5970 VRM Cooling Solutions

[st4rdust]

Most 5970 owners probably know that the VRMs are a bottleneck to overclocking and cooling. Though they were on the market for some time, I know that the Accelero aftermarket HSF has been discontinued for some time. While I'm sure water cooling is an effective way around this problem, I'm limited to air for the time being so I want to cool these chips as best I can.

I plan on trying a couple of ideas I've read about so far. One is placing a fan below the GPUs at the bottom of the case, blowing air up to the cards. I guess the idea behind this is that it provides an additional stream of cool air in addition to the case fans blowing from front to back. If this turns out to lower the temperatures significantly, I might consider adding more fans along the way to keep this bottom to top flow moving with an exhaust fan at the top of the case and perhaps another on top of the upper GPU pulling the air from the cards and up to the exhaust.

An idea that I had which is more on the ridiculous side of the spectrum is to somehow recycle my old accelero x2 coolers (these adorned my nvidia 6800gt cards many years back) to give the VRMs their own heatsink, fan, or both. Maybe a more practical solution would be to add some copper RAM heatsinks to the chips on the top side. As for the chips under the stock HSF, I'm at a loss right now for additional cooling. Anyone experienced with the 5970 card have any pearls of wisdom?

EDIT: I finally took apart one of my cards to try and reverse-engineer my way to more efficient cooling and so far it looks promising. I've removed the stock fan and the outer shell casing, leaving the heatsink exposed.







The heatsink fins are bent at the very top, perhaps in order to channel airflow through the fins and out the exhaust. While I could just strap a few 120mm fans across the top of the card a la Arctic Cooling's 5970 HSF, what I estimate would be the most effective way to cool the heatsink would be to channel the airflow of a single 120mm fan down through the fins, possibly cutting out a makeshift duct out of cardboard to adapt the fan down to the opening of the heatsink. More updates to follow.

Another edit: I added two 120mm fans on top of the heatsink but barely got to test it out. While the GPU temps looked very promising, the VDDC sensors ran up to the 130s so I shut it down right away. I sent for some new thermal pads in the hopes of bringing these temps down.

[JL421]

Try to pump as much air as possible into those things, or try to suck it out of them through the back. I also replaced the thermal pads with some 1mm thick Phobya thermal pads, it cooled those suckers down about 10C with 20% less fan...other than that, water cool them if you can.

[st4rdust]

Try to pump as much air as possible into those things, or try to suck it out of them through the back. I also replaced the thermal pads with some 1mm thick Phobya thermal pads, it cooled those suckers down about 10C with 20% less fan...other than that, water cool them if you can.

Aftermarket thermal pads definitely crossed my mind. 1mm pads seem to be the right fit for the 5970 VRMs?

Once my air cooling solutions are exhausted, then I might have to go to water. If nothing else, then just for the top card since the one below it hasn't given me any trouble as far as temperatures or OC limitations.

[MadHacker]

myself I'm going to watercool.
attach my rads to cold air vent and dump the heat in the house.
come summer i can easily move the rads outside.

[st4rdust]

Updated with new pictures.

[MadHacker]

The heatsink fins are bent at the very top, perhaps in order to channel airflow through the fins and out the exhaust. While I could just strap a few 120mm fans across the top of the card a la Arctic Cooling's 5970 HSF, what I estimate would be the most effective way to cool the heatsink would be to channel the airflow of a single 120mm fan down through the fins, possibly cutting out a makeshift duct out of cardboard to adapt the fan down to the opening of the heatsink. More updates to follow.

looking at the Arctic Cooling's 5970 HSF, the fins are vertical not horizontal so when the 120mm fans blows on it teh hot air escapes top and bottom.

the stock heatsink is vertical so adding more fans wouldn't help that much.
adding a 120mm fans and making a makeshift duct for it i think would be the simplest & best option.

[P4man]

I have a thermal take VRM R5 to cool the VRMs of my 5870. Im not sure its compatible with the 5970, but if it is, get one, its incredible. VRM temps dont exceed 55C no matter what (down from 110+ with the stock cooler). For good measure I added a spitfire cooler for the GPU itself, and now those temps dont exceed 50C. Actually, at this very moment..  as its cooling outside (the rig is in the shed, ambient temps inside around 15C) a stunning 38.5C @900 Mhz.



GPU 0 is the 5870. The other one is a 5850.
Well, a spitfire wont work on a 5970, but the VRM cooler might. Here is a pic:

http://s8.postimage.org/paa2d2ef7/SDC10725.jpg

The VRM cooler is the "small" one with the old orange fan. Fan is completely unnecessary btw.

[MadHacker]

I have a thermal take VRM R5 to cool the VRMs of my 5870. Im not sure its compatible with the 5970, but if it is, get one, its incredible. VRM temps dont exceed 55C no matter what (down from 110+ with the stock cooler). For good measure I added a spitfire cooler for the GPU itself, and now those temps dont exceed 50C. Actually, at this very moment..  as its cooling outside (the rig is in the shed, ambient temps inside around 15C) a stunning 38.5C @900 Mhz.



GPU 0 is the 5870. The other one is a 5850.
Well, a spitfire wont work on a 5970, but the VRM cooler might. Here is a pic:

http://s8.postimage.org/paa2d2ef7/SDC10725.jpg

The VRM cooler is the "small" one with the old orange fan. Fan is completely unnecessary btw.

nice cooler but will be hard pressed to get 3 or 4 5970's on a single PC...
longer extention cables perhaps.

[st4rdust]

Added another update

[P4man]

Judging by pictures, the VRM R-5 could fit your card.
http://www.thermalright.com/new_a_page/product_page/vga/vrm/product_vga_cooler_r5.html

Now  I cant find any official confirmation, because obviously the GPU cooler that goes with it (spitfire) is not compatible with a dual card.  I had originally bought the R5 to cool the VRMs of a 5850, and I cut the baseplate in half, so it still covered the RAM and the small memory VRM, while making place for the R5 to cool the VRMs. Worked fine (but is kind of irreversible.) I have it on my 5870 now, and clocked at 1 GHz, the VRMs have yet to exceed 62C (stock cooler, stock speed  was 115+C).

Next time you remove the baseplate, can you make some pictures of the VRMs with a ruler next to it for size? It looks exactly like the 5870 row of VRMs.

Then again, like the spitfire, the R5 is not a good idea if you have multiple cards. Its kind of big. You might be able to bend the heatpipes though, so they stand straight up.

[MadHacker]

Looking at my temps... I'm leaning towards watercooling.

[st4rdust]

Next time you remove the baseplate, can you make some pictures of the VRMs with a ruler next to it for size? It looks exactly like the 5870 row of VRMs.

Not sure if these are the VRMs you wanted, but here goes:

[JL421]

Next time you remove the baseplate, can you make some pictures of the VRMs with a ruler next to it for size? It looks exactly like the 5870 row of VRMs.

Not sure if these are the VRMs you wanted, but here goes:

There is also a second set of VRMs for the other core above the crossfire bridge logic chip. You can see it in the picture, about 1.5 inches above that second large chip like thing in the background.

[DeathAndTaxes]

And that bank of VRM is actually 2 sets of VRM.  The smaller one provides power for the RAM (which runs at different voltage).  The other for GPU1.  The bank at top of the card provides power for GPU2.

This photo (link only because it is large) shows it more clearly.
http://lab501.ro/wp-content/uploads/2009/11/HD5970-Inside-09.JPG

The CPLA-3-50 and CPL-2-50 are inductors for VRMs.  The small squares next to each inductor are the VRM in that group.

[RandyFolds]

The heatsink fins are bent at the very top, perhaps in order to channel airflow through the fins and out the exhaust. While I could just strap a few 120mm fans across the top of the card a la Arctic Cooling's 5970 HSF, what I estimate would be the most effective way to cool the heatsink would be to channel the airflow of a single 120mm fan down through the fins, possibly cutting out a makeshift duct out of cardboard to adapt the fan down to the opening of the heatsink. More updates to follow.

looking at the Arctic Cooling's 5970 HSF, the fins are vertical not horizontal so when the 120mm fans blows on it teh hot air escapes top and bottom.

the stock heatsink is vertical so adding more fans wouldn't help that much.
adding a 120mm fans and making a makeshift duct for it i think would be the simplest & best option.

I don't think you will be able to get any decent flow with 120mm fans. They aren't pressure fans, so they will more than likely just spin in place instead of pushing air through the ducting. The reason they use those squirrel cage fans is because they can handle the backpressure of forcing air through that style of heatsink.

[DeathAndTaxes]

Yup.  CFM =/= pressure.  You can have a high CFM fan with no pressure or a lower CFM fan w/ higher pressure.

If you did want to try w/ some 120mm fans you would want some 38mm thick Deltas designed for high air pressure situations.  It isn't going to be quiet though.

[P4man]

The VRM R5 should fit to cool that rear row of VRMs. For those 3 in the middle you would have to cook up something else. Either the baseplate, that you would have to cut to make room for the R5, and on which you can glue lots of sticky heatsinks, or something else.

Im not sure which VRMs do what on this card, but on a  5870 the rear row has all the VRMs that supply the GPU, and closer to the front of the GPU is a VRM that powers the RAM chips, and as a result doesnt get nearly as hot (still needs a heatsink, but a simple ram cooler with some hotglue or even zip tie works well enough). But if those 3 VRMs provide power to the second GPU, you will want something better than a tiny ziptied heatsink.

Edit: I see 3 banks. One large one, near that rules (6 VRMs), one with 3 in a row near the top, presumably for the second GPU, and I see a single one inbetween, also near the top, which I would guess is for the RAM.

[Mr.Coin]

If you open up the fins on the heat sink, your temps would drop a lot more.  Love the idea.

[manet]

I have made some solution for cooling VRMs on my 5970. First, after opening the card I removed original VRM thermal pads, which fell apart and seemed dried.  I replaced them with some 1mm thick, 6W/mK thermal conductive pads from thermagon inc., but in result the VRM temps haven't improved.
Later I decided to experiment more and replace the thermal pads with copper plate and thermal grease.  I bought a 0.3mm thick copper plate and cut from it two strips to cover the VRM banks. I put nonconductive thermal grease (zm-stg2) on the surface of each VRM in the two banks and on the radiator side where VRM thermal pads had been previously stuck. Then, I carefully stuck the copper strips to the radiator side and put the card together. Results:
Pros:
- vrm temps dropped about 10-15 degrees C and stayed below 120 degrees C (threshold temp) even with high core voltage like 1.15V,
- overclocking capability increased significantly – voltage could be risen much higher compared to pads
Cons:
- gpu temps increased about five 2-3 degrees C, probably partly because of the better conductivity of heat from the vrm bank between gpu cores to the radiator which in result heats more when cooling gpus, but maybe also the original grease on gpus was better than my zm-stg2.
- I have read that producer uses thermal pads because during manufecturing the vrms could be soldered/placed at different hight and thermal pad could compensate this differences. It implies that the copper strip could not stick well to all vrms, or the one sticking out to much could get smashed maybe.
- quite risky - copper strip shouldn't touch any element...,  the pcb area close to vrms gets dirty with grease

Overall, I think that for bitcoin mining this modification could not be needed, because increasing voltage is not economical. And standard vrm cooling (with pads) is rather enough for economical (slight) oveclocking.

Remarks:
I also tried putting only thermal grease instead of pads and resulting vrm temps were better than with pads, so vrms are well tightened to radiator in my card at least.
The vrm bank with three vrms located between gpu cores heats more, in my card about 10 degrees C more, than the other bank (above fan), contrary to what this article says (I think, they misidentified the vrm banks): http://www.anandtech.com/show/3590 . But otherwise it's worth looking at, there are good pictures of the card.

[plastic.elastic]

I have made some solution for cooling VRMs on my 5970. First, after opening the card I removed original VRM thermal pads, which fell apart and seemed dried.  I replaced them with some 1mm thick, 6W/mK thermal conductiv pads from thermagon inc., but in result the VRM temps haven't improved.
Later I decided to experiment more and replace the thermal pads with copper plate and thermal grease.  I bought a 0.3mm thick copper plate and cut from it two strips to cover the VRM banks. Then, I put non conductive thermal grease (zm-stg2) on the surface of each VRM in the two banks (the VRM area on pcb would get dirty with grease) and covered it carefully with the copper strips. And also, I put thermal grease on the radiator side where vrm thermal pads had been sticked. Results:
Pros:
- vrm temps dropped about 10-15 degrees C and stayed below 120 degrees C (threshold temp) even with high core voltage like 1.15V,
- overclocking capability increased significantly – voltage could be risen much higher compared to pads
Cons:
- gpu temps increased about 5 degrees C, probably partly because of the better conductivity of heat from the vrm bank between gpu cores to the radiator which in result heats more when cooling gpus, but maybe also the original grease on gpus was better than my zm-stg2.
- I have read that producer uses thermal pads because during manufecturing the vrms could be soldered/placed at different hight and thermal pad could compensate this differences. It implies that the copper plate could not stick well to all vrms, or the one sticking out to much could get smashed maybe.
- quite risky - copper plate shouldn't touch any element...

Overall, I think that for bitcoin mining this modification could not be needed, because increasing voltage is not economical. And standard vrm cooling (with pads) is rather enough for economical (slight) oveclocking.

Remarks:
I also tried putting only thermal grease instead of pads and resulting vrm temps were better than with pads, so vrms are well tightened to radiator in my card at least.
The vrm bank with three vrms located between gpu cores heats more, in my card about 10 degrees C more, than the other bank (above fan), contrary to what this article says (I think, they misidentified the vrm banks): http://www.anandtech.com/show/3590 . But otherwise it's worth looking at, there are good pictures of the card.

That is also because the GPU die got less contact pressure or lack of contact due to the copper strip being too thick?
I dont know but i assumed you did use precision measurement of the thickness needed for VRM?