Watercooled 3080/3090 VRAM temps - Copper Shim installation

Soldato
Joined
18 Oct 2002
Posts
7,983
Location
New Eltham, London
Have just ordered a CoolMyGPU copper ram shim from the US for $75 delivered. Looks like a neat way to solve the problem of ram height and thermal pads by using a shim that surrounds the core. thermal paste on the ram chips, shim on, thermal paste on the shim, waterblock installed. Can't wait to try it.
 
Associate
Joined
8 Nov 2006
Posts
244
Location
Scotland
LMFj3mM.png


3080 37~38 under load at 0.95v 2040 or 0.981v 2100. That jump in voltage didn't seem to matter.

Memory 50~55.

FdJGwjI.png


Weather has taken a turn in the last 24 hours lol, did someone say it's cold?
 
Associate
Joined
8 Nov 2006
Posts
244
Location
Scotland
@Audioboxer what cooling do you have on that card?

Just an EK Vector block and their nickel backplate (B-stock from OCUK! lol).

I've got 4 rads in a Lian Li O11 XL. 3x360 and 1x120. Two XR5, one XR7 and an EK SE Slim to fit in the rear back chamber (along with rear fans, 30mm rad is too thick, 27mm just fits with 25mm fans ~ Side is therefore push/pull with just the QLs in picture in the front of case). So low temps on the GPU are likely a combination of good ambients right now in the UK and the radiator space.

ibMPPEd.jpg


Semi-recent picture, though since it was taken I've changed the top to intake instead of exhaust and switched the rear to exhaust so I have at least 1 exhaust. I tried ALL intake briefly but due to the glass front/side I feel all intake creates a bit too much of a hotbox inside. Not enough holes and gaps in this case IMO to have all intake and just rely on the all the air finding its own way out.

Temps might even be a degree or two better if I dumped all the QLs from push/pull and used all P12s, but hey, a little bling keeps the doctor away.... As long as you don't run QL120 fans on rads on their own (they're awful).
 
Last edited:
Soldato
Joined
18 Oct 2002
Posts
7,983
Location
New Eltham, London
Well the shim has arrived.



This will replace all the thermal pads around the gpu core covering the RAM chips.

Current RAM temps if mining (stopped doing that but it seems a very good stress test for RAM) are around 100C. When [email protected] (normal behaviour) with card power limited to 80%, RAM temps are 80C, GPU hot spot at 53C with card at 1950 and 98% gpu load.

Will plan an urgent teardown and rebuild.
 
Soldato
Joined
18 Oct 2002
Posts
7,983
Location
New Eltham, London
Should be starting this tomorrow.

As a benchmark I have fired up NiceHash and while the GPU toddles along at 975MHz and a hotspot of 47C, the RAM at 9751MHz has a Memory Temperature (GPU-Z 2.45.0) of 96C.

I plan to make as few unnecessary changes to the loop as possible but I will be replacing the CPU block which is after the gpu in loop order.
 
Soldato
Joined
18 Oct 2002
Posts
7,197
Location
Scun'orp
Going to have some really nice flowrate on those, what, maybe 3 central grooves? The rest of the grooves are on their own. Bit of a joke there.
 
Soldato
Joined
18 Oct 2002
Posts
7,983
Location
New Eltham, London
TLDR. shim does not work with my block and card. basically there is no contact with the core if the shim is in place. I have gone back to some Gelid 1mm pads on the ram but am still seeing 90c on the ram when mining.
 
Soldato
Joined
18 Oct 2002
Posts
7,983
Location
New Eltham, London
Possible to lap the shim at all?
it's very thin already and would need to go down very evenly across four interconnected rectangles by the same amount.the big problem is the amount of time and error it would take to get it to where it needs to be. If anyone wants it to try with their card, I will happily send on.

When I had it fitted, if only the core was hot but the memory was cool I would have maybe worked harder on this but the RAM temps were over 100C and so I am unconvinced this thing works at all.
 
Associate
Joined
20 Dec 2020
Posts
290
TLDR. shim does not work with my block and card. basically there is no contact with the core if the shim is in place. I have gone back to some Gelid 1mm pads on the ram but am still seeing 90c on the ram when mining.
Thats a shame as I was hoping it would work out for you.
 
Soldato
Joined
18 Oct 2002
Posts
7,983
Location
New Eltham, London
Update: This morning, the 3080 is no longer in device manager and there is no display whatsoever. My PC works off its internal cpu graphics.

Having completed yesterday my 10 billionth point for [email protected], perhaps it is time to stop. I have a reference water-cooled 1070 lying around so will drop that in and inspect the 3080 but I think I may have killed it.
 
Soldato
Joined
18 Oct 2002
Posts
7,983
Location
New Eltham, London
UPDATE! She lives! spent some time trying to diagnose why the card was dead. Drained and connected up just the cpu, refilled. Tried a 670. Didn't get seen. Removed the vertical PCIE extender and tried again with the 670. Saw it but as a Microsoft Basic VGA adapter. Tried a 1070. Saw it OK. refitted the PCIE extender and tried the 1070 again. Saw it OK.

Refitted the 3080. Saw it! WTAF!

So since I had the card out of the loop, I fitted the Gelid 1.5mm thermal pads, drained the loop, reconnected up the loop, pressure tested, refilled and restarted. Nicehash. Horrible temps. Immediate 90 on both core and ram. WTF?

Drained, removed block. cleaned paste, refitted block super carefully and re-installed. Nicehash. Currently showing core at 1080MHz and 50C with a hotspot of 68.3 and the RAM at 1269MHz and


wait for it


68C! That's 24C lower than with 1mm pads. half a frigging millimetre makes 24C difference.

I'm now worried that my test of the shim was invalid and may have been the result of poor fitting. Not gonna mess with it today as have spent waaaay too much time on this draining and refilling maybe 6 times!
 
Last edited:
Associate
Joined
20 Dec 2020
Posts
290
UPDATE! She lives! spent some time trying to diagnose why the card was dead. Drained and connected up just the cpu, refilled. Tried a 670. Didn't get seen. Removed the vertical PCIE extender and tried again with the 670. Saw it but as a Microsoft Basic VGA adapter. Tried a 1070. Saw it OK. refitted the PCIE extender and tried the 1070 again. Saw it OK.

Refitted the 3080. Saw it! WTAF!

So since I had the card out of the loop, I fitted the Gelid 1.5mm thermal pads, drained the loop, reconnected up the loop, pressure tested, refilled and restarted. Nicehash. Horrible temps. Immediate 90 on both core and ram. WTF?

Drained, removed block. cleaned paste, refitted block super carefully and re-installed. Nicehash. Currently showing core at 1080MHz and 50C with a hotspot of 68.3 and the RAM at 1269MHz and


wait for it


68C! That's 24C lower than with 1mm pads. half a frigging millimetre makes 24C difference.

I'm now worried that my test of the shim was invalid and may have been the result of poor fitting. Not gonna mess with it today as have spent waaaay too much time on this draining and refilling maybe 6 times!
Great news :D
 
Top Bottom