Hi all. I decided to add my experience in EVGA 3090 FTW3 Ultra maintenance. A little history about the card:
Spoiler
- Why 3090 - I chose it because I wanted something universal for working with graphics and games, and here we have 24GB of memory and a GPU in performance approximately equal to 4070ti (except for tests with DLSS 3.0, which the 3090 does not have.).
In addition, the price for it on the secondary market in my country is lower than for the new 4070ti
- Why EVGA? Simply because Strix or MSI did not fit in the length of my case, and there was no money to upgrade the case.
- I'm the third owner. The first one was in Germany, how the card was used is unknown. The second one is mainly gaming and a little bit of mining.
- I bought it in the spring of 2023, it took me half a year for 3 service attempts, and another half to write this post.
- I'm not going to mine on it. (as it turned out later, everything was OK with the chip, the textolite didn’t burn out, which means it wasn’t overloaded on a long-term basis, everything was OK with the memory, so it was a good purchase)
close
Because I am the third owner, the card is several years old and knowing its temperatures and operating modes, it was decided to service it.
For tests, I mainly used Cyberpunk 2077 in 4k and ultra settings, sometimes turning on RTX to completely put the video card to work. I won't be able to gather test stats during this time, so I'll rely on my memory and notes.
The card was immediately
undervolted and runs at
1830MHz at 0.825V and still has those temperatures.
The temperature on the GPU in my cramped ZALMAN X3 case did not rise above 74 degrees, the memory - about 90 degrees. The previous owner posted videos of tests and had no more than 70-71 on the GPU.
I hoped that after servicing I would also get temperatures up to 70, this would allow me to adjust the rpm curves of the coolers and achieve a quiet and efficient PC. There were a total of 3 attempts to service the card and the best result was with such consumables.- GPU - thermal paste with phase change Honeywell PTM7950, slice 0.2 mm thick.- MOSFET - liquid thermal pads Honeywell HT10000 10W - about 10ml (HT7000, Laird 607 also good)- VRM- Thermalright ODYSSEY Pad 12.8W - 120x20mm, 3.0mm thickness- Memory - Thermalright VALOR ODIN Pad 15W - 95x50mm, 2.5mm thickness For those who want to read the full story, there will be spoilers below. ATTEMPT #1
Spoiler
I watched enough videos about copper plates and decided to try this "OP thing". Ordered the following consumables:
- copper plate for memory around the GPU 1.8mm and 1.8mm copper separate plates for memory on the reverse side.- on the GPU, thermal paste with phase change Honeywell PTM7950, slice 0.2 mm thick.- Thermalright ODYSSEY Pad 12.8W - 120x20mm, 1.0mm thickness on MOSFET- Thermalright ODYSSEY Pad 12.8W - 120x20mm, 3.0mm thickness on VRM- large tube of MX-4 and MX-6 paste When disassembling the card, I saw that silicone was leaking from the original pads and many elements in the card were covered in fine dust, which was potentially dangerous. So it was definitely necessary to carry out maintenance. From the good - the textolite was light emerald in color, which meant that the GPU was not subjected to strong and prolonged loads.
I isolated and limited the dangerous areas with insulating tape. Between the plates, chip and radiator I smeared MX-4 paste.
In the process, I
failed both attempts to carefully apply a slice of Honeywell PTM7950 (at that time I did not yet know about all its properties and that even slightly torn it would have melted and spread as needed at 50+ degrees), so I had to smear MX-6 on the core
After assembly, the clamping of the plates and gaskets seemed to be good. Temperatures on the memory dropped to 80 degrees, but on the GPU they rise to 80+. This didn't suit me. Possible problems in my opinion:- due to the fact that the temperature from the memory was now better transferred to the radiator, it did not have time to dissipate as much heat and the GPU had to take on part of the heat.
- the problem could be that the GPU has MX-6 paste, not a phase transition.
- the copper plate is too thick and the GPU does not have good contact with the heatsink. (nope)
It was decided to try again.
close
ATTEMPT #2 is the most unsuccessful
Spoiler
I bought a 1.2mm thick copper plate and VALOR ODIN Pad 15W as a keepsake if things don’t work out with the plates.
consumables:
- copper plate for memory around the GPU 1.2mm and 1,8mm copper memory strips on the reverse side.- on GPU, MX-6 paste- Thermalright ODYSSEY Pad 12.8W - 120x20mm, 1.0mm thickness on MOSFET- Thermalright ODYSSEY Pad 12.8W - 120x20mm, 3.0mm thickness on VRM- large tube of MX-4 and MX-6 paste- Thermalright VALOR ODIN Pad 15W - 95x50mm, 2.5mm thickness for memory After disassembly. I saw that the pressure was good everywhere. MX-6 on the GPU simply spread out in a couple of weeks, the chip was almost dry. And the whole board was in MX-4.
In the process of replacing the 1.8 mm plate with a 1.2 mm one,
everything was in MX-4 paste - me, my desk, room, cat. I also saw that with a 1.2mm plate there are too large gaps that cannot be compensated by the paste,
besides, it will leak.
It was decided to wash off all this crap and install regular VALOR ODIN pads After assembly and tests. I didn't see any improvements. The same 80+ degrees in the core, only the memory again became about 90 degrees. Possible problems?-Most likely this time I did not achieve good GPU clamping,
-Plus this time I didn't have phase change thermal paste.
It was decided to go for the third round. At this time, I tried not to overload the card until new consumables arrived.
I also bought an aluminum radiator, sawed it and glued it to MX-4 paste on the backplate. It dissipated heat well and was very hot, but according to tests with or without it, the difference was 1-2 degrees, so I won’t say that it made much sense.
close
ATTEMPT #3 After all of that, I got temperatures approximately the same as when I bought the card,
it was up to 74 degrees, it became up to 76 on GPU but the temperature on the memory is lower and does not rise above 90. Thats meant I'm back to the starting point. But now at least there is no risk that the card will die due to dust soaked with silicone.
In total, for 3 attempts I spent about $130 on consumables and a lot of time. Conclusions:- Don’t think that you are smarter than the manufacturers and don’t think that standard thermal interfaces are bad.- It only makes sense to install a copper plate if you are going to mine.- brake cleaning fluid is your friend Now I use it in games in 4K resolution with a limit of 60 frames. I adjusted the curve with the cooler and the curve of all coolers in the case.
So that the PC remains quiet under load. I’m not going to go into 4th attempt, I’ll try to hold out until the 50th series of Geforce or the price drop for the 40th series.
This was my experience. Thanks everyone for reading and good luck! Bonus. Pictures with pads sizes. (taken from this topic)