Well, I've seen people complaining about the copper plate, for fun, what happens if you use the air cooler? does the memory stay in check?

Lordred
Well, I've seen people complaining about the copper plate, for fun, what happens if you use the air cooler? does the memory stay in check?

You mean the stock air cooler or the mini-fan that people have suggested?

oninaig
On my second hybrid kit for the 3090 FTW3 after I had the RMA the first one. The first one I had did not have this issue but the refurbished one they sent me seems to have a very hard time keeping the memory junction temperatures under control and this isn't just during benchmarks either. During games like Tarkov or any heavy DX11 game the memory junction temperature creeps up to 104-106 and at that point the RPM readout for the GPU2 fan goes to 0 while the fan actually spins up to 100% in real life until the temperature drops down below 100.
 
Here's an example: 
 
I've taken apart and re-assembled the hybrid kit multiple times to verify that the thermal pads are all making good contact. The only thing I can think of is that the thermal pads that are on the copper plate that covers the VRAM surrounding the chip itself are too thick, maybe the unit I got (refurbished) was originally returned with non-factory pads that were too thick?
 
How thick should the pads that are on the copper plate be? What do I have to do to this card to get its MJT under control for just daily use?
 
The card itself is all stock, no overclock at all.
 
Edit: I've seen people place big heatsinks on the back of their cards or smaller fans that blow air directly over the back of the card but most of the people I see doing that are mining 24/7 and I am just gaming with the card at stock settings so I feel like I shouldn't be seeing these temps.
 
 

If you're taking it a part multiple times, are you remembering to repaste the memory cold plate?

You have to add thermal paste in-between the AIO and memory cold plate.

https://i.imgur.com/S7Oqoh6.jpeg

KingEngineRevUp

oninaig
On my second hybrid kit for the 3090 FTW3 after I had the RMA the first one. The first one I had did not have this issue but the refurbished one they sent me seems to have a very hard time keeping the memory junction temperatures under control and this isn't just during benchmarks either. During games like Tarkov or any heavy DX11 game the memory junction temperature creeps up to 104-106 and at that point the RPM readout for the GPU2 fan goes to 0 while the fan actually spins up to 100% in real life until the temperature drops down below 100.
 
Here's an example: 
 
I've taken apart and re-assembled the hybrid kit multiple times to verify that the thermal pads are all making good contact. The only thing I can think of is that the thermal pads that are on the copper plate that covers the VRAM surrounding the chip itself are too thick, maybe the unit I got (refurbished) was originally returned with non-factory pads that were too thick?
 
How thick should the pads that are on the copper plate be? What do I have to do to this card to get its MJT under control for just daily use?
 
The card itself is all stock, no overclock at all.
 
Edit: I've seen people place big heatsinks on the back of their cards or smaller fans that blow air directly over the back of the card but most of the people I see doing that are mining 24/7 and I am just gaming with the card at stock settings so I feel like I shouldn't be seeing these temps.
 
 

If you're taking it a part multiple times, are you remembering to repaste the memory cold plate?

You have to add thermal paste in-between the AIO and memory cold plate.

https://i.imgur.com/S7Oqoh6.jpeg

 
So I opened it back up and applied thermal compound to the cold plate. Does it have to be thermal paste? 
 
After doing so, I ran furmark 3840x1080 for 20 minutes and I got the following results:




 
 

 
This is running the fans at 80% on the radiator and 80% on the card itself (fan 2).
 
If I drop the fans down to where they were on the air cooler (quiet preset), I get the following results:


 

 
I am using 2mm gelid pads on mem1, 2, and 3 and 1mm pads on the mosfets. I am using the stock pads for the VRM because I did not have 3mm pads on hand.
 
Could this also be because I used compound instead of paste for the contact between the cold plate and the AIO?

Update: I took it apart again and noticed that the left mosfets thermal pad did not have indents so I removed and replaced the pad on that side and put it back together without any paste/connections just to see if I could get some indents and I was not able to. I checked all over the card and could not find any obstructions in that area where the mosfets are so I am unsure why there would not be 100% pressure there. On top of this I don't want to screw the 4 back screws that screw into the copper block too tight.
 
As a test solution I added some MX-4 compound to the mosfet pads that touch the radiator side of the shroud and I ran my furmark test with fans at 80% across the board and this was the result:


 

 
Better than my first result as my max memory junction temperature dropped 4 degrees but my MEM1 and GPU2 temps are still upwards of 15 degrees hotter than my GPU1 temp, what else could be causing this?

So after tinkering with this some more I was able to get the Memory Junction Temperature down to a max of 98C during a furmark 3840x1600 test with the fan profile set to a max of 30% which is the noise level I like (nice and quiet). However, I realize that while the hybrid kit can get the GPU core down to a temperature lower than the air cooled card, it cannot (by itself) cool the memory down as well as the air cooled card did since the air cooled card had 3 fans blowing cool air directly over those modules. 
 
The hybrid kit tries to solve this with the backplate that sucks heat away from the memory chips and VRM via a series of thermal pads attached to those modules. There is also the cold plate that takes heat away from the 3 memory modules surrounding the core itself with the help of the AIO. 
 
So while I was able to successfully get my MJT down to <100C with a repadding and repasting job, I still wanted to see if I could cool the memory even more without increasing the speed of the fan which would generate more noise. So I took inspiration from the many posts from people who are mining with this card and applied some of their solutions with some success. I had a few Noctua NF-A4s lying around from my 3d printer builds and I also picked up some of the heatsinks that were recommended on the posts regarding cooling the card while mining. It turns out that the heatsinks are pretty much the perfect size to fit an NF-A4 right on top:


 

 

 
I used these together with some thermal pads I had from the repadding job and placed them directly over the memory modules that were giving me the most trouble (MEM1 and MEM3):




 
After this I re-ran the furmark test and my temperatures dropped down to a more comfortable 94ish for the MJT and ~79 for the MEM1, MEM2, and ~70 for MEM3:



I am still unsure why MEM1 and MEM2 are so much hotter than MEM3 because I was unable to find any obstructions in that area that would cause the thermal pads to not make good contact and I was unsure if it was safe to use thermal compound or putty in those spots to make up the difference. If anyone has a suggestion as to which brand of thermal putty would work I would be glad to hear it.
 
I also am unsure why PWR4 is so much hotter than the rest and as I said above I was unable to find any obstructions in the PWR4 area that would cause any issues with contact. However, I dont know if PWR4 being 80+ causes any throttling so I don't even know if its actually an issue.
 
Again, this is all with the GPUFAN 2 profile set to "quiet" in X1 which maxes it out at 30%
 

oninaig

KingEngineRevUp

oninaig
On my second hybrid kit for the 3090 FTW3 after I had the RMA the first one. The first one I had did not have this issue but the refurbished one they sent me seems to have a very hard time keeping the memory junction temperatures under control and this isn't just during benchmarks either. During games like Tarkov or any heavy DX11 game the memory junction temperature creeps up to 104-106 and at that point the RPM readout for the GPU2 fan goes to 0 while the fan actually spins up to 100% in real life until the temperature drops down below 100.

Here's an example: 

I've taken apart and re-assembled the hybrid kit multiple times to verify that the thermal pads are all making good contact. The only thing I can think of is that the thermal pads that are on the copper plate that covers the VRAM surrounding the chip itself are too thick, maybe the unit I got (refurbished) was originally returned with non-factory pads that were too thick?

How thick should the pads that are on the copper plate be? What do I have to do to this card to get its MJT under control for just daily use?

The card itself is all stock, no overclock at all.

Edit: I've seen people place big heatsinks on the back of their cards or smaller fans that blow air directly over the back of the card but most of the people I see doing that are mining 24/7 and I am just gaming with the card at stock settings so I feel like I shouldn't be seeing these temps.

If you're taking it a part multiple times, are you remembering to repaste the memory cold plate?

You have to add thermal paste in-between the AIO and memory cold plate.

 
So I opened it back up and applied thermal compound to the cold plate. Does it have to be thermal paste? 
 
After doing so, I ran furmark 3840x1080 for 20 minutes and I got the following results:

 
This is running the fans at 80% on the radiator and 80% on the card itself (fan 2).
 
If I drop the fans down to where they were on the air cooler (quiet preset), I get the following results:


 
I am using 2mm gelid pads on mem1, 2, and 3 and 1mm pads on the mosfets. I am using the stock pads for the VRM because I did not have 3mm pads on hand.
 
Could this also be because I used compound instead of paste for the contact between the cold plate and the AIO?

When you say "compound" do you mean a thermal pad? I wouldn't have used a thermal pad there, it was clearly designed with thermal paste in mind.

If your memory maxes out below 100C then don't worry about it... If you are running benchmarks and you DON'T see the memor thermal throttling and dropping memory speeds, then I wouldn't worry too much.

90C is acceptable. 100C is questionable since you need some headroom. 105C and memory speeds throttling, definitely a concern.

oninaig

KingEngineRevUp

oninaig

KingEngineRevUp

oninaig
On my second hybrid kit for the 3090 FTW3 after I had the RMA the first one. The first one I had did not have this issue but the refurbished one they sent me seems to have a very hard time keeping the memory junction temperatures under control and this isn't just during benchmarks either. During games like Tarkov or any heavy DX11 game the memory junction temperature creeps up to 104-106 and at that point the RPM readout for the GPU2 fan goes to 0 while the fan actually spins up to 100% in real life until the temperature drops down below 100.

Here's an example: 

I've taken apart and re-assembled the hybrid kit multiple times to verify that the thermal pads are all making good contact. The only thing I can think of is that the thermal pads that are on the copper plate that covers the VRAM surrounding the chip itself are too thick, maybe the unit I got (refurbished) was originally returned with non-factory pads that were too thick?

How thick should the pads that are on the copper plate be? What do I have to do to this card to get its MJT under control for just daily use?

The card itself is all stock, no overclock at all.

Edit: I've seen people place big heatsinks on the back of their cards or smaller fans that blow air directly over the back of the card but most of the people I see doing that are mining 24/7 and I am just gaming with the card at stock settings so I feel like I shouldn't be seeing these temps.

If you're taking it a part multiple times, are you remembering to repaste the memory cold plate?

You have to add thermal paste in-between the AIO and memory cold plate.

So I opened it back up and applied thermal compound to the cold plate. Does it have to be thermal paste? 

After doing so, I ran furmark 3840x1080 for 20 minutes and I got the following results:


This is running the fans at 80% on the radiator and 80% on the card itself (fan 2).

If I drop the fans down to where they were on the air cooler (quiet preset), I get the following results:



I am using 2mm gelid pads on mem1, 2, and 3 and 1mm pads on the mosfets. I am using the stock pads for the VRM because I did not have 3mm pads on hand.

Could this also be because I used compound instead of paste for the contact between the cold plate and the AIO?

When you say "compound" do you mean a thermal pad? I wouldn't have used a thermal pad there, it was clearly designed with thermal paste in mind.

If your memory maxes out below 100C then don't worry about it... If you are running benchmarks and you DON'T see the memor thermal throttling and dropping memory speeds, then I wouldn't worry too much.

90C is acceptable. 100C is questionable since you need some headroom. 105C and memory speeds throttling, definitely a concern.

Oh no I literally mean compound. For the connection between the cold plate and AIO I used Arctic MX-5 which is a compound like you would use on a CPU/GPU core itself.

That's a paste also then. 

Can speak as a FTW3 3080 HYBRID owner, been mining non stop for 7 months now, my temps are 96c-98c at the most, i have yet to change the stock pads, as a side hussle i changed pads for miners on all 3080s, ive done over 80 cards in the last few months, everything from a FE 3080 to a 3090 MSI SUPRIM, the problem with the hybrid is that the vrams only get cooled from the copper surrounding the pump, and they dont get active cooling like they would on an air cooled card with direct cooling to the VRAMs, hybrid card is good for gaming 55c core temps, but for mining not the best, a  air cooled FTW3 with new 12.9 WMK Thermal odessy pads will sit comfortable on 84c when mining 

You would get even lower temps probably with a heatsink similar to this and a 92mm or 120mm fan,

https://www.amazon.com/dp...ncoding=UTF8&psc=1

I have that one on my 3090 and a 92mm fan. It drops the mem junc temp by about 12-14C. This is a waterblock though with a solid backplate. Too bad EVGA likes to put all of those cutouts in it.