ordimans

edgeofsanity

CBSloth
Hey, I'm just wondering what version 3090 ftw3u you had. I'm currently waiting for a replacement 3090 ftw3u due to an unbalanced power delivery on the early version card(nov 2020) I'm rmaing which had me pulling 85+W through my mobo pcie slot. With the 500W xoc bios I was only able to reach 448W max board power draw.

Did you go through the rma process for one of the newer revisions of the card or did you purchase a 3090 more recently?

 
This is an RMA card, I think I did it back in February of this year. I also had a Nov 2020 card for my original. Fan controller broke on the original right before my block came. Serial on the new one starts with 2114.
 

ordimans

edgeofsanity
I picked up where the OP left off on this today. I had an Optimus block on my ftw3 3090 and needed to return it to stock today. I have a strix that is a much better overclocker that is replacing it now that I have another Optimus block.

Anyways, I've repadded several 3080s/3090s already and I use calipers to measure factory pads before replacing. Memory is 1.9mm-2.1mm with an average of 2mm. This includes backside, no difference. Always measuring the thickest part of each pad. Vrms are 2-2.3mm with an average of 2.2mm for the thickest parts. I suspect the vrms were the OPs problem.

I use thermalright oddessy 12.6wmk pads, these work great but are stiff and don't like to squish so you cannot use measurements for squishy pads.

For my repad I used 2mm on the front and back mem, 2mm on the vrms, and 1mm on the mosfets. For the VRMs you MUST use a bead of thermal paste or even better if you have thermal putty. You will not get good vrm contact if you don't do this. I used arctic mx4 for this. A 2.5mm pad is too thick. The paste fills the tiniest gap you'll have. This also let's the mosfets make contact with the pad unlike the OP where they have a gap. Core was pasted with TG kryonaut.

I'm still testing but results are great so far. Looping heaven in my open air testbench with the xc3 bios and power sliders maxed(~520w)core temps top out at 70c, mem junction 78c. Hottest vrm reading is 69c. Mining eth tops out mem junction at 92c. I can easily drop this 10c by pointing a fan at the backplate.

Hello bro
 
Can you share a picture ?


I put thermal putty from TT Global on VRM
ANd i used gelid extreme for others, but i am at 108C during mining (with hybrid kit)....
 
On the backplate, should i use thermal putty too ?

Sure. I didn't use putty on the backplate or anywhere else. Just the 2mm and 1mm thermalright pads and that bead of arctic MX4 over the VRM pads.
 

 
Not the best pic but there isn't much squish on that VRM pad. 

 
One of my mining rigs, GPU #3 is this one. 90F ambient, fan blowing directly at the backplate. GPU 1 an 2 are stock, GPU 0 was a repad that needs to get redone. Mem temps are good but core contact isn't great so its 10C hotter than the others. 
 

Wahoo
 
No pad no putty on backplate ?
 
But you have airflow
Maybe i should test to put air back, than hybrid kit.

2mm pads on the back memory, thats it.

edgeofsanity
 
2mm pads on the back memory, thats it.

Thanks bro
Did you put just back VRAM or you put back of mofset too ?
 
I put 2mm Odyssey on back vram.
Still 110 after some minutes on trex...


I need help, not possible i cannot get good temperature on FTW3 as original.

why the fck did i just spend 500 on thermal pads for 3 cards? 
 
i'm using fujipoly 1.5 and 1.0, 1.5 on everything did not work obviously because of the dimensions mentioned before so i'm going to stack 1.0 and 1.5 on everything. 

So GPU temp and VRAM temps are 60 and 80 respectively but my problem now is that I'm getting a hotspot temp of 104 that's throttling the card. i can't locate the sensor so I don't know what the problem is. great.

I only add a heat sink to the backplane.
Run 2077 Game VRAM Junction temperature 85°C



 

do the heatsink reduce your temps alot?

Just repasted/repadded my 3090 FTW3 Ultra according to the info in edgeofsanity's post. I appreciate you actually using calipers and reporting actual temperatures from mining (along with power consumption) - too many posts here are just "I put new pads on and now my temps are better/worse lol good luck finding out what I did." In case I can save someone else some hassle in replacing the thermal pads on their EVGA 3090 FTW3 Ultra, here's all my info from the process:

I got 1mm and 2mm Thermalright Odyssey 12.8 W/mK pads. Replaced all mem pads (front and back) with 2mm, both VRM pads with 2mm + thermal paste (also Arctic MX4 but that's just because I have like 10 tubes from CPU coolers), and both MOSFET pads with 1mm pads. I also previously had a couple aluminum heat sinks just sitting on the backplate with no thermal pads or any TIM, 4x 20x20x11 and one big 150x70x10 (rough dimensions, can't remember the exact numbers) - those are staying on, because they did make about a 4 or 6C difference in memory temperatures.

Unfortunately I didn't control for ambient temperature and took haphazard data, expecting to get the same results as you, and I was only concerned with memory temperature to begin with. But here are my results: Before repadding, in a 28-30C room while mining ETH, 125 MH/s @ 350W and 100% fans, the GPU temp stayed around 61C, hotspot 68-75C, and memory junction would hit 102-106C. When ambient temperature peaked (probably above 30C), the memory would bounce off the thermal limit at 110C but generally MH/s would stay above 115. With the air conditioning running and the room at around 74F/23C, memory would sit around 96-102C. When I replaced the pads, putty, and GPU paste, I was excited because the stock paste was thoroughly dried out (even though I ordered and received this 3090 directly from EVGA, new, only about a week ago). Normally you have to "wipe" thermal paste off; this stuff was coming off in chunks. I literally took the GPU PCB and held it upside down over a trash can to knock off some small chunks of dried thermal paste that had stuck to the board from chipping it off.

Anyway, repadding didn't work. My temps are significantly higher now. In a very cold room 72F/22C, my GPU temp is 56C (probably about what it would've been at this ambient temp). Hotspot is 70C - that might actually be higher than what it was before under air conditioning, because it was around this temperature when the ambient was more than 5C higher. And when I started mining, the VRAM quickly shot up to 108/110C and thermal throttled. Now averaging about 115 MH/s @ 320W - 10 MH/s and 30W lower than it should be under this low of ambient temp, with memory temperature 12C higher stock. What a pain. But, I want this stuff to work correctly because I'm sick of having to run my fans at 100% to mine when my PC is 3 feet from my bed. This is what I plan to try:

1) Right now, re-seating the GPU. Make sure everything is lined up correctly, visually check that the pads have impressions from the VRAM/MOSFETS/VRM (well, check that the thermal paste on top of the VRMs was making contact with the heat sink)
2) If re-seating doesn't work, I just ordered a 0.5mm Thermalright Odyssey 12.8 W/mK pad. Once it arrives, I'll cut out pads to add on top of the existing VRAM pads for a total of 2.5mm pad height (2mm installed now + 0.5mm added).

If adding the 0.5mm pads to the memory works out, I'll be curious to know how you were able to get such good temperature reductions. I used the same pads as you, have the same card, installed them the same way, ran at the same MH/s and same wattage, and got wildly different results. Not 4 or 5C difference; more than 20. Guessing this is just user error on my part during the pad replacement, but I'm also putting down all my thoughts for the small handful of people like me, wanting to repad their cards. I'll post an update once I figure out what's going on.

Just reseated the cooler on the 3090, no change. GPU 57C, hotspot 72C, mem 110C and thermal throttling. When I took the cooler off to reseat it, the imprints of the memory chips opposite the video outputs were lightly visible, and the memory chips on the video output side barely left an imprint at all; the diamond score pattern from the protective film the pads came with wasn't pressed out at all. Looks like 2mm pads aren't thick enough, at least for my particular 3090 FTW3.
 
I'm wondering if EVGA's cooler manufacturing is just that sloppy, that some cards require an extra 0.5-1mm of thermal pad thickness to make proper contact. Maybe that's why they use thermal putty for the MOSFETs and extremely soft and slightly-too-thick thermal pads for the VRM and memory; just some extra leeway to cover the large tolerances in the cooler/PCB manufacturing. I'll post an update once I receive the 0.5mm pad and get a chance to repad with 2.5mm (2mm + 0.5mm) pads.

I too question the loose tolerances. I only messed with changing the back vram pads. I mine overnight when not gaming/working, so I wanted to not burn this up early. Also, I don't like running fans that high. Its just too loud in the living room. 
Temps prior to messing with things.
Mining Eth +200 mem clock gave me 104-108C Junction temps at 85-100% fan.
New World queue screen was at 85-90C Junction.
 
Tried Gelid GP-Extreme 2mm first, but temps were way worse. The pads compressed a bit too much and didn't rebound, so contact wasn't great. I had to underclock ram to -200 to keep it from staying at 100% fan and throttling. Mining Junction would stay at 106-108C with peaks to 110C.  New World was ~90 -95C at the queue.
 
Switched to Thermalright Extreme Odyssey 2mm. Those pads didn't compress as much, but contact STILL didn't look right. Temps improved over the Gelid, but didn't get back to what stock was. I was able to mine with no overclock again, but fans just stayed above 90% all night.
 
Move to Thermalright 2.5mm. Now I've got contact again (and an ever so slight bulge on the backplate) and temps were back to what they were prior at the start of this whole thing. 
 
Taking what I learned, I decided to just go ahead and buy one of those large heatsinks from Amazon for $20 (a pair of them actually), and just lay it on the back. While mining, temps went down immediately 4-6C, and the fans dropped to ~50%... Absolutely amazing to be honest. I added some pads between the backplate and the new heatsink to help out (since the backplate isn't completely level)
 
I now have a Noctua fan just resting on the massive heatsink. Temps are completely in check. I was able to push the mem clocks to +1500... yes.. 1500. Gained +15gh/s over what I had before (peak I was comfortable with was 124.8gh/s)   Temps are 98-104C with fans at 50-70%... Much cooler, much quieter, and as a side effect I didn't think I'd see... less power draw by 20w at exact same power/clock settings. 
 
In the end, this makes complete sense since there's no active cooling on the backplate. Original pads probably would have been just fine, but w/o some place for that head to escape, it just soaks into the plate and stays there. Just changing the pads on the back will yield you no gains. You NEED to add some extra cooling help. 

edgeofsanity

CBSloth
Hey, I'm just wondering what version 3090 ftw3u you had. I'm currently waiting for a replacement 3090 ftw3u due to an unbalanced power delivery on the early version card(nov 2020) I'm rmaing which had me pulling 85+W through my mobo pcie slot. With the 500W xoc bios I was only able to reach 448W max board power draw.

Did you go through the rma process for one of the newer revisions of the card or did you purchase a 3090 more recently?

 
This is an RMA card, I think I did it back in February of this year. I also had a Nov 2020 card for my original. Fan controller broke on the original right before my block came. Serial on the new one starts with 2114.
 

ordimans

edgeofsanity
I picked up where the OP left off on this today. I had an Optimus block on my ftw3 3090 and needed to return it to stock today. I have a strix that is a much better overclocker that is replacing it now that I have another Optimus block.

Anyways, I've repadded several 3080s/3090s already and I use calipers to measure factory pads before replacing. Memory is 1.9mm-2.1mm with an average of 2mm. This includes backside, no difference. Always measuring the thickest part of each pad. Vrms are 2-2.3mm with an average of 2.2mm for the thickest parts. I suspect the vrms were the OPs problem.

I use thermalright oddessy 12.6wmk pads, these work great but are stiff and don't like to squish so you cannot use measurements for squishy pads.

For my repad I used 2mm on the front and back mem, 2mm on the vrms, and 1mm on the mosfets. For the VRMs you MUST use a bead of thermal paste or even better if you have thermal putty. You will not get good vrm contact if you don't do this. I used arctic mx4 for this. A 2.5mm pad is too thick. The paste fills the tiniest gap you'll have. This also let's the mosfets make contact with the pad unlike the OP where they have a gap. Core was pasted with TG kryonaut.

I'm still testing but results are great so far. Looping heaven in my open air testbench with the xc3 bios and power sliders maxed(~520w)core temps top out at 70c, mem junction 78c. Hottest vrm reading is 69c. Mining eth tops out mem junction at 92c. I can easily drop this 10c by pointing a fan at the backplate.

Hello bro
 
Can you share a picture ?


I put thermal putty from TT Global on VRM
ANd i used gelid extreme for others, but i am at 108C during mining (with hybrid kit)....
 
On the backplate, should i use thermal putty too ?

Sure. I didn't use putty on the backplate or anywhere else. Just the 2mm and 1mm thermalright pads and that bead of arctic MX4 over the VRM pads.
 

 
Not the best pic but there isn't much squish on that VRM pad. 

 
One of my mining rigs, GPU #3 is this one. 90F ambient, fan blowing directly at the backplate. GPU 1 an 2 are stock, GPU 0 was a repad that needs to get redone. Mem temps are good but core contact isn't great so its 10C hotter than the others. 
 

So the 1mm MOSFET pads are the skinny pads on the outside of the VRM, correct?
And the puddy goes on top of just the 2mm VRM pads, correct? (or puddy on the MOSFET one, too?) 
 
Thank you.