• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

3090 GDDR 6 temps

Associate
Joined
23 Nov 2005
Posts
719
Location
Kingdom Of Fife
Hi All,

My 3090 Inno3d Frostbyte (water block) GPU, settles around 50oC while gaming, which is great!

However, I have a feeling however the the back populated VRAM is not so well cooled, what are your experiences of this? I have seen my VRAM go as high as 92oC in cases, although it usually sits around 84oC!

Anyone know what the specs are for these micron chips? Must be forged in the fires of hell!
 
Hi All,

My 3090 Inno3d Frostbyte (water block) GPU, settles around 50oC while gaming, which is great!

However, I have a feeling however the the back populated VRAM is not so well cooled, what are your experiences of this? I have seen my VRAM go as high as 92oC in cases, although it usually sits around 84oC!

Anyone know what the specs are for these micron chips? Must be forged in the fires of hell!
I don’t think Micron have officially released the spec for their GDDR6X but the consensus seems to be they throttle in nVidia cards around 110c and suffer damage around 120c. So 84 is fine.
 
GDDR6X just seems to be very toasty VRAM, even on my Asus 3090 Strix OC the VRAM junction temp tends to be around 88c to 92c while gaming, while the core is around 67c in performance mode or around 73c in the quiet mode.
 
Surely GDDR6X temps are high because it's clocked high. Drop the clocks and the temps drop. Nvidia pushed them to the max to complete with AMD and that's what you paid for. The 3090 is maxed out by definition. If you're unhappy with a maxed out GPU then lower the clocks or get a lower tier next time.
 
Surely GDDR6X temps are high because it's clocked high. Drop the clocks and the temps drop. Nvidia pushed them to the max to complete with AMD and that's what you paid for. The 3090 is maxed out by definition. If you're unhappy with a maxed out GPU then lower the clocks or get a lower tier next time.
The best way to lower the temps is to change the power limit or do a custom power curve.
 
Undervolt your card with a custom curve. Wins all round.

Other thing to try without opening up the card to do the heat pads mod and potentially voiding warranty (seems to be no concrete answer to that!) is to fit a basic type heatsink to the top of the card if you have room. You can literally just rest one on the top, or attach with thermal tape if preferred. I've not tried it yet but a few on here have and reported a very decent drop of about 5-8 degrees on the memory junction temps I think? Someone chip in and link the post....
 
You can buy these heat sinks which come with adhesive as well.

Just bear in mind that this will make your card heavy and it will work on cards with a flat backplate, and only if the backplate is correctly connected to hot components (ie if the backplate runs hot itself).

Because the card will be heavy, you should make sure that you support it correctly or it will possibly bend or break.
 
The best way to lower the temps is to change the power limit or do a custom power curve.

The best way is to change the memory pads. I changed the pads on my 3080 Gigabyte Xtreme for Gelid Ultimate (0.5mm x 4 layers = 2mm) and it has lowered RAM temp by circa 35 Degrees C and completely eliminated thermal throttling. The Rog Strix has better pads but it will still benefit from the above.
 
The best way is to change the memory pads. I changed the pads on my 3080 Gigabyte Xtreme for Gelid Ultimate (0.5mm x 4 layers = 2mm) and it has lowered RAM temp by circa 35 Degrees C and completely eliminated thermal throttling. The Rog Strix has better pads but it will still benefit from the above.

That’s an amazing result, very tempting. But, and it’s a big old but, you’ve voided your cards warranty.

If you’re not mining, undervolting with a custom curve gets you perfectly acceptable temps in gaming, no throttling, with no real loss in performance, draws way less power so saves Leccy - all with your warranty intact.
 
That’s an amazing result, very tempting. But, and it’s a big old but, you’ve voided your cards warranty.

If you’re not mining, undervolting with a custom curve gets you perfectly acceptable temps in gaming, no throttling, with no real loss in performance, draws way less power so saves Leccy - all with your warranty intact.

Just like water blocks I believe this depends on the company. Decent ones will honour warranty if the changes didn't cause the fault. Many also will simply RMA a card and return it to the factory without stripping it at all. And there are no warranty seals on the backplate of any of the 4 x 3080 GPUs I have so far seen first hand. Anyway you have up to 6 years under UK law regardless of warranty if the fault was demonstrably inherent.

And undervolting gets you lower GPU speed boost and changes GPU voltage, not RAM voltage / heat I believe.
 
Last edited:
The warranty thing does seem to be a grey area and vary. Which sucks, shouldn’t be this way.
Anyway, sounds like you’ve never actually tried undervolting? I’ve extensively tested on my 3090fe and it significantly reduces overall power draw and therefore heat, and that does have an impact on ram heat as a result. About 10 degrees or so.

And, depending on what curves you run it doesn’t lower performance, you can actually get better than stock performance using lower volts and less watts. If you’ve never done it I’m aware that sounds bizarre and like wizardry, and it kind of is! It really does work though.
 
Last edited:
The warranty thing does seem to be a grey area and vary. Which sucks, shouldn’t be this way.
Anyway, sounds like you’ve never actually tried undervolting? I’ve extensively tested on my 3090fe and it significantly reduces overall power draw and therefore heat, and that does have an impact on ram heat as a result.
And, depending on what curves you run it doesn’t lower performance, you can actually get better than stock performance.

Heat from the GPU isn't a notable issue on cards with a decent heatsink - its thermal throttling from the RAM chips. On a card with a low end heatsink turning down the GPU might help RAM, but it makes little difference on high end cards from my experience with a couple.

OC scanning tries a whole range of voltages and the results show voltage boost is directly proportional to GPU clock boost up to the stability limits of the card unless I'm missing something?
 
Heat from the GPU isn't a notable issue on cards with a decent heatsink - its thermal throttling from the RAM chips. On a card with a low end heatsink turning down the GPU might help RAM, but it makes little difference on high end cards from my experience with a couple.

OC scanning tries a whole range of voltages and the results show voltage boost is directly proportional to GPU clock boost up to the stability limits of the card unless I'm missing something?

You can lead a horse to water but you can’t make him drink.
All I can say is try it yourself and see (though you don’t really need to now you’ve done your mod).

Other people should definitely look into undervolting though if they’re concerned about the mem junction temps and don’t want to ride the warranty Uncertainty risk wave.
 
I have tried - and as I said turning down the GPU voltage doesn't seem to stop memory thermal throttling on high end cards such as an Xtreme 3080 at least.
You can lead a horse to water but you can’t make him drink.
All I can say is try it yourself and see (though you don’t really need to now you’ve done your mod).

Other people should definitely look into undervolting though if they’re concerned about the mem junction temps and don’t want to ride the warranty Uncertainty risk wave.

I have tried - and as I said turning down the GPU voltage doesn't seem to stop memory thermal throttling being the limiting factor in my testing (before replacing pads at least).

And not clear how undervolting would improve performance - as mentioned above - that doesn't match the OC scanner results. What am I missing?

The thermal throttling is what is protecting the memory junction temps. Although component life is proportional to temperature so best to minimise it.

Just for refence - Xtreme 3080 with new RAM pads at 100% GPU + 100% memory load after 2 hours gaming and then ten minutes of ETH mining (Fans 100%) shows:

GPU Clock [MHz] 1665.0
Memory Clock [MHz] 1281.5
GPU Temperature [°C] 46.8
Hot Spot [°C] 64.2
Memory Temperature [°C] 70.0
GPU Load [%] 100
Memory Controller Load [%] 100
Video Engine Load [%] 0
Bus Interface Load [%] 2
Board Power Draw [W] 231.2
GPU Chip Power Draw [W] 86.4
 
Last edited:
I have tried - and as I said turning down the GPU voltage doesn't seem to stop memory thermal throttling on high end cards such as an Xtreme 3080 at least.


I have tried - and as I said turning down the GPU voltage doesn't seem to stop memory thermal throttling being the limiting factor in my testing (before replacing pads at least).

And not clear how undervolting would improve performance - as mentioned above - that doesn't match the OC scanner results. What am I missing?

The thermal throttling is what is protecting the memory junction temps. Although component life is proportional to temperature so best to minimise it.

ah sorry bud, didn’t sound like you’d actually done it.
Hmm, interesting. From what I’ve seen an undervolt curve should have an effect regardless of the card, but maybe not. All I can say is on my 3090fe it’s definitely a good thing and definitely lowers memory temps (as reported by hwinfo64) along with GPU.

What ‘type’ of undervolting were you trying though? There’s a few different ways of doing it but I’ve always used the curve method. This video here is a good quick guide to how I do it....

https://youtu.be/FqpfYTi43TE
 
Last edited:
Back
Top Bottom