The Performance Impact Of Intel's Register File Data Sampling "RFDS" Mitigation

Written by Michael Larabel in Software on 15 March 2024 at 03:00 PM EDT. Page 2 of 5. 21 Comments.
QuantLib benchmark with settings of Configuration: Multi-Threaded. Vulnerable was the fastest.

It didn't take long at all before seeing performance hits as a result of this new default mitigation on the Intel Core i9 14900K.

DaCapo Benchmark benchmark with settings of Java Test: Apache Tomcat. Vulnerable was the fastest.
DaCapo Benchmark benchmark with settings of Java Test: jMonkeyEngine. Vulnerable was the fastest.
DaCapo Benchmark benchmark with settings of Java Test: Apache Cassandra. Vulnerable was the fastest.
DaCapo Benchmark benchmark with settings of Java Test: Apache Xalan XSLT. Vulnerable was the fastest.

In most Java OpenJDK workloads there wasn't much or any of a performance difference.

DaCapo Benchmark benchmark with settings of Java Test: FOP Print Formatter. Vulnerable was the fastest.
DaCapo Benchmark benchmark with settings of Java Test: Apache Lucene Search Engine. Vulnerable was the fastest.
DaCapo Benchmark benchmark with settings of Java Test: Avrora AVR Simulation Framework. Vulnerable was the fastest.
DaCapo Benchmark benchmark with settings of Java Test: BioJava Biological Data Framework. Vulnerable was the fastest.
DaCapo Benchmark benchmark with settings of Java Test: H2O In-Memory Platform For Machine Learning. Vulnerable was the fastest.

But for some Java workloads, the mitigated state of RDFS was causing a measurable performance difference.

Timed Linux Kernel Compilation benchmark with settings of Build: defconfig. Vulnerable was the fastest.
Timed Linux Kernel Compilation benchmark with settings of Build: allmodconfig. Vulnerable was the fastest.

For code compilation workloads like compiling the Linux kernel there was a minor hit with the RFDS mitigation active. Again, all other (non-RFDS) mitigations were at their respective boot time defaults.


Related Articles