A couple of weeks ago, I received an interesting email from a reader who had observed high CPU utilization after upgrading VCF Automation (VCFA) to the latest VCF 9.1 release. They were running VCF 9.1 on the popular Minisforum MS-A2 platform, which uses AMD Zen4/Zen5 (Ryzen) consumer processors which are not supported on the Broadcom Compatibility Guide (BCG). As I have covered previously (here and here), these processors have been known to exhibit slower entropy generation that can lead to higher than expected CPU utilization without the appropriate optimizations.
With the help of Claude to analyze the issue, the reader arrived at the following conclusion:
Five JVM-based services were dying repeatedly during startup because FIPS_MODE=strict was slowing down TLS handshakes (both database and internal service-to-service HTTPS) enough to occasionally exceed connection timeouts. Every restart meant a full JVM boot: Spring context initialization, Liquibase migrations, Hibernate, index rebuilding, which is expensive. The crash loop itself became the CPU load. Setting FIPS_MODE=disabled on the affected deployments stopped the crash loop entirely, and CPU usage returned to near-idle.
Interestingly, around the same time, I also heard from an internal colleague who had independently observed that disabling FIPS for specific VCFA services significantly reduced CPU utilization. Since FIPS is enabled by default for VCF 9.1 deployments, and there are known performance implications with FIPS and OpenSSL 3.x, these effects can be amplified on AMD Zen4/Zen5 (Ryzen) consumer processors due to their slower entropy generation, resulting in higher than expected CPU utilization.
Fortunately, both individuals had identified workarounds for the issue, but their recommendations did not completely overlap. Before sharing the guidance more broadly, I wanted to spend some time validating the solution to identify the minimal set of changes required to consistently achieve the same outcome.