Noctua NH-U9S Performance For The AMD Ryzen 9 3950X + Ondemand vs. Performance Governors
For those that may be looking to run an air-cooled AMD Ryzen 9 3950X especially in a rack-mount 4U chassis, here are some recent results I did from some testing using a Noctua NH-U9S with two 92mm fan configuration. Additionally, these results contain performance metrics from both CPUFreq Ondemand vs. Performance governors as an additional point of interest.
These results are for reference purposes of the Noctua NH-U9S in a dual fan setup for this 16-core / 32-thread 3.5GHz (4.7GHz boost) CPU rated with a 105 Watt TDP.
The NH-U9S was tested as one of our go-to heatsinks given that it fits within 4U height requirements that we are often running within for our many rackmount benchmarking systems.
The Ryzen 9 3950X testing was done on Ubuntu 20.04 daily with its default Linux 5.4 kernel at the time. Tests were done out-of-the-box with the CPUFreq Ondemand governor and then when switching over to CPUFreq Performance governor.
On the thermal front, ondemand vs. performance governors for the Ryzen 9 3950X didn't yield much of a difference, similar to what we have seen in the past -- contrary to popular belief that keeping "performance" on will lead to much higher operating temperatures normally isn't the case for at least desktops. Under hours of load, the average temperature was 62 degrees with a peak of 84 degrees. The idle temperature was around 32 degrees.
As for the performance between governors, librsvg and dav1d had the most profound benefit from the performance governor while GIMP, vpxenc, DeepSpeech, and others saw less of a benefit. For the web browsing performance, using the CPUFreq performance governor only led to ~3% better performance. In numerous workloads, there isn't a measurable difference between ondemand and performance governors. Those wanting to dig through all these cooling and governor performance tests conducted on the Ryzen 9 3950X, you can find all the data via this OpenBenchmarking.org result file.
These results are for reference purposes of the Noctua NH-U9S in a dual fan setup for this 16-core / 32-thread 3.5GHz (4.7GHz boost) CPU rated with a 105 Watt TDP.
The NH-U9S was tested as one of our go-to heatsinks given that it fits within 4U height requirements that we are often running within for our many rackmount benchmarking systems.
The Ryzen 9 3950X testing was done on Ubuntu 20.04 daily with its default Linux 5.4 kernel at the time. Tests were done out-of-the-box with the CPUFreq Ondemand governor and then when switching over to CPUFreq Performance governor.
On the thermal front, ondemand vs. performance governors for the Ryzen 9 3950X didn't yield much of a difference, similar to what we have seen in the past -- contrary to popular belief that keeping "performance" on will lead to much higher operating temperatures normally isn't the case for at least desktops. Under hours of load, the average temperature was 62 degrees with a peak of 84 degrees. The idle temperature was around 32 degrees.
As for the performance between governors, librsvg and dav1d had the most profound benefit from the performance governor while GIMP, vpxenc, DeepSpeech, and others saw less of a benefit. For the web browsing performance, using the CPUFreq performance governor only led to ~3% better performance. In numerous workloads, there isn't a measurable difference between ondemand and performance governors. Those wanting to dig through all these cooling and governor performance tests conducted on the Ryzen 9 3950X, you can find all the data via this OpenBenchmarking.org result file.
17 Comments