permitting for heterogenous platforms and elevated customisability for shoppers.
RDNA 3 can be seen in next-generation ‘Navi 3’ discrete GPUs, launching as the following Radeon RX collection of merchandise later this yr. This structure will construct on RDNA 2, which has been utilized by a number of recreation console producers this technology, hinting that the following technology will proceed to leverage AMD’s {hardware}. RDNA 3 may also be built-in into upcoming ‘Phoenix Point’ cell CPUs based mostly on the Zen 4 structure in 2023, with ‘Strix Point’ following in 2024.
For skilled markets, the CDNA 3 structure and XDNA, which is the results of AMD’s acquisition of Xilinx, will assist drive vital efficiency features for brand spanking new merchandise in the AI and high-performance compute area. New Instinct MI300 accelerators for AI coaching and Alevo sensible NICs for confidential computing are additionally anticipated to add to the corporate’s knowledge heart portfolio.
AMD sees development in PC, console and cloud gaming in addition to interactive metaverse purposes and 3D content material creation. Xilinx IP can be built-in throughout product strains to drive efficiency in AI inference and coaching workloads for customers and enterprises. The firm can be engaged on a unified AI software program roadmap to make improvement extra cohesive throughout numerous merchandise.
