[SOLVED] RTX 3090 Running Very Hot

KIMBOF1

Honorable
Jan 5, 2017
14
1
10,525
I recently purchased a Palit RTX 3090 GamingPro which is running very hot and the fans are maxed out.
I know is not the best model however I am planning to install a waterblock in the future anyway so wasn't too interested in paying more for better air coolers/heat sink. For reference[ here is my system:

Case: Phanteks Enthoo 719
Bottom Fans: 3x Corsair LL120 Intake
Front Fans: 4x Corsair LL120 Intake
Top Fans: 3x Corsair LL120 Outtake
Rear Fan: 1x Noctua NFA14 Outtake

These fans are running at 1000w-1200w ish, I know I may need to increase the RPM.

I decided to try undervolting my GPU, I first tested the card on the Witcher 3 running at max settings, it was sitting at 1920MHz, 83 degrees, 4000RPM. Later I tested on Heaven benchmark, this time it was throttling and running at 1545MHz, still maxing out fans and temps.

Voltage was around 1v, I undervolted all the way to 700mV, keeping the clock speed at 1550MHz. The card still runs hot at the high 70s but the fan speed is lower now at around 2500RPM.

I got this card on eBay, is it faulty or am I doing something wrong? Any help appreciated.
 
Solution
The eBay seller didn't modify the card or use it at all? Not that you could just take their word for it... IDK. 2nd hand market can be a crapshoot sometimes.
Throttling under Heaven Benchmark? That's supposed to be an easy one to run.


https://www.guru3d.com/articles_pages/palit_geforce_rtx_3090_gamingpro_oc_review,1.html
That's a rather favorable review of this card - but it was tested in an open bench there, so obviously it's going to run a little warmer inside a boxed chassis. The reviewer does bring up noise: they're not the quietest fans.
Which gives me another idea that chassis airflow might be lacking.
-Those LL120s don't move much air - except for the 2200rpm ver, but you'd have to be able to tolerate the higher rpms to...

Phaaze88

Titan
Ambassador
The eBay seller didn't modify the card or use it at all? Not that you could just take their word for it... IDK. 2nd hand market can be a crapshoot sometimes.
Throttling under Heaven Benchmark? That's supposed to be an easy one to run.


https://www.guru3d.com/articles_pages/palit_geforce_rtx_3090_gamingpro_oc_review,1.html
That's a rather favorable review of this card - but it was tested in an open bench there, so obviously it's going to run a little warmer inside a boxed chassis. The reviewer does bring up noise: they're not the quietest fans.
Which gives me another idea that chassis airflow might be lacking.
-Those LL120s don't move much air - except for the 2200rpm ver, but you'd have to be able to tolerate the higher rpms to get that stronger airflow.
-The Enthoo 719 is a huge chassis(own one, but it's not in use ATM), and it's size can be a detriment. Air has to travel further to get to the components that need it, and that means more time for the airflow(CFM) to decrease before reaching said components.
 
Solution

KIMBOF1

Honorable
Jan 5, 2017
14
1
10,525
The eBay seller didn't modify the card or use it at all? Not that you could just take their word for it... IDK. 2nd hand market can be a crapshoot sometimes.
Throttling under Heaven Benchmark? That's supposed to be an easy one to run.


https://www.guru3d.com/articles_pages/palit_geforce_rtx_3090_gamingpro_oc_review,1.html
That's a rather favorable review of this card - but it was tested in an open bench there, so obviously it's going to run a little warmer inside a boxed chassis. The reviewer does bring up noise: they're not the quietest fans.
Which gives me another idea that chassis airflow might be lacking.
-Those LL120s don't move much air - except for the 2200rpm ver, but you'd have to be able to tolerate the higher rpms to get that stronger airflow.
-The Enthoo 719 is a huge chassis(own one, but it's not in use ATM), and it's size can be a detriment. Air has to travel further to get to the components that need it, and that means more time for the airflow(CFM) to decrease before reaching said components.

It seems to be throttling because on some games the core clock is running at 1920MHz but it drops to 1545MHz on Heaven however on the manufacturers page states these specs:

Graphics Clock​
1395 MHz​
Boost Clock​
1695 MHz​

Perhaps I should limit it to the base clock of 1395MHz until I install the waterblock just to keep the fans from getting too loud.

I know that the LL120 aren't the best for moving air, perhaps I should also increase the RPM on these until I install my water loop. My plan is to place a 480mm rad on the front and then a 360mm on the top.

What do you suggest?
 
Last edited:

Phaaze88

Titan
Ambassador
It seems to be throttling because on some games the core clock is running at 1920MHz but it drops to 1545MHz on Heaven however on the manufacturers page states these specs:

Graphics Clock​
1395 MHz​
Boost Clock​
1695 MHz​

Perhaps I should limit it to the base clock of 1395MHz until I install the waterblock just to keep the fans from getting too loud.

I know that the LL120 aren't the best for moving air, perhaps I should also increase the RPM on these until I install my water loop. My plan is to place a 480mm rad on the front and then a 360mm on the top.

What do you suggest?
Since the plan is to slap a waterblock on it anyway, returning it and getting your money back isn't a necessity.

Using either Hwinfo or Gpu-Z, monitor:
Gpu Core temperature. Maximum 91C.
Gpu hotspot(hottest sensor off the die). Maximum 110C.
Gpu Memory temperature. Maximum 105C.
These don't exactly tell you what or where the problem is though.
High core temp can be from using the 'wrong' paste, or too thick thermal pads, the latter of which would shift die to cold plate surface contact.
High hotspot temp, same as above.
High memory: can be from pads missing or too soft and mating loosely with cooler, pads are torn, and I've been told simply taking the cooler apart requires replacing the pads unless the cooler is reattached perfectly, so the former prints in the pads aren't leaving air pockets.

You can ignore the advertised BOOST clock - up to a point - as it is more of a worst case scenario. The cards can all easily boost much higher than that if parameters are acceptable... though clearly, something is wrong.

You can run the fans faster, if sanity persists; I don't know how loud they get, if at all.
 

KIMBOF1

Honorable
Jan 5, 2017
14
1
10,525
Did you get a decent deal on eBay? 10%+ less than what others are selling the same card for?
It's possible that your card is damaged. It shouldn't be throttling that much.

What PSU are you using? What's your ambient temp?
I will need to check ambient temps as I'm not sure.

PSU - Corsair RM1000i

As for the card - the starting bid was £1500, the auction ended and I won it for £1700 so it is definitely slightly cheaper than other cards I have found on eBay. I have read that the cooler used on the Palit 3090 is the exact same as the Palit 3080 which is obviously not sufficient for a more powerful card.
 

KIMBOF1

Honorable
Jan 5, 2017
14
1
10,525
Since the plan is to slap a waterblock on it anyway, returning it and getting your money back isn't a necessity.

Using either Hwinfo or Gpu-Z, monitor:
Gpu Core temperature. Maximum 91C.
Gpu hotspot(hottest sensor off the die). Maximum 110C.
Gpu Memory temperature. Maximum 105C.
These don't exactly tell you what or where the problem is though.
High core temp can be from using the 'wrong' paste, or too thick thermal pads, the latter of which would shift die to cold plate surface contact.
High hotspot temp, same as above.
High memory: can be from pads missing or too soft and mating loosely with cooler, pads are torn, and I've been told simply taking the cooler apart requires replacing the pads unless the cooler is reattached perfectly, so the former prints in the pads aren't leaving air pockets.

You can ignore the advertised BOOST clock - up to a point - as it is more of a worst case scenario. The cards can all easily boost much higher than that if parameters are acceptable... though clearly, something is wrong.

You can run the fans faster, if sanity persists; I don't know how loud they get, if at all.

The problem I am having is the fans are particularly loud - when the core is running at 83C which is the hard limit set on the card, the fans are running at 100% which equates to 4000RPM so it sounds like a jet engine. I would like to achieve lower temperatures so I don't have to put up with this loud noise until I install the custom loop.

My previous card was an MSI GTX 1070 Gaming X, the fans on this model are top notch, I never heard them ever get even close to the racket this 3090 makes even during stress tests. I had also never changed the settings on this card apart from an overclock I had for a while which suggests to me either the fans are far superior in sound level or they never had to spin as fast due to the card never reaching those high temperatures (not sure as I never really tested).

I'm convinced that this is either an underlying issue with the 30 series cards drawing too much power and generating too much heat with not enough surface area to dissipate OR I have been sold a faulty/damaged card but I really don't know how I can verify which one of these eventualities is true so I can act accordingly. Regardless of whether I am installing a waterblock or not, I do not want to use a faulty card to avoid further issues. Is there somehow I can verify that my card is working as intended or is it really just best judgement?

Thanks for replies.
 
The problem I am having is the fans are particularly loud - when the core is running at 83C which is the hard limit set on the card, the fans are running at 100% which equates to 4000RPM so it sounds like a jet engine. I would like to achieve lower temperatures so I don't have to put up with this loud noise until I install the custom loop.

My previous card was an MSI GTX 1070 Gaming X, the fans on this model are top notch, I never heard them ever get even close to the racket this 3090 makes even during stress tests. I had also never changed the settings on this card apart from an overclock I had for a while which suggests to me either the fans are far superior in sound level or they never had to spin as fast due to the card never reaching those high temperatures (not sure as I never really tested).

I'm convinced that this is either an underlying issue with the 30 series cards drawing too much power and generating too much heat with not enough surface area to dissipate OR I have been sold a faulty/damaged card but I really don't know how I can verify which one of these eventualities is true so I can act accordingly. Regardless of whether I am installing a waterblock or not, I do not want to use a faulty card to avoid further issues. Is there somehow I can verify that my card is working as intended or is it really just best judgement?

Thanks for replies.
Unless your ambient temp is really high, or your case cooling is really bad, I'd say your card is either damaged or has been taken apart and put back together without proper TIM.
Even extended run reviews I've seen don't have the GPU clocking below 1700MHz.
 

KIMBOF1

Honorable
Jan 5, 2017
14
1
10,525
I ran the heaven benchmark through a few times giving the card a run to warm up before recording hardware information. This test was carried out with stock settings i.e no undervolt etc - You can see a few things:

--Just occurred to me that the screenshot does not show column headers so it's as follows - Current Value, Minimum, Maximum--

- You can see here that motherboard temperatures aren't too hot and I know that doesn't correlate to ambient temperature but it gives you an idea.
- The system fans all run into a hub with one cable going to the motherboard so you can see they are running at around 1200-1300 RPM.
  • For reference the CPU temperatures were sitting at around 50-60C.
  • Interestingly you can see the temperatures of my two NVME drives, one of them which is sitting below the GPU is at 38C however the other which is right above the hot GPU backplate is sitting at 50C.?
  • GPU voltages sitting at roughly 800mV.
  • There are three GPU fans and two of them are grouped into one so you can see they are running at roughly 3500RPM
  • GPU core temp sitting at around 80C, memory 90C, hotspot 100C.
-Graphics clock again sitting in the 1500MHz to 1600MHz range.

Does this look normal? I have never run extensive tests on GPUs before.

PM1pOQ9.png
 
Last edited:

KIMBOF1

Honorable
Jan 5, 2017
14
1
10,525
View: https://imgur.com/WgpnmZj


That's why you're having this problem. Hotspot is bloody high.

High core/hotspot temp can be from using the 'wrong' paste, or too thick thermal pads, the latter of which would shift die to cold plate surface contact.

Do you suggest I switch out the pads/paste or just wait until I install the waterblock? I can always underclock and undervolt the card in the mean time perhaps?
 
  • Like
Reactions: bryanc723

Phaaze88

Titan
Ambassador
How far off is the waterblock project?
If it's tomorrow, or a few days from now, I'd would consider refraining from playing gpu heavy games while I wait, but if it gets that hot from just idling on the desktop or simply browsing the web(gpu acceleration), I probably wouldn't use it at all.
If we're talking weeks/months, I would try a repaste first.
Good pastes for gpu applications:
Noctua NT-H2
Prolimatech PK-3
Gelid GC Extreme
Kingpin KPx
Cooler Master Mastergel Pro V2

IF hotspot doesn't improve with the repaste, then it's the pads. This may not be worth messing with. You'll be getting a waterblock in time, and that comes with its own set of pads anyway, right?
Shopping for a different set of pads in the meantime can be a crapshoot. It's been made more complicated than it needs to what with the w/mk ratings, different testing conditions, and pads in the same size class having different levels of hardness.
You could end up going through a few sets before 'finally getting it right'.
 
Last edited:

KIMBOF1

Honorable
Jan 5, 2017
14
1
10,525
Thing is, constantly hitting that hotspot high can slowly damage your card.
When will you be getting the waterblock going?

Thanks all for replies, I'm not building the water loop just yet from a financial stand point - I will look into it in potentially a few months time when I've got some cash to spare.

Repasting may be my best option so far, as for installing new pads I've seen a video somewhere on YouTube putting pads on this very card and it seems a bit hit and miss as the stock pads seem to be a non-standard thickness I think it was 0.75mm or 0.8mm? The guy I watched had to take a thicker pad and compress it to size before seeing noticeable performance improvements.