Yeeeeeah she definitely got toasty, but she passed.  These are by far the highest temps I've seen to date on the card, but given the clocks/voltage/power draw it was running at throughout the testing, I'm not surprised:
 
https://www.3dmark.com/3dm/101305777?
 


 
I just find it fascinating that while the GPU voltage occasionally deviated from 1070mV to 1065mV, the GPU core clock was just flatlined at 2970MHz after the first 15 or so seconds of the first loop (where it started at 3000, then dropped to 2985 before settling at 2970MHz).
 
3DMark single test run result link for these settings: https://www.3dmark.com/pr/2601292 I'm pretty happy with being north of 27k with a "stock" PBO 5950X that's 'only' running 1800MHz Infinity Clock, and the GPU being air-cooled.  But at this point, these settings aren't even 'really' an undervolt, it's just using an undervolt curve to control the GPU clock & voltage instead of letting NVidia's GPU Boost muck it all up with random fluctuations in frequency & voltage.
 
Don't know that I'd run these settings for gaming necessarily even though they've been proven at least mostly stable, just because of those temps being not really to my liking... coming from a Kingpin 3090 that I had under an eVGA waterblock, and rarely seeing above 55C, I don't dig hitting over 65C - especially when this card runs at under 60C stock 99% of the time.
 
Speaking of stock/other settings I've tested, here's some screenshots of those:
 
Stock/Stock with sliders maxed for comparison of temps/power draw, but keep in mind this is a single run, cuz I'm not running a full stress test on stock/stock clocks cuz what's the point:

 
2790/2775 settings, with the VRAM at stock, I did this for an A to B comparison vs the above, but with a full stress test, so temperatures are steady state in this vs the above is a single run - funnily enough the temps are still lower, despite it being a 34 minute test vs a single run at the stock settings above lol:

 
2850/2835 settings:

 
 

Opolis
Interesting results. I have a 5950X w/ MSI 4090 Suprim X. My best Port Royal was 26,808.  The avg GPU temp was 64C (air cooler).
I briefly played around with undervolting and overclocking but just run it at stock or reduced settings for daily use.  This GPU does really well whether you reduce the power draw with a curve or just power target.  I found I can go down to 70% PT, still run 26-2700MHz range and use up to 100W less with max temps in the low 50's.
 

The 2775 undervolt only hits 55C in Port Royal with ~410W power draw.   But the scores I get are kinda wonky, at "stock" I get around 25.5k give or take, I think it's a combination of this Windows drive being on it's way out (as in it's dying) and other 'issues' with my 5950X clocking oddly.  Or maybe that's normal, I dunno.
 
Best stable score I've hit was with the 2970MHz @ 1070mV "undervolt" OC from above, https://www.3dmark.com/pr/2601371 just a smidge over 27k.  That test preceded the stability test above that verified that running at 2970MHz @ 1070mV with the +1500 mem was 'stable'.  I haven't tested it enough in games etc to dub it truly stable, I'm kinda waiting to do that until I can swap to a native 12PinVHPWR connector PSU so I'm not using the adapter anymore, since it does draw all the way up to the full 600W just about, according to the screenshots showing the "GPU Power Rail" draws.
 
I set up a fan as 'exhaust' on the 'flow through' portion of the card pointing upwards towards my CPU radiator to push air through the card regardless of temp, lowered my idles by 4-5C with the GPU's actual fans at 0% (cuz it's idle so anything below 35C I have the fans programmed to shut off). 
 
Coming from a liquid cooled 3090 Kingpin to this has been interesting thus far, obviously the 4090 is much more locked down, but the temperatures being so much cooler than the Kingpin ran when it was on the AIO before I put the HydroCopper on it, pretty impressive.  Only seeing over 60C when I'm pushing the card to really high clocks, if I'm using more 'normal' settings with an actual undervolt & thusly lowered power draw, I never see 55C, let alone 60C.  And the VRAM & HotSpot scale downward with the core temp pretty linearly.  Hotspot is almost always about 8C higher than the core readout, and depending on what fan speed I'm running, the GDDR6X is either running at the same temperature as the core readout, or 2-4C hotter, so the Gigabyte cooling solution is definitely living up to it's reviewed "pretty good memory cooling". 

Hi Dabadger84,
EK Blocked 13900K, DDR4 3600 CL14; 4090 suprim liquid X With EKblock;  PR 26282 on undervolt , TSE 20055 GRX , 13178 CPU
I run .975 @ 2820 +1120 mem for the daily , It will game all day long at 364-388w and keep my loop temp 29-30C for the long run and does not start heating up the room , I have 3 x 360s  shoe horned into a O11D , 2 SE and 1 PE , EK rads with Noctua , Lian li and Vardar fans running about 60% W/ CPU temp hits 80 they ramp up Noctuas to 100%...
My card's top score was 28500 on PR but that was starting to artifact, my mem does not like < 1300+ , I can run a stress test successfully at 3040 +1280mem but it starts at 45C and keeps climbing and heats up the room so ambient keeps climbing and levels out at about 51C , HS 62-4C but my memory stays cool, probably why it wont overclock much , OC forum says to use insulating tape on the memory so it gets in the upper 70s, mine is usually in the 40s . I have though about trying to drag it up to 2850 at the .9750 but its pretty happy where it rides at the moment .

kraade
Hi Dabadger84,
EK Blocked 13900K, DDR4 3600 CL14; 4090 suprim liquid X With EKblock;  PR 26282 on undervolt , TSE 20055 GRX , 13178 CPU
I run .975 @ 2820 +1120 mem for the daily , It will game all day long at 364-388w and keep my loop temp 29-30C for the long run and does not start heating up the room , I have 3 x 360s  shoe horned into a O11D , 2 SE and 1 PE , EK rads with Noctua , Lian li and Vardar fans running about 60% W/ CPU temp hits 80 they ramp up Noctuas to 100%...
My card's top score was 28500 on PR but that was starting to artifact, my mem does not like < 1300+ , I can run a stress test successfully at 3040 +1280mem but it starts at 45C and keeps climbing and heats up the room so ambient keeps climbing and levels out at about 51C , HS 62-4C but my memory stays cool, probably why it wont overclock much , OC forum says to use insulating tape on the memory so it gets in the upper 70s, mine is usually in the 40s . I have though about trying to drag it up to 2850 at the .9750 but its pretty happy where it rides at the moment .

See that's CRAZY to me, that my load temps with the undervolt are within spitting distance of load temps on full coverage liquid with an OC.  That's something we haven't seen in ages if ever on GPUs I don't think.
I do think I'll eventually end up putting the card under a full block, simply because I want to see if I can push it further, and how cool I can run it.  My old loop for the 3090 Kingpin had a P360 (regular sized) radiator after the 5950X in the loop then an EK XE480 (60mm thick) radiator after the GPU in the loop, and that 480mm had some HEAT comin' off it when that Kingpin was under heavy load sucking ~500W or so.  I can imagine it would fair better & do a good job cooling this 4090 as well.
 

Opolis
I'm also not used to the cool temps...always battled the heat with air cooled cards.  I'm running the Cablemod cable for EVGA PSUs with their 90deg connector.  No issues at all and this GPU is at ~100% load running Folding at Home or similar projects just about the entire time I'm not using it for gaming.  The 90 deg adapter is barely warm to the touch.
For fun I went for 3,000 MHz and saw 480W power draw, which I think is the bios limit on this model. http://www.3dmark.com/pr/2576675
Probably have a bit more in the memory if I had spent more time on it. Your score was better which is interesting.
OC not worth it for my use but still a fun exercise.
 
I'm seeing similar to you on stock settings, 25.3K in Port Royal @ 2760MHz, ~350W.
5950X is always on stock settings with a 360 AIO, ram at 3600MHz, Gigabyte X570 Aorus Master.

I'm currently doing stability testing on my new 7800X3D setup, but I'll see how the score compares once that's done... haven't even had the hardware installed for 8 hours yet & I'm doing curve optimizer -30 all core just to see if it's good or not lol so far it's running at ~4950-5000MHz set like that, and 'only' hitting 76C peak on a few cores, which considering I'm on an AIO (granted, it is THE best AIO tested by most review sites, but it's still an AIO) is pretty good... the craziest part for me about the 7800X3D so far is the power draw.  It's insane that something this fast & powerful is only drawing 83W under all core load.  I'm fixing to VERIFY that because I seriously don't believe it.
 
From the wall under load my power plug readout is saying slightly more than 200W, I know the system (because of the amount of fans I'm running, currently 15 EK Vardars) and other hardware, idles around 120W or so... at least that's what I remember, I'm waiting for the stress test to finish so I can check it... 
 
125-135W idle, so it's drawing about 65-70W more under full load, and that's not accounting for the backwards "extra power draw" the PSU is doing in comparison to what's actually being fed to the CPU... my PSU is a Platinum rated one so it's at least 89% efficient all the way through the curve, which means we can pretty much say the HWInfo readout is accurate saying there's a ~20-25W power draw at idle from the CPU & a ~83-85W power draw from the CPU under load. 
 

mech9t5
Hey Dabadger, see if this will work...a different way of undervolting, just for the 4090:
 
https://www.youtube.com/watch?v=WjYH6oVb2Uw

I actually use this exact method, flattening it etc, that's how I got the undervolt/OCs I posted in this thread.

This one needs a bit more testing but it did pass stress testing & some other checks, and I like that it maxes out in the 380W draw range:
 


 
2700MHz @ 950mV with the GPU fans locked at 80% (because that's what they're at when I'm gaming so I wanted realistic testing) - which means my temps could've been lower if I ran them full throttle.  I'm wondering if I can go even lower on the voltage while still hitting that speed, that'll probably be what I test next.  I'm trying to see how low I can go, out of curiosity, while still maintaining around 2.7GHz boost clock. 
 
Also found out my new RAM is A-Die, I'm happy about that.  Running some tighter timings already & it seems perfectly happy, and those timings do result in a pretty decent FPS gain in some games according to HWUnboxed & Buildzoid's testing.

Hey , been a while since we talked , I forgot this thread was still active , the one thing I have found with the 4090 is it seems to have V-droop depending on what you are playing,  Most 3D mark  benchmarks do not react the same as full Kombustor loads but they do seem to be really close to the way games react . DX 11 and DX 12 are also to be considered, but in the end , every card has to be FAFO'ed to find what it will do , and make sure what are the background tasks you use all the time... the day after my last post I lost 2-400 points one night but it was background programs.  It's not performance as much as it is Quality of life issues for me , that said, It is funny that my undervolt can be superior over stock conditions but try that Vulcan donut and see it bow down 14 fps, even though it held an advantage in Game loads. I Have an open Praxis wet-bench but not sure I will use it as I have dremmelled out my OD11 to fit my needs. (if you want one? pm me) 
so I tried lower V and it just sucked on my current 4090 , at less than the performance drops faster than the voltage 

The last 2 cards I blocked won't go near +2000 memory , what is the memory temp with the OC?

I tried 2730 at .930v and It crashes in the PR loop pretty quick, I think the my ram that never gets above 42C is the culprit, There is talk about using tape on the Ram pads to get those temps up a bit, on the OverClock.net forum
Good thing it's winter my card is really stable at 3010 , lol I will just use it as a aux heater 

Hi Talon, 
yes its seems like its now OC season , I tried a 1.025v @ 2910 and it was pretty good , pulled about stock power and only a few dips to 2895, but mostly steady at 2910 with 1300 on the mem and it hit 50 C !
strangely enough my most stable OC is +158c , +1250m , 1.1v  480w  , and it topped out at 49 C in an hour , room still 72 F , going to heat my house this winter with an OC . LOL 

cezarL
Mine - Suprim X air-cooled - can do 2610@0.900 and 2730@0.950.
The difference between the two is about 3-4fps in CP2077 (96-97 v 100-101 fps in-game benchmark at 3440x1440 with DLSS Quality, PT, RR and FG) at the expense of some 50W (300W vs 350W) and 5-6C temps at 2000rpm fan speed (50-52C vs 56-57C).
Not sure that's worth it though... I'll stick with the lower undervolt. Besides, none of the other games I play are as demanding as CP, 2610MHz can easily handle 120fps even with DLAA or DLDSR+DLSS Quality.

Good thing is the card can do +2000 on memory, as far as I can tell. No artifacts that I can see and scores in 3DMark keep going up for every +100MHz I add up to the limit allowed by Afterburner. Didn't really expect that...

Worth noting that I had to run three stress tests (Port Royal, TimeSpy Extreme and Speedway) to check stability. Higher clocks would pass one or two stress tests but not all three.
Are there any other stability tests (preferably free if not part of 3DMark) worth checking out, especially for checking memory?

Sounds like you got a golden sample in terms of VRAM clocks at least, most people I've seen running air cooling are lucky if they hit 1500... apparently, mine does 1500 "all day" (I've gamed on it, benchmarked it, etc, hasn't failed anything yet or artifacted) but will fail out of 3DMark PR at 1650.  
 
I've been getting in to AI/rendering, which this card is a beast at, so I'm revisiting my undervolt OC stuff to see if I can find a good medium between wattage used & speed without compromising stability, probably gonna aim for the 2850/2835 setting I had before.
 
 

I stopped using OCCT a while ago because it was so buggy, haven't picked it back up yet.
 
I hit a thermal limit on my 4090 for the first time tonight, doing some render testing at different settings on the GPU, I wasn't even running it at my full bore 2970MHz @ 1070mV either lol 
I hit around 560W of power draw, GPU Hot Spot readout was 87.6C & the GPU core clock & voltage both nosedived til it cooled off... I enabled some settings in the renderer that increased GPU/VRAM load in order to decrease render time, and it worked, but dayum that heat lol
 
I'm testing at the 2625MHz 900mV settings & it seems much happier, hot spot during the same test load is only getting in to the low 60s C, core is barely hitting 53C (vs nearly 70C when it throttled at the settings before), and I'm only seeing 380W or so of load instead of almost 600 lol
Funny part is, running that with the VRAM @ +1400MHz, I'm actually rendering just as fast, or faster, than I was at the settings that were running hotter than hell.  Seems like this is akin to many applications that care more about VRAM/memory speed on the GPU than anything else.

Most I go is 535W ,with only a 110% on the power , but I am not going to try any of the other Bios , main reason is the darn backplate EK made has no hole where you can get to the switch, so it's surgery if I needed to change it.  For what ever reason I can't go low, 900-950 mv , I set the clock to 2610 and it over shoots and crashes.  2820@ 975 all day long and 41C max and it wont heat up the room , if it's cold i use the 2910@ 1025 and it pulls stock wattage and will TSE,  PR a 99% run with 13900k and DDR4 still .  You are right about the performance difference in game with an OC these days, it really doesn't matter much now with the AAA games , It;s more about the quality of life issues around heat that makes a difference .

cezarL
560W!
 
After quite a bit of testing a bunch of clocks+voltages, I ended up with three profiles saved in Afterburner: 2610@900, 2805@0.975 and 2955@1.050 (2940 effective, it drops almost immediately) thinking I'd use the higher clocks in the most demanding games.
 
But now even with CP2077 (which eats the 4090 for breakfast) I stick with the 2610 one - low temps, very low power usage... the others aren't worth it for the meagre increase in fps.

I just did some rather extensive testing in Stable Diffusion (the AI rendering software I'm using), and the results are similar to what we see in gaming for the different GPU settings, although the gap is even less because GPU rendering is more reliant on VRAM speed (at least in SD) than GPU core:
 
For a set batch of 2 with a batch size of 8 (so 16 images rendered total) at 768 x 768, upscaled to 1536 x 1536 (for those that don't know, that's not easy to do, most GPUs simply cannot, because the VRAM required is 14+GBs for just that):
 
2625@900mV results in 48.4C max core temp, 58.3C max hotspot, 56C mem temp, 335W power draw for 4m27s render time
2985@1070mV results in 60.2C max core temp, 75.5C max hotspot, 62C mem temp, 486W power draw for 4m17s render time
 
So, again, this is for 16 images being rendered, saving 10 seconds (which is a 4-5% reduction in total time I think), it's running 12C+ hotter on the core/hotspot, 6C hotter on the Memory (probably from heat soak from the GPU core increase), and 150W more power draw.
That's the definition of "not worth" lol
 

kraade
Most I go is 535W ,with only a 110% on the power , but I am not going to try any of the other Bios , main reason is the darn backplate EK made has no hole where you can get to the switch, so it's surgery if I needed to change it.  For what ever reason I can't go low, 900-950 mv , I set the clock to 2610 and it over shoots and crashes.  2820@ 975 all day long and 41C max and it wont heat up the room , if it's cold i use the 2910@ 1025 and it pulls stock wattage and will TSE,  PR a 99% run with 13900k and DDR4 still .  You are right about the performance difference in game with an OC these days, it really doesn't matter much now with the AAA games , It;s more about the quality of life issues around heat that makes a difference .

Usually in order to get it to hit 2625 @ 900mV and actually WORK, clock properly etc, I have to set it a few times (as in set it, run something to check, then reset to 0 offset, and try again), and setting a profile then reloading it does not solve this from happening either.  I think this is an effect of Afterburner not being as well maintained as it once was & thereby not being as compatible with the 40-series cards.
 
I would say for most people, aiming for stock or slightly above stock clocks, while lowering your voltage as much as you can while maintaining stability, should be the goal, because they will result in lower temps, lower power draw, and the same performance as stock settings.
 
Perfect example is the SD rendering stats I posted just now, there's literally no reason to run the higher clocks in terms of workload-time vs heat generation etc.  It just makes no sense.  For gaming, most people with a 4090 can punch in 2775-2790MHz @ 975mV (maybe a bit higher if they have a really bad quality die) and just run with it.  The resulting (basically) 10% lower voltage will lower their power draw by at least 50-75W if not more compared to stock, run cooler by a noticeable margin, and they won't see a lick of difference in FPS.  MAYBE 1% lower, at most, because of how wonky the effective clock is when you start messing with undervolting, but the difference is well worth it for the reduction in temperatures & power draw. 
 
Even for someone running a full coverage block like yourself, I'd still say doing that is more worth it to eliminate any worries about the 12VHPWR connector if anyone is running the adapter.  For those running a PSU with native support for the connector, power draw is less of an issue though.  I plan on upgrading to a Silverstone HELA that has a built in connector that's got full 600W support, but that's a beefcake of a unit, 1200W & it's around $230, so it's not a huge priority at the moment since my current eVGA 1200W P2 is doing just fine... I just HATE the adapter so much.
I hate the connector, to be frank, but it is what it is.

ya I have just 2 undervoltz / OC , .975v for when the AC is on and 1.025v OC for when the house is 72 or less.
 
Not worth it going Low for me , just gets unstable below .950, My KPE was awesome at .925...and the 3090 Ti in the wife's machine has a .975 OC for the daily