Cooling the backside of a GPU

[jordy231]

HI All,

I have an MSI R9 280X Gaming 3G currently in my rig, one thing that I have noticed with this card is that the GDDR cooling is not great.

When i first bought this card if my core temps went higher than 55 degrees I would start artifacting all over the place.

To this end i set up an MSI afterburner fan profile that kicked up to 80% as soon as my GPU went higher than 35 degrees (i.e. as soon as i launched a game) and got progressively towards 100% as it reached 50 degrees. This worked great with many older games but still caused artifacts after hours of playing newer titles like farcry 4.

Once this started happening in farcry 4 I opened up the GFX card and replaced the thermal paste with arctic MX4, which gave me about another hour until the artifacts kicked in. So I opened it up again and put a little bit of paste on each of the VRam chips to give a better contact between the thermal tape and heat spreader plate, and voila fixed... until now.

Now playing assassins creed unity the artifacts are back, but now when my core reaches about 64 degrees.

I was wondering if anybody has tried cooling a GPU from the back? I've noticed that some aftermarket coolers come with a backplate heatsink such as the arctic accelero (http://www.arctic.ac/eu_en/accelero-xtreme-iv-280-x.html).

However it seems that you can't just get the back plate, and i'm not sure if it would be worth the £70.

Has anybody instead tried placing something such as these over the backside of the Vram, core and VR modules?

http://www.amazon.co.uk/Copper-Memory-Chipset-Heatsinks-Thermal/dp/B00KBMB76I/ref=pd_sim_computers_2?ie=UTF8&refRID=1SZ4SHTHHJASJJS762S0

Also is there any danger in trying? I would of course use a non-conducting thermal pad to attach the heatsinks to the reverse side?

 

[rolli59]

Well max temp for that card is in the mid 90's and the card should have been RMA'd before disassembly and void warranty. Looks to me you simply have a broken card.

 

[jordy231]

I understand your point about disassembly, but this card was bought as a second hand without any warranty or guarantee, so my only options were to see what i can do myself

 

[dizzycriminal]

Artifacts at 55 degrees doesn't sound right at all to me. GPUs should operate normally up to 60-70C at least in my experience. With many holding steady at 70C +. I know this isn't answering your question but is there any chance of replacing the graphics card entirely, because it sounds faulty?

 

[nukemaster]

I do not think it will hurt as long as you are using a non conductive pad like Arctic Cooling does. I do not think it is as effective as cooling the memory directly either however

The idea for them is to get enough heat by cooling the board(most boards use the traces are ground planes as a heatsink anyway). and avoid needing heatsinks that can not be removed from memory. This was always one thing users did not like. Older Arctic Cooling units had thermal glue to hold the heatsinks on the VRM(mosfets)/Memory on the card. So if you had to RMA the card or swap the cooler to a new card, the heatsinks could not be removed. With this new system, you just pull the cooler and backplate off the card and you are good to go.

I am not 100% sure about thermal paste on thermal pads, that is generally recommended against, but you seem to have had luck with it.

I do not think that card should artifact at those temperatures either, I would have RMAed the thing if it was under warranty. You may be able to stick a fan on the back of the card(prop it up with rubber pads or some other non conductive material) to aid cooing. It is cheaper and may help some.

 

[dizzycriminal]

I understand your point about disassembly, but this card was bought as a second hand without any warranty or guarantee, so my only options were to see what i can do myself

In that case try lowering your clock speeds in CCC and see if its not just the clock causing the problems.
Keeping the card under load at 55 or less is going to take a serious cooler without lower your clocks anyway.

 

[Neur0nauT]

The card would still retain at least a 3 year manufacturers warranty, so you still may have been able to RMA it for having faulty VRAM. It does sound like the VRAM is faulty if you are artifacting at 50c. I've seen it before with cards. The only thing I was able to do in the end was underclock the VRAM to prevent it overheating and causing artifacting.

 

[jordy231]

I think the issue with the 280Xs is cooling on the memory, from what research I have done many have had memory issues, although the core is reporting 55/60 degrees, the core is being effectively cooled, whilst i believe that the ram is not, especially with the "heat spreader plate" essentially a metal block obstructing all airflow directly to the ram and instead relying on this piece of flat metal to remove the heat.

I had initially thought about adding heat sinks directly to the memory but there isn't enough clearance between the Twin Frozr IV and the ram to get a heat sink in there.

With regards to getting a new graphics card, I am definitely looking to get one, the only problem is I am a student at the moment.

Hopefully when i graduate this summer and start earning i can invest in a 960 or 970. If i could sell this card on to make back half the money to get a 960/970 i would do so right away, but i wouldn't want to lump a faulty card onto someone unsuspectingly

 

[Neur0nauT]

I think the issue with the 280Xs is cooling on the memory, from what research I have done many have had memory issues, although the core is reporting 55/60 degrees, the core is being effectively cooled, whilst i believe that the ram is not, especially with the "heat spreader plate" essentially a metal block obstructing all airflow directly to the ram and instead relying on this piece of flat metal to remove the heat.

I had initially thought about adding heat sinks directly to the memory but there isn't enough clearance between the Twin Frozr IV and the ram to get a heat sink in there.

Since you have already voided the warranty (if there still was one) you could consider getting a water block for the card, going down the watercooling route. This would probably allow you to overclock the card after fitting one, without getting overheating issues.

 

[rolli59]

I think the issue with the 280Xs is cooling on the memory, from what research I have done many have had memory issues, although the core is reporting 55/60 degrees, the core is being effectively cooled, whilst i believe that the ram is not, especially with the "heat spreader plate" essentially a metal block obstructing all airflow directly to the ram and instead relying on this piece of flat metal to remove the heat.

I had initially thought about adding heat sinks directly to the memory but there isn't enough clearance between the Twin Frozr IV and the ram to get a heat sink in there.

I had memory issues with my HD7950 and fixed it by RMA. New card had no issues at all.

 

[jordy231]

I think the issue with the 280Xs is cooling on the memory, from what research I have done many have had memory issues, although the core is reporting 55/60 degrees, the core is being effectively cooled, whilst i believe that the ram is not, especially with the "heat spreader plate" essentially a metal block obstructing all airflow directly to the ram and instead relying on this piece of flat metal to remove the heat.

I had initially thought about adding heat sinks directly to the memory but there isn't enough clearance between the Twin Frozr IV and the ram to get a heat sink in there.

Since you have already voided the warranty (if there still was one) you could consider getting a water block for the card, going down the watercooling route. This would probably allow you to overclock the card after fitting one, without getting overheating issues.

I had contaacted MSI previously about the fault but there response was to contact the seller, to which the seller never responded, so I suspect they may have been aware of the issues.

I have since contacted MSI to which there responses were all trivial such as try a different HDMI, try a different display, update drivers etc.

This was the point at which i decided to take things into my own hands

 

[rolli59]

I think the issue with the 280Xs is cooling on the memory, from what research I have done many have had memory issues, although the core is reporting 55/60 degrees, the core is being effectively cooled, whilst i believe that the ram is not, especially with the "heat spreader plate" essentially a metal block obstructing all airflow directly to the ram and instead relying on this piece of flat metal to remove the heat.

I had initially thought about adding heat sinks directly to the memory but there isn't enough clearance between the Twin Frozr IV and the ram to get a heat sink in there.

Since you have already voided the warranty (if there still was one) you could consider getting a water block for the card, going down the watercooling route. This would probably allow you to overclock the card after fitting one, without getting overheating issues.

I had contaacted MSI previously about the fault but there response was to contact the seller, to which the seller never responded, so I suspect they may have been aware of the issues.

I have since contacted MSI to which there responses were all trivial such as try a different HDMI, try a different display, update drivers etc.

This was the point at which i decided to take things into my own hands

I hope you can fix it but I do not want to give much hope in that direction since the artifacting is a sign that the memory is bad and corrupting the textures.

 

[jordy231]

I think the issue with the 280Xs is cooling on the memory, from what research I have done many have had memory issues, although the core is reporting 55/60 degrees, the core is being effectively cooled, whilst i believe that the ram is not, especially with the "heat spreader plate" essentially a metal block obstructing all airflow directly to the ram and instead relying on this piece of flat metal to remove the heat.

I had initially thought about adding heat sinks directly to the memory but there isn't enough clearance between the Twin Frozr IV and the ram to get a heat sink in there.

Since you have already voided the warranty (if there still was one) you could consider getting a water block for the card, going down the watercooling route. This would probably allow you to overclock the card after fitting one, without getting overheating issues.

I had contaacted MSI previously about the fault but there response was to contact the seller, to which the seller never responded, so I suspect they may have been aware of the issues.

I have since contacted MSI to which there responses were all trivial such as try a different HDMI, try a different display, update drivers etc.

This was the point at which i decided to take things into my own hands

I hope you can fix it but I do not want to give much hope in that direction since the artifacting is a sign that the memory is bad and corrupting the textures.

Thankyou very much for your hope!

I suppose I will give it a go and place heatsinks on the reverse of the ram as it will only be ~£5 to experiment, and for now I will just have to stick with aggressive fan profiles and just try and eek 3 more months out of it, or maybe just 2 and ill treat myself for graduation.

 

[Neur0nauT]

I would waterblock it, unless you want to just get rid of the card.

http://www.coolingconfigurator.com/step1_complist?gpu_gpus=1186

 

[jordy231]

I would waterblock it, unless you want to just get rid of the card.

http://www.coolingconfigurator.com/step1_complist?gpu_gpus=1186

If as you stated previously the card is faulty, I would rather not spend that amount of money on it, as it would be better put towards a new card with a full warranty.

I would rather just go for a cheap fix until I can take the risk of shelling out £200 on a new graphics card

 

[nukemaster]

In general the memory does not really need much cooling and the metal base plate is more than upto the job. It is just defective for sure.

You may be able to gently lower the memory frequency. It will have a performance hit, but you may find a spot that the artifacts stop happening. in some cases a defective gpu can cause these issues as well(With my first GTX 670 I think I had to take the card down by 26mhz to stop all problems[A bit different since it was the cards own boost feature causing these issues], the memory was fine). Now I RMAed that card and the new one not only runs perfect, but boosts higher(luck of the draw).

 

[Neur0nauT]

Yeah like Nukemaster says, I also decreased the clock on VRAM when I had artifacts due to overheating VRAM and I was able to use the card until my next upgrade. I did lose some performance, however I found a happy medium by overclocking the GPU slightly until it was just under bottleneck, and the card worked well. Still does.

Just bad luck with VRAM sometimes.

 

[jordy231]

Yeah like Nukemaster says, I also decreased the clock on VRAM when I had artifacts due to overheating VRAM and I was able to use the card until my next upgrade. I did lose some performance, however I found a happy medium by overclocking the GPU slightly until it was just under bottleneck, and the card worked well. Still does.

Just bad luck with VRAM sometimes.

Yeah one other thing I have tried is underclocking, this does seem to help with the VRAM, and I've also noticed I can push an extra 100Mhz out of the core without making anything worse