That is just the temperatures causing slight performance loss on the card as it heats up, it's throttling slightly to prevent overheating, which is the built in protections on the card doing it's job.
 
My first recommendation would be to setup a custom fan curve - the card is programmed by default to be silent, and keep the card under the "temperature target" which for Ampere I believe is 83C (pretty darn hot, imo, but also not "dangerous" since the Die can handle up to something like 92C before hard throttling or shut down).  Use Afterburner or Precision X1, turn the fans up 10% at a time until they get "too loud" for the noise level you like.  Once you get to that, turn them back down until it's no longer annoyingly loud, by about 5% at a time.  Whatever % you end up at should be your maximum fan speed in your curve.
Then you just set it up to run at say 30% at lower temperatures, ramping up to whatever maximum speed you came up with at 55C - this will allow the card to run it's fans as fast as you'd like them to go (noise wise) and keep itself cooler, probably by a fair bit, compared to the stock fan curve, which likely tops out at something like 55%, for the sake of noise.
If you don't care about noise at all, do what I did with my 3080 FTW3 Ultra when I had it. First step in the fan curve was 50% at 30C, 80% at 55C & 100% at 70C.
 
This will most likely help with your vRAM temps as well, just keep in mind, the 3090 has VRAM chiplets on both sides of the PCB, it's likely the ones getting hottest are on the back.  As far as the VRM temp goes, those will likely go down once you fix the fan curve to be more aggressive.
 
You can always do what I did, find a PCB shot of the backside of the 3090 FE, and position heatsinks over the areas of the backplate that have them behind it... or just get thermal pads, pop the backplate area off & put them on thermal pads directly on the vRAM chips: https://forums.evga.com/3090-Kingpin-Get-that-backside-VRAM-cool-Also-Thermal-Pad-size-backplate-side-m3219773.aspx 

Thank you @

Dabadger84
That is just the temperatures causing slight performance loss on the card as it heats up, it's throttling slightly to prevent overheating, which is the built in protections on the card doing it's job.
 
My first recommendation would be to setup a custom fan curve - the card is programmed by default to be silent, and keep the card under the "temperature target" which for Ampere I believe is 83C (pretty darn hot, imo, but also not "dangerous" since the Die can handle up to something like 92C before hard throttling or shut down).  Use Afterburner or Precision X1, turn the fans up 10% at a time until they get "too loud" for the noise level you like.  Once you get to that, turn them back down until it's no longer annoyingly loud, by about 5% at a time.  Whatever % you end up at should be your maximum fan speed in your curve.
Then you just set it up to run at say 30% at lower temperatures, ramping up to whatever maximum speed you came up with at 55C - this will allow the card to run it's fans as fast as you'd like them to go (noise wise) and keep itself cooler, probably by a fair bit, compared to the stock fan curve, which likely tops out at something like 55%, for the sake of noise.
If you don't care about noise at all, do what I did with my 3080 FTW3 Ultra when I had it. First step in the fan curve was 50% at 30C, 80% at 55C & 100% at 70C.
 
This will most likely help with your vRAM temps as well, just keep in mind, the 3090 has VRAM chiplets on both sides of the PCB, it's likely the ones getting hottest are on the back.  As far as the VRM temp goes, those will likely go down once you fix the fan curve to be more aggressive.
 
You can always do what I did, find a PCB shot of the backside of the 3090 FE, and position heatsinks over the areas of the backplate that have them behind it... or just get thermal pads, pop the backplate area off & put them on thermal pads directly on the vRAM chips: https://forums.evga.com/3090-Kingpin-Get-that-backside-VRAM-cool-Also-Thermal-Pad-size-backplate-side-m3219773.aspx 

Everything you say makes sense and sounds good. One thing I was curious about was your thoughts on Furmark in general...I've had people say not to run it and I've had other says it's fine and a good stress test for GPU. I've heard it "kills" GPU or even a power supply that isn't up to the task which I assure you mine is (it's an AX 1600i from Corsair). The one thing I've read is that it is harder on the VRM but it sounds like it does that on purpose. I've also remember at least the 2014 version of OC Scanner by EVGA on their site had a similar (and from what I've gathered identical) stress test / gpu benchmark built. The latest version of Precision X1 for the 2000 and 3000 series I don't know if it has it. It sounds like I can use this tool (for general monitoring of the card / fan curve etc.?) even though mine is a FE model?
 
To another point you mentioned it sounds like intended throttling (that anyone with this card would likely get) but:
 
I did have HW Info up during the benchmarks and the VRM reached 92C on the card (When doing Far Cry 5 benchmarks) which is as I understand it within spec. I also noticed the GPU reached maybe 63 C at most. Also definitely fine. Here's what's more interesting.
When I run the 1080 TI SC2 and run the benchmark and then repeat the benchmark without exiting the application the score stays the same and the frames rendered go up with every pass but the average stays the same.
When I do the same thing with the 3090 FE the average drops by 1 an the rendered frames drop by 100 or less and as the test runs continue the drops by 100 shrink by drops of 60 or 70...here's where it gets interesting.
If I close the application and I'm in the desktop and then immediately run it again...say Far Cry 5.. right after getting the last low benchmark score I will get a benchmark identical to the first pass I ran the last time - I got 8740 frames and 148 average. This is repeatable... So what does that mean...memory leak??? Driver immaturity / issues?
 
It's not a cooling issue (at least for the GPU die) because I opened the side panel of my case and that didn't change these results. That 104 C was in Furmark after a few minutes in. I ran a second pass for only 10 min because I didn't think I had any worries because the only temp I could see in Furmark or anywhere was the GPU temp and then I learned about the VRM issue and started to get a bit nervous that I had "damaged" the card. Then I read that Nvidia and AMD have driver precautions in there that limit the cards from getting too hot but I'm not sure if that's true for the VRM. My other reading over the web had to do with what Igor's Lab had identified as maximum TDP / thermal levels for GDDR6X modules and VRM. It sounds like GDDR6 is 105C while 6X is designed to go to 110 TDP and then throttles with a Tmax of 120C. So in "that sense" and only that sense during the two runs of Furmark would I have been there. Seems like games keep it below 95 C on stock curve for VRM and GPU is way cool at like 66-68C or less. The only other finding I've heard about but not tested is heavily Ray Traced titles or Mining (the latter which I haven't done with the card and don't have any intention on doing...because I've heard it can prematurely age it? Is that still true with NiceHash?)
 
Finally - I've noticed similar behavior in "repeat runs" on benchmarks whether it be Far Cry or Tomb Raider or 
 

kevinc313
Use some small 4" zip ties to mount a 80mm or 92mm fan to the back of the card blowing on the mem chips.  Also make sure you have good air flow into the case, no front AIO with a powerful air cooled card.

Thank you :) I do have a Phanteks P600S case and have the front panel on but have all Noctua 140MM fans (3) up front on 140mm in the back and a Noctua D15S cooler. I could remove the front intake panel to increase intake. However, even with that off and the side open I still noticed the benchmark behavior mentioned in my post on subsequent runs. Sounds like the VRM might just get / stay stupid hot regardless under certain workloads?


GREATLY APPRECIATE all the insight and information provided here. It's easing the worry!