AMD Chief Technology Officer Mark Papermaster said at Wells Fargo’s 6th TMT Summit in 2022 that the company’s next-generation Instinct MI300 accelerated processing unit for data centers and high-performance computing has been implemented in AMD’s Up and running in the lab.
The AMD Instinct MI300 APU adopts AMD’s Zen 4 and CDNA 3 architecture and will be used in the El Capitan supercomputing in the United States. The supercomputing is expected to be put into use in 2024, and the FP 64 peak computing performance will exceed 2 ExaFLOPS.
AMD’s Instinct MI300 is a multi-chip APU featuring Zen 4 general-purpose x86 cores and a CDNA 3-based compute GPU that share a unified pool of packaged memory, including Infinity Cache cache and HBM shared memory design. AMD will use TSMC’s N5 (5nm-class) manufacturing process to produce CPU/GPU chipsets for the MI300, but the company did not disclose the number of CPU cores and GPU stream processors for the Instinct MI300.
Papermaster also revealed that AMD is expected to launch an EPYC Bergamo processor competing with Arm servers such as the AWS Graviton series in the first half of 2023, using the Zen 4C architecture.
AMD EPYC Bergamo processors will feature up to 128 cores and will be optimized for HBM-powered Xeon chips, as well as server products from Apple, Amazon and Google, using the SP5 socket, optimized for higher-throughput workloads.
Additionally, the AMD Genoa-X CPU, expected to be in production late Q3/early Q1 2023 and to launch around mid-2023, will feature a similar architecture to the Milan-X chip with 3D V-Cache design method. The Genoa-X CPU will be equipped with more than 1GB of L3 cache and up to 96 cores based on the Zen 4 architecture.
