NOVA - Optimizing perf, noise and temperature

[FerrariVie]

I finally got the time to make note of Firestrike measurements when limiting GPU clocks:
View attachment 23825

I know this is not scientifically ideal since I've run each max clock just once, but it's kinda time-consuming to do them all 5 times or so an then take averages, and there's also the probability that running 35 tests non-stop will make components hotter and bring results down with time. But at least you can compare your results with mine

Took advantage of the colder weather this week to run the benchmarks again after repasting, as it was easier to manage the room temperature. The ambient was still 1°C above the temperature of the stock paste test, but since we're on the Spring already I'm sure this is going to be the closest that I can get to 20° in my room for a few months.

Before:

Now:

In summary, GPU temperatures in Firestrike dropped 1°C and its scores were improved at the same time, so it was worth the €15 investment and around 3 hours of my time for the whole process (yes, I'm slow but was doing things with attention to detail and taking pictures at the same time). Regarding CPU temperatures, I also monitored them using Hwinfo (max and avg of each run) and they're about the same as I had before repasting, so whatever paste PCS used on my CPU was indeed well applied and of good quality.

 

[maxff9]

Thanks for the good job (you do nice plots on Excel he !).
Indeed the repasting doesn't drastically change the thermals but that's a good news for PCS work.

On my side, I have received the paste but will wait a few months as I want to trust the good work of PCS. And also I don't have much time during this period (maybe I'll never have time again ?). I'll investigate more this topic soon I hope.
And I'll definitely use your recommendations for the repasting.

I played Outer Wilds (seriously, that's a 10/10 game for me, you should buy it) quite a lot recently and I noticed some FPS drops once. I don't know why, nothing change on the PC except some background processes/updates maybe. Only happened once. I should monitor all my gaming sessions, it's really just one click. I always play with my headset so the noise isn't TOO annoying but if that's the price to pay to have good perf, well that's fine.

Always playing with a max clock of 1755 MHz.
Nothing on the software side to cool down the CPU ? Except increasing the fan speed I mean.

 

[FerrariVie]

Thanks for the good job (you do nice plots on Excel he !).
Indeed the repasting doesn't drastically change the thermals but that's a good news for PCS work.

On my side, I have received the paste but will wait a few months as I want to trust the good work of PCS. And also I don't have much time during this period (maybe I'll never have time again ?). I'll investigate more this topic soon I hope.
And I'll definitely use your recommendations for the repasting.

I played Outer Wilds (seriously, that's a 10/10 game for me, you should buy it) quite a lot recently and I noticed some FPS drops once. I don't know why, nothing change on the PC except some background processes/updates maybe. Only happened once. I should monitor all my gaming sessions, it's really just one click. I always play with my headset so the noise isn't TOO annoying but if that's the price to pay to have good perf, well that's fine.

Always playing with a max clock of 1755 MHz.
Nothing on the software side to cool down the CPU ? Except increasing the fan speed I mean.

Hahaha, thanks for the compliment of my excel graphs . I've never heard of Outer Wilds, I'll have a look at some videos about it

Regarding the CPU... undervolting does help, but I'm not sure how Control Center will behave when you start changing voltage and clocks outside of it. Last time I tried, Control Center always changed the voltage back to its profile settings after a few minutes, so I got rid of CC and I'm using third party software to control fans and keyboard LED. But Clevo has released a few updates for the CC that might have helped with that issue.

If you want to try undervolting while still using CC to see if it sticks, the first step would be to have both CC and Ryzen Master updated (both should have been installed by PCS by default, but you can download the latest versions respectively from Clevo and AMD directly). Go into Ryzen Master and note the current voltages that you're getting on the current CC profile that you're on. The voltages will go up and down depending on CPU load, but it's good to have an idea about how high it is going when under heavy and low load.



You could also rely on Hwinfo (Core VID data point) to easily get the max voltage when gaming, but I'm not 100% sure how accurate that is. From the few tests that I've done here while writing this post, it's been matching Ryzen Master on low loads.

After getting an idea about the highest voltages that your CPU is getting at stock (or CCC's active profile), you can start tweaking that down on the manual overclock in Ryzen Master. You should set the clocks to your highest stock boost and start lowering voltages a bit every time. My CPU at stock would go all the way to 1.25V on max load (e.g. running cinebench), so I turned it down as much as I could while keeping it stable. My current settings are:


This means that the maximum voltage that it will go to is the one that I'm setting on the manual profile, in my case 1.00625V. If I go one step lower (1V) it's not 100% stable and the PC will restart on some specific high usages. Cinebench is one indicator of stability (if you can run it till the end without crashing, it's a good sign), but my personal experience says that games are even better indicators for that same purpose. The one that worked best for me was GTA V, as I've noticed that some other games would be stable while GTA was crashing. 1V doesn't crash Cinebench for me @ 3.9Ghz, but it does crash on GTA and other CPU heavy games.

With that high undervolt, I can use my CPU with stock boost clocks even when doing everyday light loads, while still keeping fans quiet. They just ramp up a little bit for about 5 seconds when opening Chrome from scratch. Keep in mind that the lower you set your max voltages, the lower your temperatures will be, but you'll also lose a bit of performance. Differently than with the GPU curve, in here we're keeping the stock clocks and just lowering the voltage (undervolt with no clock change), so the impact on performance is not that huge. Sometimes, when the person is having issues with CPU temps, using a lower max voltage might even mean better performance, but only testing and benchmarks will tell what's the best voltage for you.