Intel 5th Gen Xeon Performance Benchmarks: Impressive Efficiency Gains With "Optimized Power Mode"

Written by Michael Larabel in Processors on 18 December 2023 at 10:00 AM EST. Page 2 of 6. 21 Comments.
Intel Optimized Power Mode Xeon Platinum Benchmarks

The dual Intel Xeon Platinum 8592+ was running Ubuntu 23.10 with the Linux 6.5 kernel and using the Intel P-State performance governor across all of the testing. Again, the "default" run is the out-of-the-box state while the Optimized Power Mode is with the only hardware/software change being enabling the Optimized Power Mode setting from the system BIOS.

QuantLib benchmark with settings of Configuration: Multi-Threaded. Optimized Power Mode was the fastest.
QuantLib benchmark with settings of Configuration: Multi-Threaded. Optimized Power Mode was the fastest.
QuantLib benchmark with settings of Configuration: Multi-Threaded. Optimized Power Mode was the fastest.

With the multi-threaded QuantLib benchmark for that quantitative finance software, for this workload that does a good job keeping all of the CPU cores busy there was no difference in the benchmark result nor the combined 2P power consumption for the Xeon Platinum 8592+ processors....

QuantLib benchmark with settings of Configuration: Single-Threaded. Default was the fastest.
QuantLib benchmark with settings of Configuration: Single-Threaded. Default was the fastest.
QuantLib benchmark with settings of Configuration: Single-Threaded. Default was the fastest.

But when running QuantLib in its single-threaded benchmark mode immediately showed an interesting difference: the performance was effectively the same but on average the dual Xeon Platinum 8592+ server dropped its power consumption on average by around 20 Watts. The peak was much lower at 191 vs. 274 Watts. Activating Optimized Power Mode improved the single-threaded QuantLib performance per Watt by 10% for this simple change.

OpenFOAM benchmark with settings of Input: motorBike, Execution Time. Default was the fastest.
OpenFOAM benchmark with settings of Input: motorBike, Execution Time. Default was the fastest.

When running the OpenFOAM CFD software with the rather simple motor bike test, the Optimized Power Mode performance was about 1 second slower but the CPU power consumption on average for this dual Emerald Rapids processor configuration was nearly 80 Watts lower and a peak power consumption 117 Watts lower.

OpenFOAM benchmark with settings of Input: drivaerFastback, Small Mesh Size, Execution Time. Optimized Power Mode was the fastest.
OpenFOAM benchmark with settings of Input: drivaerFastback, Small Mesh Size, Execution Time. Optimized Power Mode was the fastest.

With the more demanding drivaerFastback model, the Optimized Performance Mode result was similar to the out-of-the-box performance but with the average 2P power consumption was 33 Watts lower while the peak power consumption was 37 Watts less.

OpenRadioss benchmark with settings of Model: Chrysler Neon 1M. Default was the fastest.
OpenRadioss benchmark with settings of Model: Chrysler Neon 1M. Default was the fastest.

Or with Altair's OpenRadioss Chrysler Neon 1M model that does a typically good job saturating the CPU, with Optimized Power Mode enabled it delivered the same performance on the Xeon Platinum 8592+ 2P server while doing so at around 30 Watts less... Again not bad for a workload keeping the CPU busy and delivering the same performance with a simple BIOS switch to enable said savings.


Related Articles