Defeating thermal paste pump-out in high power GPUs

[yoadknux]

Good news, PTM7950 arrived today. I thought it would take longer due to the Iranian attack last week, glad to hear I was wrong.

So, first and foremost. When opening the card I see the exact same pattern in the center as before:



Anyway, I cleaned up the paste and put the PTM7950 on it.



The application was kinda annoying. You need to peel off two sides. The bottom left and right broke a bit, but it's no big deal I think.

As for temperatures, I read that it needs time to "cook". So I'll run a light load on it for a while and then run the stress test and update with the results.

EDIT
After letting it "cook", meaning doing several hours of gaming with load-on / load-off regularly, here's what I'm getting with the benchmark (Cold start, TimeSpy Extreme, Power Limit set to 600W):


In terms of "Day-0" temps, it seems to slightly outperform the CM MasterGel Maker by about 2c, and match the Kryonaut Extreme. That's pretty impressive considering its not an actual paste.
Will it pump out? I'll have to wait and see how it holds up over time. A week for now should be a good time to repeat this test.

 

[yoadknux]

I'll post some overclock results with PTM7950.

The card manages to pass Superposition at a core of 3060MHz (+270MHz on MSI AB). Memory is a bit on the disappointing side at +1100MHz. Anything more and I start seeing artifacts.

Temps were really low at 52c core and 59c hotspot. That's "cheating" though as I had the fans at 100% (sounds like a jet engine). At regular profile it's about +10c and lower clock because of thermal throttle.

 

[yoadknux]

PTM7950, after one week:



Completely identical to the result from 7 days ago (top post on this page). So far, no sign of ANY degradation!

 

[yoadknux]

Day 14, TimeSpy Extreme benchmark (same as posts #41 + #43), 600W PL, identical temps. No paste was able to achieve such consistency over time for me, only PTM7950. Magic. Will keep updating weekly until the end of the month.

 

[yoadknux]

Day 21, same benchmark, same power limit, same temps.