Trinity turbo clock based more on temps than load?

Apr 20, 2008
10,067
990
126
Hello everyone,

After changing out the CPU fan on the laptop in sig, I've noticed that performance is higher than it was. I know with Fritz the CPU would lock onto the base 1.9Ghz clock with all four cores loaded in everything, especially Fritz. Now that I replaced the seized CPU fan, the CPU locks itself to 2.2-2.5Ghz at full load, a whole 300-600mhz higher.

3BmSrKt.png


At load of two threads it is now hitting the full 2.8Ghz whereas before it was 2.1-2.5ghz.

It was my understanding that at full CPU load it would only be the base clock of 1.9Ghz. If temps greatly affect CPU performance like I can see, would getting a small cooling pad/fan increase performance even more? I bought the laptop used so I don't know what it was like brand new.

Edit: Also, I don't fully understand thermal margin. Coretemp/HWinfo and all of the non-AMD temp programs read this laptop incorrectly (120C, for example) Is it temp difference from a determined safe zone? When I load all four cores AMD overdrive reports that the GPU portion tops out at 56C.
 
Last edited:

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
A8-4500M has 2.8GHz 1CU turbo and 2.3GHz 2CU turbo.

If AOD shows thermal margin of -3.5°C, your CPU is already overheating (103.5°C).
 
Apr 20, 2008
10,067
990
126
A8-4500M has 2.8GHz 1CU turbo and 2.3GHz 2CU turbo.

If AOD shows thermal margin of -3.5°C, your CPU is already overheating (103.5°C).

I thought the max temp for FX CPUs was 70C and the thermal margin was set at 60C? I know Intel has a max temp of 100C for mobile CPUs, but AMD does not. FX shuts down at 95C IIRC.

I just replaced the fan, and the new one spins much faster and quieter. I also applied new thermal paste (Arctic Silver Ceramique 2) which doesn't have a burn-in period. If it was overheating I would think the keyboard where it's at would be hot as hell. Right below the ~ and F1 key is where the HSF is.
 
Last edited:

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
A8-4500M is a mobile part and it has tCTLMax of 100. Desktop parts have tCTLMax of 70.

Thermal margin = tCTLMax - tCTL.
 
Apr 20, 2008
10,067
990
126
I'm having a very hard time TM is 100C. When I have a 100% load on the CPU with Fritz, the GPU shows 57-60C. When I have 100% GPU load with Unigine, the Thermal Margin get to -1 and -2 on the CPU, but show 61C on the GPU. When I load both CPU and GPU at the same time, I get 59-63C on the GPU and and 1 to -3 on the thermal margin.

v6xmOtJ.png


6UjduRZ.png


If it was 100C on the thermal margin calculation, that would be a 40 degree Celsius spread from the CPU to the GPU, or 104F difference. That doesn't seem possible. The thermal margin has to be set to 60C as it mirrors GPU load temps the closer it gets to 60C. If the CPU was really 100C it would be torching the GPU side too. You can test this with any MCM CPU setup, very similar to APUs. When a load is put on two cores, the other two cores ramp up in temperate as well (but not fully).
 
Last edited:

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
And what kind of temperatures does the GPU read when the system is idle? Close to zero or even negative?

The stock tCTLMax for this APU is 100, period.

Use HWInfo to monitor the current tCTL value and ditch the garbage AOD.
 
Apr 20, 2008
10,067
990
126
Idle temp on the GPU is 39C, as shown in screenshot.

Sjy0GZl.png


It appears as if thermal margin might be 70C on the APU even though the max temp is 100C. Whenever I add the celcius reading from HWinfo and add it up with AOD, it's 69.9C.

FwjNEXi.png
 

Erenhardt

Diamond Member
Dec 1, 2012
3,251
105
101
Thermal margin is nice. New stuff hit power margin which is infuriating. Hacking your way though it is pain in the rear. But if you do, it works nicely, which means the power margin is too conservative (at least on some boards).