New AMD ERAPS Feature Yields Additional Performance Gains On Zen 5
For workloads involving I/O or a lot of context switching there tended to be small but consistent improvements seen when running on the Linux kernel build patched with AMD Enhanced Return Address Prediction Security and thereby removing the explicit stuffing/filling on context switches. This also removes explicit RET stuffing/filling on VMEXITs as a win too for virtualization use.
The Apache Hadoop improvement was rather significant but in general that Hadoop benchmark is on the noisier side but in any event still resulting in a win.
The AMD EPYC 9005 series performance is already stellar and leading against the Intel and Arm server competition for a vast majority of workloads. With AMD ERAPS it looks to be some mostly subtle but consistent improvements from these pending Linux kernel patches for introducing AMD ERAPS and relaxing the relevant software mitigations. It will be interesting to learn more about Enhanced RAP Security when the official AMD documentation is published.
As mentioned I tested my patched ERAPS kernel on Ryzen 9000 series desktop processors and the feature was active there. So while the Linux patches have been talking about ERAPS in the context of EPYC 9005/Turin, the desktop/mobile Zen 5 CPUs appear to stand to benefit too. I will be conducting some ERAPS desktop workload benchmarks on Ryzen this week.
With not being part of the TIP x86/cpu branch at the moment, it looks like AMD ERAPS will be missing out on the Linux 6.13 cycle but hopefully we'll see this kernel code baked and ready for the Linux 6.14 cycle in early 2025.
If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.