(this second part of Building a Low-power Ryzen HTPC series focuses on underclocking, the first part is here)
I finally managed to get wattage readings from my Ryzen HTPC, and spend some time figuring out the best way to underclock it. I used PeakTech 9035 to measure power draw. This power meter also keeps note of the highest observed power value, which comes in handy when doing stress tests.
My objectives are:
Minimize power draw when idling, which is what the HTPC will be doing most of the time. It makes the most sense to optimize for this first.
Maximize performance when needed. I will be using the HTPC as my development machine. When I’m doing a lot of compiling, I am more than willing to trade off power efficiency for performance and save some human time instead.
I also need to make sure I stay within the following constraints:
Max CPU core temp must be lower than 90° C (194° F) at all times. The case is very space-constrained, with the only fan being the low-profile fan on the CPU. With limited air flow within the case, cooling performance of the fan is questionable. I want to make sure that I’m not frying the CPU during sustained load.
Max power draw must not go over 80% of PSU declared power. The PSU I got has max declared power at 85W. I don’t know how much power different components draw, so I want to leave some headroom for intermittent power spikes. With declared efficiency rating of 87%, the highest power draw I should be able to read at the mains is
(0.8 * 85W) / 0.87 = cca. 78W.
For load testing, I used a 30-minute long
stress-ng load test that combined
CPU hogs with I/O workers:
stress-ng --cpu 6 --io 4 --hdd 2 --vm 2 --vm-bytes 128M --fork 4 --timeout 1800
I will take note of maximum CPU core temperature reached, maximum CPU frequency observed, maximum power draw at the mains, and maximum PPT (AMD Package Power Tracking) value, which roughly translates to how much power the CPU is drawing1.
Before I list the values, I have to give a shoutout to this great article on Ryzen CPU undervolting. I took the advice on PPT tweaking and it worked great.
Baseline: No underclocking
no underclocking ================ idle power draw (mains/PPT): 15.2 W / 4.4 W max power draw (mains/PPT): 68.7 W / 36.8 W max CPU temp: 95 C max CPU freq: 3952 Mhz
I was pretty impressed that idle power draw with no underclocking at all is as good as idle power draw when CPU was severely underclocked. AMD power management is pretty amazing. Additionally, PPT max value is very close to declared TDP of 35W for my Ryzen CPU.
However, CPU temperature is unacceptably high during stress testing. Max power draw was also uncomfortably close to my acceptable limit.
Take one: Undervolting and frequency limiting
2000 Mhz max freq, 0.95 VID =========================== idle power draw (mains/PPT): 15.5 W / 4.2 W max power draw (mains/PPT): 33.3 W / 13.8 W max CPU temp: 60 C max CPU freq: 2000 Mhz
This was my setting from the time I first set the HTPC up. It is very conservative, and I knew up front that this configuration prioritizes efficiency over performance. Both power draw and CPU temperature fall well within acceptable limits, but performance is very hindered. It still works for Kodi and editing, but falls short when it comes to compiling.
Take two: PPT 20 W
PPT 20W ======= idle power draw (mains/PPT): 16.4 W / 4.2 W max power draw (mains/PPT): 41.4 W / 20.0 W max CPU temp: 76 C max CPU freq: 3283 Mhz
The next thing I tried is to adjust Package Power Tracking (PPT), as suggested in the article I mentioned above, and it worked out pretty well! Both power draw and CPU temperature are fine, yet I was able to see CPU peaking at ~3.3 Ghz which made a big difference in performance.
At this point, I became a bit greedy, and wanted to see if I could squeeze a bit more out of this.
Final take: PPT 25W
PPT 25W ======= idle power draw (mains/PPT): 16.9 W / 4.6 W max power draw (mains/PPT): 48.0 W / 25.0 W max CPU temp: 87 C max CPU freq: 3604 Mhz
This is my final underclocking configuration. To make sure CPU temperature during load is stable, I ran the load test again for 1 hour, and the temperature kept steady. Besides this, power draw during load testing remained pretty efficient as well. Adjusting PPT for +5W resulted in about ~10% time savings when doing intensive cpp compilation, when compared to PPT = 20W.
This tiny PC is remarkably efficient. Even if it was running at full load during the entire month, it would contribute ~1 EUR to my electricity bill.
Ryzen CPUs have great power management. While idling, cores were spending over 98% time in C6 sleep state. Power draw when idling was pretty much the same across all underclocking configurations, including baseline.
Adjusting PPT is an amazing way to rein in CPU power consumption and get the most performance out of it.
I was able to read package power thanks to this awesome Ryzen SMU kernel driver provided by