Cooling & VRM's

[ShadowXP]

So here I sit, with an XFX HD6870 that I swapped out the cooling on. Went from the stock cooler (noisy and not all that great), to an Arctic Cooling Accelero Extreme with the VR001 heatsink-set (quiet and very good). All fine and dandy...until I plugged it into the computer.

There seems to be some sort of short on the card, because the comp won't even turn itself on when both power-cables are installed. And as you might know, running the card on one rail = no post, no image etc. Having tried most things, and already having written the card up as a loss due to my own stupidity, I decided to try something today, mainly to see if I could get the card running again in some form or function. First thing I checked was taking off the heatsink/fan assembly to see if I could find any sort of damage to the card hidden by it. Nothing found. Strangely enough, though, is that the testbox would fire up when the GPU heatsink/fan-assembly was NOT on the card. Still no image, but it looked like the rig did go through it's POST-sequence.

The heatsinks for the RAM and VRM's are all glued on with thermal adhesive, but I noticed that the short and little spark came from the area where the VRM's are mounted (yep, there was a little spark ). Thus I figured that it was a good idea to remove the heatsinks to see if they caused the short between components. Not really an issue, since applying a little bit of pressure to the sinks made them pop right off. Here I made three discoveries:

1: The thermal glue, which I thought was a solid, hard mass, is actually more akin to the silicon sealant used by plumbers (I know, not the same at all )

2: Between the VRM's, there's signs of damage on two small components

3: on the underside of the heatsinks, there's marks that looks like small scorchmarks in the aluminium.


Now then! To questions:

- Just how important is the cooling of the VRM's?
- Could it be that the components that looks damaged has been the cause of the short? And if so, can they be replaced? See attached screenshot for a (poor) image of the component.

 

[pantherx12]

Cooling the VRMS is essential, may I ask when you installed the non stock cooler did you remove the stock ram/vrm cooling plate?

If yes, why XD it works nicely and supports the card.


If you installed your own heatsinks to the card vrm area there is a chance you caused the short yourself, some of the componants by that area are a little taller than the vrms themselves, so using anything wider than the stock heatsink could cause a short.

As for the componants, they are replacable : ]

Should be able to nick them off another gpu.

 

[AsRock]

Don't use those heatsinks you get with them coolers as your much better of using the stock plate on the card.

EDIT: As Panther said you have to cool them VRMS they get hotter than the GPU most of the time.

 

[ShadowXP]

On the XFX-card I have, the main cooling-plate covers both RAM and VRM's (see extremely poor pic). I also now see clearly where I messed up, as on the stock cooler, there's notches to make sure that there's no contact between the heatsink and the components. The VR001-set from AC I installed had no such clearance, thus causing the short.

Also checked on another dead vidcard I have, and with a little soldering and careful tinkering, I should be able to replace the two fried components. And hopefully the card should work after that.

We'll see. Will update the thread after more testing

 

[micropage7]

cooling is pretty important, but its kinda harder to put heatsink on vrm than on processor
when you have some vrm on one row and its covered by heatsink you may notice one or two vrm that not pretty touch the heatsink so it may lead to poor cooling

 

[ShadowXP]

Just refit the card with new heatsinks for the VRM's. Had to trim them down to fit the chips (yay for Dremel ), and paid very close attention to the placement and spacing between the heatsinks and the other components. The card will be left to cure for a few hours, before I apply goop to the GPU and mount the rest of the cooler.

Will test it tomorrow to let everything get a chance to set properly. Built a testrig to handle it, but will monitor it VERY closely once testing sets in. A few hours of idle followed by an increasing amount of load on the card. Would think that World Computing Grid would be a nice way to thoroughly stresstest the card.

Will post results tomorrow Thanks for all tips etc.

 

[ShadowXP]

Well, after a few hours of tinkering with the heatsinks for the VRM's, the card has now been chugging away at both the Dx11 Aliens vs Predators benchmark and 3DMark11. The VRM-heatsinks were a few millimeters too long, so had trim them down a bit.

Benchmarks are all implying that the card works well, but the real test will be when I try it out when I get home and get it installed in the gaming-rig. It's looking promising, though!

Thanks to all for replies and tips

 

[silkstone]

Grats Shadow. It's never fun killing an expensive card. Whenever i put non-stock heatsinks on different components, i am alwas careful to check what they are touching for that reason....... which reminds me, i have some crappy ram sinks stuck with thermal adhesive on my mobo's VRM's... i'd better check they aren't falling off.

 

[AsRock]

Just use the stock plate if ya can see if the temps are better. At least with using the stock plate heat allways has a better place to go than some crappy heatsink.

 

[pantherx12]

+ one to using stockplate if it came to one, it will protect the card from warping under the weight of a after market heatsink as well.

 

[ShadowXP]

Here's the deal, AsRock and Panterx12: On an XFX-card, there IS no stockplate. The stockplate, if you can call it such, is incorporated into the main cooler of the card. When you unbolt the main cooler, the stockplate follows. And since that part of the stock cooler is just over an inch thick, you cannot fit both the stockplate and the new cooler on the card.

The good news is that the card performs as it should now. The short caused one of the DVI-ports to die, but with both an HDMI-port and two DisplayPorts in addition to the remaining DVI-port. Temps are good as well, with the highest recorded GPU-temp of 54c, and highest VRM-temp of 49c. This is both after an extended session of gaming (mostly World of Tanks and Just Cause 2, both maxed, but also Age of Conan in DX9-mode, all settings maxed). All temps measured with GPU-Z. Have also run both Furmark, 3DMark11, Heaven Benchmark, AvP DX11 and Just Cause 2's benchmarks without much variation in the temps. So things are looking good =)