AMD revealed its CPU server roadmap at its Monetary Analyst Day 2022 and likewise unveiled 4th-Gen EPYC Genoa-X processors with 3D V-Cache, the sting and telecommunications-focused EPYC Siena, and AMD’s first knowledge heart APU, the MI300X. As well as, AMD teased the Fifth-gen EPYC Turin chips that can launch earlier than the tip of 2024.
AMD EPYC CPU Server Roadmap
Right here we will see the AMD server roadmap till the tip of 2024. We even have the CPU core roadmap within the second slide, which you’ll be able to learn extra about right here. The server CPU roadmap reveals a major growth of AMD’s goal markets with the 4th-Gen EPYC household. These chips all include the Zen 4 microarchitecture.
AMD will now have general-purpose server chips with its customary EPYC Genoa silicon, a high-performance optimized EPYC Genoa-X with 3D V-Cache, cloud-optimized chips with the 128-core 256-thread Bergamo, and edge- and telecommunications-focused Siena processors. Additional beneath, we’ll dive into these new chips, together with AMD’s first knowledge heart APU, the Intuition MI300.
AMD introduced that the Fifth-generation EPYC Turin household will come to market in 2024 however did not reveal any additional particulars.
The 128-core 256-Core Bergamo and 96-Core Genoa
AMD shared a demo of a 96-core EPYC Genoa offering a >75% uplift over the current-gen EPYC flagship in a Java SPECjbb benchmark. This achieve comes from the transfer to the 5nm Zen 4 structure, together with the 12-channels of DDR5 reminiscence. That’s to not point out the additional 32 cores over the previous-gen EPYC. Genoa can also be notable for supporting CXL-attached reminiscence. AMD says Genoa stays on observe for launch within the fourth quarter of this yr.
The 128-core Bergamo comes with the density- and scale-out optimized Bergamo comes with as much as 128 Zen 4c cores and an astounding 256 threads in a single socket. Bergamo drops into the identical SP5 socket because the Genoa processors and helps the complete Zen 4 ISA.
The Zen 4c cores are conceptually just like the effectivity cores (e-cores) that we see with different varieties of chip architectures in each Arm and x86 flavors. These ‘c’ cores are smaller than the usual Zen 4 core that can debut in Genoa, with sure unneeded performance eliminated to enhance compute density. The chips have a density-optimized cache hierarchy to extend core counts, thus addressing cloud workloads that want larger thread density. This might imply that the chips have a smaller cache(s), or maybe a cache stage has been eliminated, however AMD hasn’t shared particulars. The Zen ‘c’ cores help the complete Zen 4 ISA — in contrast to Intel does with Alder Lake, AMD will not disable some options like AVX.
AMD EPYC Genoa-X and Siena
AMD revealed that it’s going to have a model of its upcoming EPYC Genoa processors that can come armed with as much as 1+ GB of L3 cache. The 5nm Genoa-X processors will include as much as 96 cores, similar to the usual Genoa merchandise, and drop into the identical socket. In impact, these are the Zen 4 equivalents to the 3D V-Cache outfitted EPYC Milan-X processors.
AMD can also be increasing into the telecommunications and edge markets with its new Siena, a streamlined chip structure with as much as 64 Zen 4 cores on supply. AMD hasn’t shared many particulars on these chips but, however they’re optimized for energy effectivity and characterize an vital growth of AMD’s addressable market.
Each Genoa-X and Siena arrive in 2023.
AMD Intuition MI300 APU – Zen 4 Meets CDNA 3
The AMD Intuition MI300 is the corporate’s first APU for the information heart, that means it should have each x86 CPU and GPU cores built-in onto the identical chip package deal. AMD makes use of the 4th-gen AMD infinity Structure to tie collectively the Zen 4 CPU cores and CDNA 3 GPU cores with a coherent reminiscence interface. The chip makes use of 3D packaging tech, with CPU, GPU, cache, and HBM within the single-chip package deal, however it isn’t clear if all of those elements are stacked vertically or that solely sure elements are 3D-stacked. AMD claims this accelerator gives as much as an 8X efficiency enhance in AI workloads over the MI250X. You may learn extra about this chip right here.
The AMD occasion is ongoing; we’ll replace as we study extra…