Worth it to replace Mid-Range GPU Thermal Compound? My experience (with pics)

[dragonhart6505]

I've already taken the leap to do this, so here's my experience so far. I have an XFX 4gb RX 460 Double Dissipation. I should mention that I have applied the "Unlocked Shaders" bios modification to my card, so temperature readings as described below are a few degrees higher than with the stock RX 460 unmodded. Thought I'd clear that up beforehand.

In a recent thread I posted about benching said RX 460 (my very first post here) I was introduced to HWbot.com, a community of competitive and enthusiastic overclockers. TL;DR, the thread quickly got out of hand between me and one other member of the LTT/HWBot community about going fast and pushing my CPU (Phenom II x4 955) to its limit. I fiddled around and got a max clock up to 4.114ghz that I recorded to CPUZ validator. Cool stuff!

During the process of trying to OC my CPU to the max, I screwed up and realized I don't have any thermal paste to replace on my heatsink. So I went out and got the best I could find in my immediate area, some Arctic Silver 5. Applied, validated my CPU clocks...good to go. Then I got to thinking...what would it do for my RX 460?

 

 

Last night I wasted no time getting temperature readings of it at stock and OC'd before tearing into my GPU. I used 3DMark Fire Strikes' introductory scene just to get the GPU ramped up and hot, so I don't have any scores to speak of yet.

So stock temps at idle were anywhere between 51c-56c with the default fan curve, which Radeon Crimson leaves off until it thinks they're needed. This was the same case with my OC applied, more or less (a recent update to MSIAB disabled sticking the GPU clock maxed out on unofficial overclock).

 

 

On load the fans ran at 1700rpm keeping the temp at 67-68c.

 

 

With my OC, the fans ran around 2200rpm and temps hit 71-72. There was also quite a bit of artifacting during the bench. I am aware that my overclock was ambitious, to say the least, and I didn't need NEARLY this high of a voltage bump. I didn't realize my mistake until after the fact...we'll get to that later.

 

 

So let's take it apart:

 

It was really easy to tear down. 4 screws on the back of the PCB holding the block down

 

 

and 2 on the rear of the support shroud.

 

 

Here's what XFXs' application looked like

 

 

and here is mine

 

 

Close it all back up and slap it back in.

 

Results? Well, this is the tricky part to me. I ended up wasting another couple applications of my AS5 because turns out that wasn't quite enough paste. Temps were actually HIGHER at idle and it became even more unstable than before during OC. So I tore it down 3 more times and applied paste differently each time. Second application was spread, third application was a dot, and final application was a larger dot. I compared temps after each on load and at idle with the same settings and results were minimally different from each other and the initial baseline test. However my final "bigger dot" method seemed to do a bit better than the rest. I have come to terms that the fan curve provided by Radeon Crimson is crap. You need the fans to spin at least a LITTLE bit to get some kind of cooling result, otherwise the GPU die is relying solely on the heatsink and that alone is only going to do so much.

 

Currently with 25% fan speed, idle temps rest at about 36c-40c. Cooling while on load is where the real result comes in. I can actually keep the GPU cool enough below 65c while overclocked on load at 40% fan speed, or 1800rpm, to finish the same Fire Strike test, this time with more reasonable settings.

 

So back to the point I added earlier about my OC settings. Turns out at 1340mhz/1920mhz I really only need to bump the voltage up to +12 and remain stable enough with minimal artifacting, though that's still not ideal. I don't plan to keep the clocks that high for any other purpose other than benchmarking and even then it's giving diminishing returns in terms of actual performance gain. This is why you test and learn, people .

 

I think that just about covers my experience. I have found a happy medium in temps and maximum performance gain out of my card at a more reasonable 1330mhz/1920mhz and +6 on voltage to get about 62c with no artifacting throughout the Fire Strike benchmark. I do also know that the Arctic Silver 5 needs some time to cure, so my temperatures may drop a degree or two more when it finally sets. Here's hoping!

 

 

I hope you enjoyed this little experiment of mine. Do you think it was worth it to do this on a budget card? Please leave your thoughts down below. I'd love to hear your own experience and whether or not you might try this after seeing these results.