MiztahSparklez

Dabadger84
 
You have to keep in mind, there's vRAM on both sides of the PCB on the 3090, so there's memory on the top of the card.
 
Try using HWInfo64 and check the "Memory Junction Temperature" they recently added, it's possible you're getting close to thermal limit on the chiplets and that's what's causing the crash.
For a comparison on temp delta between the units, here's a screenshot from a Folding @ Home unit I finished earlier's temperature readouts:
 
So there is' some serious variance between the 3 readouts even on a Hybrid Kingpin card.  I imagine that "difference" is even larger on air-cooled cards.
 
Edit: Right side of the readouts is the maximum temps seen, obviously.

Yeah, there's still thermal padson the back side of the card as well, it just doesn't get the copper heat plate that the front has.  I saw some folks attach heat sinks with thermal tape to try and help with the cooling back there.
 
That actually looks pretty close to my temps in terms of ratio for your max temps.  mem2 is highest generally speaking, followed by mem1.  GPU is super cool in comparison, due to the water block and kryonaut, similar to the mem3 reading.
 
My GPU2 temp would almost track in line with mem1.
 
I may just leave it as-is then.. I might just be reaching the limit, as you mentioned.
 
Interesting to note that your max temps are probably only 5 degrees or so less (however, you're probably running a more aggressive OC).
 
I was originally running a +750 mem +145 gpu with 112 power.  119 power would cause crashing, probably because of heat.
I've decreased this now to +200 mem +145 gpu with 110 power. seems fairly stable.  I've actually even seen my FPS remain more consistent, vs jumping up and down when at the upper clock.  My bench scores have decreased maybe by 200-300 in time spy, but FPS in game (5120x1440) is 90-120, vs 75-140.
 
As a sanity check, stock clock , I would not see any crashing for the most part and could play for hours.  Almost in line with my new +200 clock, just 10-20fps less on average.
 
Voltage is at 0 for all scenarios.   While increasing voltage let me get higher stable clocks, it negatively impacted my gaming performance, sometimes reducing it below stock FPS rates, hence the strange stop at 145, vs 150.  150 would crash for me after time without voltage.

MiztahSparklez
 
This particular example was Call of Duty BO. However, I see similar peak temps from benchmarking, etc.
 
Memory Junction temp is HOT!  max is 86C, average 79C.  idle is 60C.  Where is this reading from?  Sounds like it's not getting any cooling....
 
GPU temp is max 63C, average 56C, idle 35C.

RickJamesBish

MiztahSparklez
(it was HOT when it was still air cooled, easily into the mid to high 70s under load).  I would just have expected to have everything follow the same trend.  Perhaps HWINFO is not reading this sensor correctly and is reading 20 over?
 
Even at stock speed, temps are maybe -2 from the OC temps at idle, but at max, it's still roughly the same, with maybe even a couple of the readings above the OC speed (2 degree variance).  I'd probably just say that's good as equal though, just minus 20fps or so without the OC.
 
Rickjames has some crazy low temps.. must be an air conditioned room

 
I don't know the temp is right around my case. It cold here so we heat to about 72F.  I do run the aggressive fan curve in PX1 with all three fans set identical and I do not OC. You mentioned mid 70 under load when yours was air cooled. So this is the addon AIO? Mid 70 on air seems high to me. When I first got this card I was low 70 when gaming but I changed cases to a much better air flow and it made a huge difference.

Attached Image(s)