AMD claims Zen 2 has 29% higher IPC than Zen 1 in certain workloads
Update: AMD has made a statement to Notebookcheck clarifying the nature of the IPC improvement cited in the footnotes of its document, reiterating that this IPC uplift was calculated in a specific task that specifically used integer and floating point operations. It is not an aggregate for all applications.
As we demonstrated at our Next Horizon event last week, our next-generation AMD EPYC server processor based on the new “Zen 2” core delivers significant performance improvements as a result of both architectural advances and 7nm process technology. Some news media interpreted a “Zen 2” comment in the press release footnotes to be a specific IPC uplift claim. The data in the footnote represented the performance improvement in a microbenchmark for a specific financial services workload which benefits from both integer and floating point performance improvements and is not intended to quantify the IPC increase a user should expect to see across a wide range of applications. We will provide additional details on “Zen 2” IPC improvements, and more importantly how the combination of our next-generation architecture and advanced 7nm process technology deliver more performance per socket, when the products launch.
While AMD made strong statements about Zen 2's performance and power thanks to several architectural improvements (such as improved branch prediction and execution pipelines), the company never gave out hard numbers that could tell us how much Zen 2 improved upon Zen 1. AMD did show a performance demo of Rome, but they did not disclose the clock speeds, and it's hard to tell if every one of Rome's 64 cores scales perfectly, meaning architectural specific performance improvements are hard to determine. Thankfully, AMD did quietly disclose some information on Zen 2's IPC (or instructions per clock) improvements in the footnotes of an article they published on the day of the Next Horizon event. IPC is a very important figure for hardware because it boils down many architectural improvements into one number that is easy to compare. However, keep in mind that IPC is workload specific.
In this case, AMD claims a 29% increase in IPC when comparing Zen 2 to Zen 1 in "combined floating point and integer benchmarks." Floating point operations are commonly known as FLOPs. This is a common way to quickly describe how fast a CPU or GPU may be in a floating point heavy workload. According to AMD, in this type of workload the IPC uplift will be 29%, which is decently high considering AMD achieved a 52% IPC uplift when comparing Zen to Piledriver, based on the highly infamous (for its poor performance) Bulldozer architecture.
Of course, don't take AMD's word for granted, the IPC increase in reality may be slightly lower in floating point and integer workloads and it could be much lower in other workloads that do not utilize the FPU. FLOPs are important for the data center, so it makes sense AMD would have improved the ability of Zen to push those FLOPs. This may not translate to additional performance for games, web browsing, video encoding, and other common usages for CPUs that exist within the professional and consumer spaces. Ultimately though, AMD is claiming a significant improvement in IPC that would be impressive if it materializes.