The memory bank temperature difference between banks bothered me so I decided to order up some thermal pads and see what could be done.  I was surprised by what I found.  The coolest memory bank on these cards is the only one from the factory to have a thermal pad connection to the heatsink assembly.  The other two, much hotter banks are uncoupled and rely on the good grace of the monster anodized black plate that seems press fit attached to the back plate.  Maybe it's removable but that is an investigation for another time.  I just wanted to see what I could do with a quick stab at the configuration.  Before and after temps heatsoaked, using the same work unit for f @ h GPU, same 65% all card fans speed.
 
Factory configured

 
 
After with the addition of some 1mm thermal pads (fuji poly 11) and a thin layer of Kryonaut added atop all thermal pads, factory included.
 

 
Not a huge difference and I have my doubts this is going to improve anything other than required fan speeds to keep the memory out of the red zone but I am kind of unpleasantly surprised by there being no pads to thermally couple two of three banks of memory for obvious reasons.  It is worth mentioning that with the lay out of the heat pipes and the location of the memory banks the coverage for 12 o'clock and 3 o'clock ends up being an effective 60% direct coverage guesstimated. (edit: the following sentence added) They are coupled to the HSF assembly via heat pipe for 12 and heat pipes for 3 o'clock.
 
I am disappointed by the cooling solution as I feel it leaves performance on the table or at the very least will lead to my needing to run fans higher than I otherwise would have to in order to keep my memory within design spec while clocked up.
 
Now I have to figure out what I am going to do with this thing.  I am definitely going to look for tear downs of the MSI and Asus options... maybe see about getting a Palit if they are even distributed in N.A. If EVGA releases a hybrid card while I am still within the Step-Up period, is that a sufficient enough upgrade to qualify?

I didn't think to check stock on my 1.5mm fujipolys and didn't have enough to do both banks.  I have an order coming in Monday. I am going to check thermal paste squeeze out @ the added pads and swap in for 1.5mm if appropriate.  I'll take some pictures while I am at it, all I bothered to snap was the cleaned up TU102.  Kudos to assembly on the GPU die paste job.  It spread and looked a lot like Kryonaut; the temps are pretty much the same upon reassembly.  Just the right amount of squeeze out around all around the die, too.  Nothing really to be gained there without LM.  I do like how simple removal/disassembly of the HSF assembly is, and the rest of card, too.  By far the easiest video card I've ever dealt with.

scorpion73
Hello
i didn't understand well the work you have done but i am interested in that. Can you tell me if you added pads to the back plate between the same and the heatsink in the areas where memory banks are located? Or did i understand bad? Can you post some pics of the board in the parts where you put the pads?
Thanks

I removed the heatsink fan assembly and between it and the front plate (forgive the incorrect terminology) on the GPU and memory side of the card I laid a strip of 1mm thermal padding directly above the memory at the 12 o'clock and 3 o'clock poisitions as displayed by the EVGA Precision X1 graphic. 
 
Next go around, probably November 19, or 20th I am going to disassemble again and survey thermal pad compression, Kryonaut squeeze out on all thermal pads, and asses if 1.5mm fuji poly pads would help out further.
 
I apologize for not having taken pictures this go around but will take them next time I have it apart.
 
Near as  I can tell, the back plate and front plate aren't coming off without a press.  Although I removed all of the backplate screws, removed the fan shroud and fans, I did not notice user serviceable fasteners that were holding the front and back plates onto the card assembly.  Neither were even close to budging from relative placement to the card. Near as I could tell, EVGA set these cards up to have the entire HSF assembly retained by only the four spring screws common to the the GPU die.  I had all of the backplate screws removed owing to lack of familiarity with modern GPU design but I do not believe that to be necessary on the FTW3 Ultra; just remove the 4 spring screws and gently twist and pull should be all that is required to separate them.  Just make sure you loosen up the fan and RGB plugs before unscrewing.  They are quite snugly connected and jerking them off would definitely damage them.  The RGB plug is pretty easily accessible and can be ignored, more or less, upon reassembly but you are going to have to align the three fan plugs into their sockets carefully or you won't be able to seat the HSF assembly to the card and the cooling fins may cut through the jacket if you try forcing the issue.  There is very little clearance on the fan header side of things.
 
 

I was quite disheartened to see this omission in production. The placement of the heat pipes only allows partial coverage of the "North" and " East" memory banks. The east side may only be about 25% direct coverage because of the bend of the heat pipes up and away. Some design tweaks of the hsf assembly and much more coverage could have been had and no clearance issues would be present and especially for the east bank, I can't see how such a modification would affect heat pipe performance.

My concern with going with 1.5mm is that it may cause the heat pipes to flex and cause the contact of the vrm heat pad to be lost. I want to think, though, that more heat can be safely and effectively pulled off of the N and E memory banks so I am going to try it if at all sensible.

As it stands, my two GPU temperature measurements are now never more than 2c off from each other and I can get the same temps and clock speeds with a +60 offset vs +100 and fans at 50% vs 65% pre mod. Well worth the effort but I agree with you and GTXJackBauer that these should have come from the factory this way. I bought this card more for the hope of lower fan speeds than any designs on better overclocking.

It's odd to see the temp variations on this card, Some people get 80c on the memory, others get 75c, etc. all running under very similar conditions even in the same cases. and at stock/oc. Mine for example doesn't really go over 70c even under the heaviest OC i could put on it during the typical benchmarks. and during gaming it barely ever hits 65c. and i have terrible airflow in my Evolv X case, I'm using old fans that probably do very minimal. (My GPU is using stock fan-curves, It's very quiet and rarely do all 3 fans even get used during stress)
 
but other people get 80c just stock playing a normal game even with higher fan curves 

rchiwawa
The variation is interesting for sure... I run mine at a set fan speed. It was, before the addition of pads 65% all around but now I get away with 50%same peak thermals and clocks. I had to adjust my core offset lower to keep it from bugging out via driving the core faster than it is actually stable. I am curious is your copy was issued with thermal pads where mine wasn't. I haven't used the stock curve since I am noise averse at a certain point but I am going to give it a go and see what happens when I get home. Your temps seem good. Do your memory banks, east, north, and west all run about the same temps? If not, what is your typical differential?

In X1 the west (Left) memory bank mostly always runs at 30c + lower than the north and east banks. In a 850/125 oc running the card around 2070/2055 clock during the benchmark, I hit 70c on the top bank. the east one stays very close to the top. However the left one still was down at around 40c. I checked the fans during the benchmark at it was around fluctuating around 50% usage of all 3. in some benchmarks though they don't really spin even at 50% and i get similar temps.
 
The other odd thing i noticed was that allot of people can get 1000 - 1100 oc on their memory... I can't even get 900 stable on any benchmark. I have to push it down to 850 on the memory. So maybe some how my memory isn't getting as hot as other's are, but i also can't OC any-where near them... which is counter intuitive sounding...
 
The gpu clock I run at 125 which seems to be my max... I was running 130 for earlier benchmarks but now all of a sudden that crashes TS extreme. Maybe I got some stuff running in the background now, not sure.
 
I know every card and every overclock is different but considering my temps are lower, I was suprised to find i can't oc it as far as the majority can. To be fair though my card hit 2100 with just the 850/130 oc. But some people are bench marking at 1100/150 and 1000/175, That's far from where I can get.
 
I have tried cranking my fan speeds to max and i stay in the low 60's through all benchmarks but again, didn't help my oc's go any higher because temp isn't a issue for me. I'm happy with low temp's don't get me wrong, but it kind of sucks to never really hit 2100mhz on this thing.

CraptacularOne
Adding thermal paste on top of thermal pads is just adding another layer for any heat to transfer through and lessening heat transfer to the heatsink. 
 
As for the memory temps, even the 73C range you posted is well within spec for GDDR6 which is rated to 95C. Not sure why people obsess over this as even if it were running at 85C it would still be fine. 

When memory is warmer than a GPU, that's of a concern.  Being 10c away from threshold, is also of a concern and not comforting. 
 
Heat kills longevity.  The longer they stay cool, the longer they'll last but to each their own.

KaptCrunch
yes heat is a killer electronics, so the cool obsession is needed for the heat is like no oil in your motor, things seize up, no data flow when hot
 
next level of cooling will be T.E.C. (peltier)  

I know someone who does that.  TEC cooling aren't that efficient as they use up a lot of energy to cool.  I'd rather stick to ambient liquid cooling which is fine.

scorpion73
which thermal pads are better? 1mm or 1,5mm?

It's not better.  That's based on getting the right size so its a decent fit.  If it's too big or too small, it's neither good.  Got to find the closest fit.  What you should be looking for is thermal conductivity rating of the pads.  The higher the number the better heat transfer. 
 
Fujipoly to me are the best thermal pads you can get but they ain't cheap that's for sure.

scorpion73
Gelid thermal pads have a highest conductivity than Fujipoly.
12w/m-k vs 11w/m-k.
But i was asking because the user that did this try has already choosen the thickness of the pad, so i was asking if he chose 1mm or 1,5mm in order to see which one fits in order to make a correct contact.
 

Oh I see.  You were asking them a question to see which was the best fit.

As for Fujipoly, they got 17's.  Yup, I've used them on a GPU before and my older M.2.   

GTXJackBauer
As for Fujipoly, they got 17's.  Yup, I've used them on a GPU before and my older M.2.  

He mentioned FujiPoly 11 with which he obtained that decrease, so as i don't find easy the Fujipoly here in Italy where i live but i find easy the Gelid i was comparing and the consideration is that if with 11 he obtained 10°C less (more or less), with 12 i can stay relaxed i can obtain the same.
Then if there is another pad with 17 i think is still better :) but i cannot find here.
As soon as the user can indicate the tickness of the pad between 1mm and 1,5mm i will proceed to place the order so that i also can do this operation when i will have success in ordering one of these boards (if EVGA decides to make them available instead of let them appearing like ghosts....)

Yes but he told he should confirm if it was correct 1mm or better 1,5mm.
This is very important.

You can post a pic of the board for how it is now without unmounting, for us to see how the added dpads are in contact with heatsink in the memory area.
Why are you talking about flexing heatpipes in the "east" memory bank?. The dpad is not in contact there by leaving them as they are?