ROG GL702ZC fluctuating GPU frequencies

[deksman2]

My GPU apparently cannot reach its full potential for an unknown reason due to clocks fluctuating wildly (though in all honesty, I have no idea if this is hardware or software related).

Running Furmark GPU stress test for a limited period of time, I've noticed that the GPU cannot sustain maximum clocks at all.
Its clocks wildly fluctuate from minimum to medium to maximum (you can see screenshot of running Furmark benchmark how the GPU frequency graph is fluctuating and an example of how low the frequencies are at a point during benchmarking).
This is NOT normal behaviour and it explains why Darksiders III was performing poorly too (same fluctuating clocks throughout the game).

I don't understand how/why is this happening.
Even before the second RMA, the GPU was displaying similar behaviour in Furmark (fluctuating clocks, but they were fluctuating from middle to high as opposed to minimum to high) but only after it went past certain temperature levels (above 73 degrees C) - and upon closer examination same is happening now, only its more pronounced.
And as you can see, the GPU is nowhere near what I would call 'dangerous' temperature levels, and this GPU clock fluctuation is happening when the GPU is stressed).

Something seems to be preventing the GPU from leaving its minimum state and maintaining its boost clocks on a consistent basis... I tried other drivers (with same results).
I tried turning off Power efficiency and Chill off in Radeon drivers... same thing is happening.
I tried even increasing power limit to 50% with no difference.
I tried undervolting, which lowered the temperatures to around 73-74 degrees C but clocks were still fluctuating like mad (so, no difference in actual performance).

The GPU might be throttling... or is it?
I don't know.
Could it be a Windows issue?
I hadn't tried reinstalling the OS... so that might also be contributing to or causing the problem (but as you may know, reinstalling the OS is a last resort as its a time consuming process - I can do it of course, but I need help in trying to narrow down what's causing the problem).

EDIT: Someone on AMD forums had a similar problem with their desktop GPU and the only way they solved the problem was to open the GPU and replace the thermal compound (and of course, Asus does NOT allow people to remove the cooling assembly and replace the thermal comound themselves).
The person on AMD forums said their GPU was overheating and that the displayed temperature was wrong.
My issues begin few seconds after running FurMark GPU stress test... which I don't know if its enough time to cause overheating.

[JustinThyme]

Do you have game center installed and clocks set accordingly?

[deksman2]

JustinThyme wrote:
Do you have game center installed and clocks set accordingly?

I have Asus Rog Gaming centre installed, but I cannot affect the clocks through that (and it shouldn't matter because ROG gaming centre is not responsible for that).
My GPU clocks are basically set to default settings, and they go up or down depending on the load... my problem is that the GPU cannot reach or maintain its maximum boost clocks when its stressed.

[deksman2]

Note: Star Trek Online works fine.
Luxmark benchmark also works fine (gpu clocks are maintained at maximum levels)... but Darksiders III just refuses to work properly.
I also ran Superposition benchmark as well to see if I can replicate the problem, and the clocks fluctuate there as well.

[jels]

I suspect its power limiting to lower power state clocks open up gpu z and check perfcap reason you could also open up afterburner press ctrl +f to see current frequency curve. Running a windowed benchmark while curve is open will allow you to see whats going on, also make sure you got fans on overboost as fan profile also seems to effect gpu frequency regardless of temp.

[deksman2]

jels wrote:
I suspect its power limiting to lower power state clocks open up gpu z and check perfcap reason you could also open up afterburner press ctrl +f to see current frequency curve. Running a windowed benchmark while curve is open will allow you to see whats going on, also make sure you got fans on overboost as fan profile also seems to effect gpu frequency regardless of temp.

I seem to have discovered a problem here.
The fans don't seem to kick high enough to cool the GPU properly... but the temperatures do not exceed 75 degrees C... which is odd because this unit shouldn't produce oscillating frequencies or reduced performance at those temperatures...

Under auto-fan control in ROG gaming centre, I get 2500 score in Superposition benchmark with 1080p High settings, whereas I should get about 4960 (or close to 5000).
The fans didn't ramp up under Superposition benchmark and 'auto-fan' profile...
When I set the fans to 80% manually, I score about 4600 points in Superposition benchmark.
With a core undervolt set to -93mV, +50% power limit and fans set manually to 80%, I get 4862.

This indicates something is wrong with either the fan control regulation, the GPU might be defective, or the cooling assembly might not be touching properly on the VRM's or GPU (the laptop underwent 2 RMA's by now).

So, for my laptop to need these kinds of 'mods' to achieve 'stock' levels of GPU performance is absurd.
It's going back.
I'm sending a signed for letter to LAptopsDirect and will ask for a full refund under 2015 consumer protection laws as this unit was fraught with problems and STILL doesn't perform as it should.