As I eradicated the cause of stuttering and frame skipping in all games, and also accidentally lowered the temperature by 15 degrees

I will be brief, the reason was the inconstancy of the frequency of the GPU (more) and CPU (less). Using MSI Afterburner, HWiNFO64, and RivaTuner experimentally, I found out that frame drops are due to the powerlimit in gpu and pl4 limit in cpu. Limits are and always will be. Even if we remove the limits for power and current, we rest on the temperature limit, which is even more blatant than the previous ones, because the frequency changes with sharper jumps. But there is a solution. I ask under the cat.

Shuttering of this type is found even on the most powerful computers from 2080ti, 1080ti, and especially in sli configurations.

I am using the GTX GTX1070 GPU, CPU 6700K, 1080p @ 120hz screen, p750dm2 clevo (but most likely, everything mentioned below will also apply to desktops).

My laptop device doesn’t turn to call the language because of the desktop processor and the corresponding cooling.

Just lowering the frequency to somewhere 1-5% from jumping 1750-1680Mhz to 1671Mhz stable in time using the curve function in Afterburner, stuttering almost completely disappeared.
And after turning off the turbo boost at 6700k (and fixing the frequency at 4Ghz), it disappeared completely. At the same time, the temperature of the GPU fell from 88 to 73 degrees.
Next, I conducted testing with the processor at the same frequency, 4Ghz. The effect is even stronger if you take into account the effect of disabling turbo boost.

Testing conducted in the game Rise of Tomb Raider mk it has a built-in benchmark.
Before:


After:


Benchmark result:
Before:


After:


Graphics settings are average:




And most importantly, the curve for my video card, with overclocking +148:


I activated the OSD in the MSI Afterburner Core clock, CPU clock, Power limit settings, and through the plug-in to HWiNFO64 I used the throttling CPU indicadors:


Just playing for about 15 minutes, determined the minimum CPU and GPU frequencies after a drawdown due to heating, and set the horizontal line 10 Mhz lower. Power limit (and other limits) should always be either active or not, and not rattling every few seconds.

A small note, many monitoring tools, the same task manager in Windows, and especially HWiNFO64 are very clearly visible on the graphs, so be careful, frame drops at regular intervals are the norm with these programs during testing. I recommend setting the update interval of the charts every 5 seconds.

The result is a drop in performance of only 0.14% (calculated from the total number of benchmark frames), an increase in average fps (top 95%), and a decrease in the temperatures of the GPU and CPU by 15 degrees when measured on the GPU (they have a common radiator) instead of smooth gameplay.

Results:
CPU full power, full power GPU - 88 C, 72 fps
CPU low power, full power GPU - 80 C, 72 fps
CPU low power, GPU low power - 73 C, 71 fps
In more detail, I unfortunately do not have time to test.

PS:
About the CPU. It is necessary not to turn off the turbo boost roughly, but wisely. Before switching to 4.0Ghz, I tried 4.4Ghz (overclock), 4.2Ghz, and 4.1Ghz. Without exception, they are not stable in time, and produce drawdowns to 4.0Ghz. You may have a different situation. It also depends on the game, so turning off the turbo boost as a universal solution is not recommended, rather a manual selection of the maximum frequency stable in time.

I was inspired to write this article - habr.com/en/post/419043 .

If you notice errors, please write to the PM)
If you like the article, I will translate it into english.