RIP-Felix
Senior Member
Just to get back to the diminishing returns that Mitsu has been talking about. I showed this graph in the video, but you can see that each increase is having less and less effect. I haven't quantified how much yet. Like is 750 core 90% of the FPS gain's as 950, but you can see that even a modest OC of 700 is a very nice improvement.
Here are the stats for power consumption as frequencies increased. In each of these tests I held either the VRAM or CORE at one frequency while overclocking the other, to gauge the effect of just that frequency, not both of them together. As you can see, the core frequency increases power consumption the most. Which makes sense, it increases the duty cycle of the processor. More frames are being processed and at a faster rate. That translates to work being done. Work is a measure of power. Power is Watts.
The last 3 data points were for voltage increases. That's why the power jumps up at a new rate. VRAM increases barely increased the power. I see them as basically free performance.
And of course all this extra work and power has to go somewhere. It is converted to heat, which is basically a mirror of the power graphs. They show how closely frequency is tied to power and heat. And that increasing voltage quickly increases both.
EDIT:
Something I didn't test or mention in the video is that I suspect the VRAM temps plateaued. I don't think that means they reach a certain point where the performance can increase, power can increase, work and thus heat should increase, but don't. Instead what I suspect is happening is that the thermal adhesive is being overwhelmed and there is a thermal bottleneck preventing the heat from escaping into the IHS and transfer to the die where it's would be registerd by the thermal diode inside. So it looks like a plateau.
I think that delidding and cooling them more efficiently might yield a higher VRAM OC and stability. The reason I don't like recommending this is that the thermal adhesive provides structural support that improves reliability of the solder joints and TIM bond-line on the die. It stiffens the processor package preventing warping stresses during thermo-cycling, increasing the mean time before failure.
The only way I can think to replace the adhesive with a more conductive material is to get a 0.2mm high w/mK thermal pad and use a hole punch in the center to act as a reservoir for the thermal adhesive (Stars 922). The adhesive will solidify and provide the structural support, but the thermal pad around it will still conduct some of the heat away.
I haven't tried it yet. Super hypothetical ATM. It's probably not going to help and be a waste of time. So I didn't think it was worthy of mentioning in the video. That and it takes a decent amount of time to explain (as the wall of text above proves).
The last 3 data points were for voltage increases. That's why the power jumps up at a new rate. VRAM increases barely increased the power. I see them as basically free performance.
EDIT:
Something I didn't test or mention in the video is that I suspect the VRAM temps plateaued. I don't think that means they reach a certain point where the performance can increase, power can increase, work and thus heat should increase, but don't. Instead what I suspect is happening is that the thermal adhesive is being overwhelmed and there is a thermal bottleneck preventing the heat from escaping into the IHS and transfer to the die where it's would be registerd by the thermal diode inside. So it looks like a plateau.
I think that delidding and cooling them more efficiently might yield a higher VRAM OC and stability. The reason I don't like recommending this is that the thermal adhesive provides structural support that improves reliability of the solder joints and TIM bond-line on the die. It stiffens the processor package preventing warping stresses during thermo-cycling, increasing the mean time before failure.
The only way I can think to replace the adhesive with a more conductive material is to get a 0.2mm high w/mK thermal pad and use a hole punch in the center to act as a reservoir for the thermal adhesive (Stars 922). The adhesive will solidify and provide the structural support, but the thermal pad around it will still conduct some of the heat away.
I haven't tried it yet. Super hypothetical ATM. It's probably not going to help and be a waste of time. So I didn't think it was worthy of mentioning in the video. That and it takes a decent amount of time to explain (as the wall of text above proves).
Last edited:
((
