AMD announced a comprehensive datacenter strategy to compete directly with Intel and Nvidia across CPUs and GPUs.

AMD has emerged from its long defensive crouch to taking the fight directly to Intel and Nvidia, backed by a company racking up wins. Coming on the heels of a record-setting quarter, AMD announced new EPYC server CPUs, a new line of Instinct brand GPUs it says are faster than Nvidia's best, the next generation of its CPU architecture, and a deal with Meta, formerly known as Facebook. AMD CEO Lisa Su introduced the EPYC Milan-X processors, an iteration of its third-generation server processors featuring a 3D-stacked L3 cache called 3D V-Cache, which reduces physical die size while increasing density by stacking transistors vertically instead of sprawling them horizontally.

The EPYC Milan-X chips are memory-dense, with up to 768MB of total L3 cache per chip for the 64-core design, meaning a dual socket server would have 128 cores and 1.5GB of L3 cache. AMD shared benchmarks showing up to a 50% performance improvement over previous chips. The chips will come to market in Q1 2022 in 16-, 32-, and 64-core variants and are drop-in compatible with older Naples and Rome-era server sockets, requiring only a BIOS upgrade. Su noted that EPYC is now designed into the data centers of 10 of the world's largest hyperscalers.

The second major announcement was the Instinct MI200 series, the second generation of AMD's GPU accelerators for data centers using the CDNA 2 architecture. The MI200 represents a major change by using multi-die packaging with chiplets connected via high-speed Infinity Fabric links operating at 25Gbps, providing up to 100GBps of bi-directional bandwidth between GPUs, with eight links offering 800 GBps of bandwidth between the two chiplets. AMD claims the Instinct MI200 series provides up to 4.9x higher peak performance for high performance computing workloads (FP64) than the Ampere A100 and 95.7 TFLOPS of peak double-precision matrix floating-point performance compared to Ampere's 19.5 TFLOPs peak. The MI200 along with EPYC processors will be used in the upcoming Frontier supercomputer at Oak Ridge National Labs, expected to be the first U.S. exascale supercomputer when deployed in 2022.

Su disclosed information on the fourth generation of Zen, AMD's microarchitecture launched in 2017 that brought the company back from the brink. Zen 4 will debut in the Genoa family in 2022, supporting DDR5 memory, PCI Express Gen 5, and CXL—a cache coherency link between memory and processors that speeds up transfers between main memory and the CPU. Unlike the first three compatible generations, Genoa breaks socket compatibility due to architectural changes. Su stated: "When introduced, we expect Genoa will be the world's highest performance processor for general purpose computing. It's designed to excel across a broad range of data center workloads from enterprise to HPC." Built by TSMC using a 5nm process, Genoa will have 96 cores. AMD also announced Bergamo, featuring a special core architecture called Zen 4C optimized for cloud-native applications with 128 high-performance cores, socket compatible with Genoa and on track to ship in the first half of 2023.

AMD has added Meta, formerly Facebook, to its list of hyperscaler wins. AMD and Meta worked together to define an open, cloud-scale, single-socket server designed for performance and power efficiency, based on the Milan processor, with further details to be discussed at the Open Compute Global Summit.