AMD’s earlier this week hosted a presentation at the SC19 conference, an International Conference for High-Performance Computing, Networking, Storage, and Analysis. AMD made some interesting remarks about ZEN3 and Epyc in regards to development as well as IPC increases.
First off, AMD mentions the ZEN3 architecture design phase has been finished. Nowadays you can look at AMDs design and release phases as the traditional tick-tock release schedule. Here you may expect an iterated update of ZEN2. That’s said, ZEN3 should be seen as a new architecture is mentioned, which is interesting. We know that ZEN3 is a 7nm+ based product based on the below roadmap slide that was shared a while ago.
Zen1 compared to Zen2 brought a clock-for clock IPC increase of 21%, AMD says it careful, but hints at yet another 15% IPC increase for ZEN3, that’s again clock for clock and not counting faster clock frequencies. The necessity of more cores has not ended, and the future design path will be based on more cores and a greater compute density as well as a focus on memory bandwidth and I/O connectivity.
AMD qualified the remarks by pointing out that Zen 2 delivered a bigger IPC gain than what's normal for an evolutionary upgrade - AMD has said it's about 15% on average - since it implemented some ideas that AMD originally had for Zen but had to leave on the cutting board. However, he also asserted that Zen 3 will deliver performance gains "right in line with what you would expect from an entirely new architecture."
Despite earlier rumors, AMD also confirmed that ZEN3 and Epyc CPUs will not use a maximum of four threads per core, so you can 'bin' that idea. On the Epyc side, there was news as well, Amazon will be deploying Epyc 2 technology for their server farms and thus web services servers, which is huge for AMD of course. 64-core Rome processors will also be used in Microsoft's Azure service.