Heatsinks on VRAM?

[SteveGrabowski0]

I have the EVGA GTX 970 SC ACX 2.0 (2974-KR) and they made some puzzling design choices. For instance, they moved the vrms to the other side of the core from reference so that using a Kraken G10 + AIO is a complete no go, since the fan will be blowing on the right of the card while the vrms are on the left. They also put four vram chips on the top of the PCB so that they receive no cooling whatsoever. I'm wondering if anyone has managed to get better memory overclocks in such a situation by putting small heatsinks on the vram chips so they at least get some passive cooling. You can see the four uncooled ram chips here on top of the pcb:

 

 

If you get their backplate you can see thermal pads here where the vram chips are:

 

 

I also wonder if I might be better off with the backplate. Would that end up being treated like a big heatsink that could be more effective for cooling the vram chips than buying some small copper heatsinks like these:

 

 

Or will neither be likely to do anything for memory overclocks? I ask because I can get a pretty significant boost in my framerate (3 fps or so) on Witcher 3 by giving my vram a 100 MHz overlock on my 970, and it's actually pretty noticeable. But I get driver crashes sometimes with that 100 MHz memory overclock, so I'm wondering if getting some passive cooling to those four exposed vram chips might be the little bit needed to keep that memory overclock stable.

[TheRandomness]

I'm pretty sure some larger individual heatsinks would work better but don't take my word for it.

[don_svetlio]

VRAM usually doesn't need cooling unless you are overvolting it

[Stefan1024]

These heat sinks are not made for passive cooling. They rely heavily on airflow from a nearby fan. Also try to find >10mm high once, the ~3mm once you showed off are pritty bad.

[SteveGrabowski0]

These heat sinks are not made for passive cooling. They rely heavily on airflow from a nearby fan. Also try to find >10mm high once, the ~3mm once you showed off are pritty bad.

 

Do you think the 10+ mm heatsinks would be likely to be a better solution than using a metal backplate with thermal pads? I do have a 200mm fan blowing over the GPU, but it's in a large case (Phanteks Enthoo Pro) so it might not make a huge difference. I can install the top drive cage and put a 120mm fan that should get reasonably close to the GPU though. I just wonder if it'll actually do anything, and whether temperature is a big driving factor towards limiting memory overclock.

[Stefan1024]

Do you think the 10+ mm heatsinks would be likely to be a better solution than using a metal backplate with thermal pads? I do have a 200mm fan blowing over the GPU, but it's in a large case (Phanteks Enthoo Pro) so it might not make a huge difference. I can install the top drive cage and put a 120mm fan that should get reasonably close to the GPU though. I just wonder if it'll actually do anything, and whether temperature is a big driving factor towards limiting memory overclock.

The backplate with thermal pads should be as good as most VRAM heat sinks.

Put you can test the OC potential easely by using a fan that is blowing full spped directly on the VRAMs. It's loud but effective. When you can OC higher than try to cool them quieter. If not, I don't think you will gain performace by cooling.

[SteveGrabowski0]

The backplate with thermal pads should be as good as most VRAM heat sinks.

Put you can test the OC potential easely by using a fan that is blowing full spped directly on the VRAMs. It's loud but effective. When you can OC higher than try to cool them quieter. If not, I don't think you will gain performace by cooling.

 

Sounds like a plan. I'll try mounting a 120 mm fan right in front of the GPU later on, run it full blast, and see if it helps at all when testing overclocks on say Valley or Heaven. I probably should have done that before creating this thread lol.

[tomas710]

 

 

 

The tiny heatsinks work, the problem is they can fall, and if they got into the PSU it could be really bad. I would do nothing, but if you really want cooling, a backplate is a much better idea.

[SteveGrabowski0]

The backplate with thermal pads should be as good as most VRAM heat sinks.

Put you can test the OC potential easely by using a fan that is blowing full spped directly on the VRAMs. It's loud but effective. When you can OC higher than try to cool them quieter. If not, I don't think you will gain performace by cooling.

 

I have been benchmarking for a couple of hours now testing various settings, and getting that 120 mm fan blowing across the top of the card has made a pretty big difference for my memory overclock. I think it was due to the memory chips on top of the card being uncooled, because putting that 120mm fan on the top hard drive cage did absolutely nothing for my core overclock (still stuck at 1418 MHz there and only got about 2C lower temp on the core), but my memory overclock went from a very unstable +100 MHz (I really needed to set it to 0 for stability) and now a rock solid +400 MHz bump, so my effective speed goes from 7 GHz to 7.8 GHz on my vram. Cranking it up to +500 I crashed pretty quickly, but at +400 it's stable.

 

I can't believe that, why did they put vram chips on an uncooled section of the card? It's not like my initial setup offers bad cooling: I mean, it's an Enthoo Pro with the top drive cage removed, so it should be really strong at cooling a GPU according to these results.

 

[Stefan1024]

- snip -

 

Everybody likes to hit on AMD for theier bad cooling. But most of them forget Nvidia isn't better at all only because the core temperature is a bit better.

[ThomasD]

I would not called them uncooled, as they are connected to a rather large heatsink by thermal pads.  But those thermal pads are also darn thick - which means they probably do not transfer heat well at all.  

 

What might be more interesting is replacing those pads with copper shims of suitable thickness along with adding TIM on both sides (or super thin transfer tape.)  Given the even spacing around the pcb it shouldn't be too hard to get good even contact. 

 

 Not only would that increase the total mass of the heat sink, done properly it also improves heat transfer.