Ampere Laptop GPU Overclocking Guide 101
What I´m aiming for is finding the maximum frequency for every voltage which is in the range of being hit on loads.
My most chill way to test for stability was with using 3DMark´s Stress test, FireStrike and TimeSpy (2 displays are needed – One Display with stresstest open and the other one with MSI AB open)
ALT Tabbing
3DMark benches do stop when you alt tab out. Other games, benches or alike will go down with fps and load when being in background, Hence avoid alt tabbing during benchmarks
Consequences of ALT Tabbing
Consequent Alt TAB will make the GPU switch over to a higher frequency when being on load after a short time.
So with non constant load and consistent Alt TAB might need to wait this out every time depending on how fast you are.
How does this exactly behave?
You have applied a certain curve . After some time on load (sometimes seconds on me sometimes close to one minute) this whole curve will be shifted up by 30Mhz.
This means when you are testing clocks on load and type the exact same clocks later in the curve when being in idle the load, clocks will be 30Mhz higher leading to instabilities
Frequency Clocks
The frequency steps are a bit annoying too. It´s mostly 7. Sometimes 8 in mine. The load curve shift is mostly 30. Sometimes 29.
Because of this you have to find out the exact clocks you are using in the curve later on. If you type another number it is gonna be changed to a number available from the steps.
This might lead to instabilities too if not taken care of. You can read out these numbers easily thru OSD. Just write them all down so you know each step you went across.
Actual testing
- Open MSI AB.
- Shift click the whole curve as far you can downwards with beeing able to still see the first dot in the line.
- Click the first frequency dot on 0,7V and pull it up to 1400MHz. This should be stable to start with and make the curve flat at 1400 MHz over the whole graph.
Thus the GPU will be picking only the first voltage from where the frequency isn’t climbing anymore. - Apply
- Open Stress test (Explained previously)
- Let it run until curve shifts 30 MHz up, Should stay at 1430 MHz now.
- Now increase the Curve by 7 MHz. Wait one or two cycles of the test. Increase by 7 MHz again.
- Repeat until you get crashes/artefacts. Artefacts can be identified by „green squares“ „yellow arcs“ „white circles“ and so on and so on.
- Due to artefacts go 7 MHz back. Wait 3-5 cycles for crashes/artefacts (If you get a driver crash: try to click in the now black stress test windows FAST and click „ESC“.
This should close the stress test after a maximum of 1 minute and you can restart it thru 3DMark and keep on tweaking. If you cannot get 3DMark to react you have to force close every component o fit thru the task manager and open again. - The GPU using small steps makes it very pleasant to tweak (at least for me) because the number of full driver crashes were at around 4 or 5 in a 3 hour session. Artefacts just need fast clock adjustment and you can keep it running.
- Now when you didn’t find artefacts or crashes after 3-5 cycles write down the achieved clock with the used voltage. I would recommend going at least 3 steps (21 MHz) below the short-term stability you just tested in your final curve to have a pretty stable OC.
Now repeat with the next voltage and so on and so on.
Follow the above mentioned steps until you reach around 0,85V. Because at this voltage you will start to get tdp throttling in certain loads.
Further stability testing is really time consuming and should be happening over time when gaming. Because you will hit tdp throttling very often at above 0,85V the voltage will fluctuate by a lot.
Most modern games will make very good use of 0,85V maxing out closely around it and you will already have close to max performance.
Some example values to help (This was performed on a 3060) (3070 and 3080 varies)
— Tidles0