SMT Proves Worthwhile Option For 128-Core AMD EPYC "Bergamo" CPUs

While the AMD EPYC 9754 "Bergamo" processor is impressive for having 128 physical Zen 4C cores, it also has Simultaneous Multi-Threading (SMT) to provide for 256 threads per socket. Meanwhile with Ampere Altra Max and AmpereOne there is no SMT and it's likely Intel's upcoming Sierra Forest will also lack SMT (Hyper Threading) given it's an E-core-only design. But that led to my curiosity over the SMT impact for Bergamo on power and performance when leveraging SMT for the 128-core flagship EPYC 9754. Today's Bergamo benchmarking is looking at SMT on and off for both 1P and 2P server configurations.

With the new Zen 4C core I was curious about the benefit of SMT with these very dense processors. SMT at least for lower core counts does tend to provide some measurable benefits across a range of workloads while the main downside of having Bergamo (Zen 4C) with SMT is that there could end up being some public cloud service providers that leverage SMT threads as part of their vCPU rating and advertising. At least from some of the major cloud service providers in recent years we've seen them leverage AMD EPYC servers with SMT disabled so each vCPU is backed by a physical CPU core. SMT off can also be better in the name of security. But in any event I wanted to see what benefits are there for bare metal workloads if tossing in an extra 128 threads via SMT or 256 threads for the dual EPYC 9754 configuration.

Long story short, SMT did still provide measurable benefit for numerous workloads -- while having only a small impact on power consumption. For workloads heavily reliant on system memory bandwidth, disabling SMT tended to deliver better performance while for workloads less sensitive to memory bandwidth and more compute bound there tended to be at times very significant advantages to running the EPYC 9754 with SMT enabled.

This round of testing was with SMT on and off as toggled from the system BIOS while testing the EPYC 9754 in both single and dual socket configurations on the AMD Titanite reference server. Ubuntu 22.04 LTS with its stock Linux 5.19 kernel was running on the system throughout all of this testing. No other changes were made to the server state during testing besides swapping 1P for 2P CPUs and toggling SMT.

Besides looking at the raw performance results, the total CPU power consumption as measured via the RAPL/PowerCap sysfs interfaces were also monitored for seeing the SMT impact on CPU power consumption with these Zen 4C cores.