Overclock.net banner
1 - 11 of 11 Posts

Wolverine2349

· Registered
Joined
·
1,367 Posts
Discussion starter · #1 ·
What is the real maximum safe temps. I have searched and cannot find any answer.

I have a Gaming OC RTX 4090 and I set a custom fan curve to keep noise down while getting the fan speed higher to like 70% when it hits 75C with a curve going all the way to 100% at maybe like 84 to 90C

It seems to keep fan quiet enough.

The GPU temp maxes out at like 78C though hotspot hits 89C and once hit 91C

Memory junction temp hit 82 to 83C.

Are these too high or good temps.

And is there any risk of damage if temps get too high or will the card auto throttle its speeds to protect itself from damage?
 
The thermal limit is 84c.
Above that and it will throttle automatically.

I´d say don´t worry too much.

If you want to get rid of a few degrees the backplate does dissipate heat quiet well and appreciates a little help:

Motor vehicle Computer hardware Audio equipment Radiator Machine


What you see helps a lot with simple games (stays below fan activation)
Full tilt this is only worth around 4 degrees

Or you go all out with a dual tower gpu backplate :)
Automotive lighting Circuit component Computer cooling Hardware programmer Electrical wiring
 
Discussion starter · #3 ·
The thermal limit is 84c.
Above that and it will throttle automatically.

I´d say don´t worry too much.

If you want to get rid of a few degrees the backplate does dissipate heat quiet well and appreciates a little help:

View attachment 2621401

What you see helps a lot with simple games (stays below fan activation)
Full tilt this is only worth around 4 degrees

Or you go all out with a dual tower gpu backplate :)
View attachment 2621402

That 84C thermal limit is for GPU temp not the hotspot nor memory junction which run hotter than GPU temp? Am I correct? The GPU temp has never hit 84C though it did get to 79C or 80C. Its the hotspot and memory junction sensors that read 5-15C hotter?
 
That 84C thermal limit is for GPU temp not the hotspot nor memory junction which run hotter than GPU temp? Am I correct? The GPU temp has never hit 84C though it did get to 79C or 80C. Its the hotspot and memory junction sensors that read 5-15C hotter?
I faced similar questions when I got my Gigabyte-G-OC late October '22. The hotspot temps were getting too high, IMO, even if the stated general GPU temp max was not reached (88C on my original AD102-300 / 1.1V /600W). My VRAM temps were much lower though than what you mentioned. The stated max temp of 88C is general GPU temp (not hotspot) as far as I could find out. Memory junction temp shows the highest temp by any of the VRAM.

In any event, I usually water-cool everything anyway (but especially a 600W GPU) and besides, per upper pic below, the air-cooler did not even fit in my dual mobo custom build. Water-cooling the 4090 (bottom right below) dropped > 30 degrees on average. If everything is seated correctly from the factory (there can be issues), the Giga-G-OC air-cooler works reasonably well, but given hot summer temps (it is 29 C ambient here right now) and the 600W envelope, I would consider water-cooling if you can.

Purple Automotive lighting Entertainment Electricity Automotive design
 
Discussion starter · #5 ·
I faced similar questions when I got my Gigabyte-G-OC late October '22. The hotspot temps were getting too high, IMO, even if the stated general GPU temp max was not reached (88C on my original AD102-300 / 1.1V /600W). My VRAM temps were much lower though than what you mentioned. The stated max temp of 88C is general GPU temp (not hotspot) as far as I could find out. Memory junction temp shows the highest temp by any of the VRAM.

In any event, I usually water-cool everything anyway (but especially a 600W GPU) and besides, per upper pic below, the air-cooler did not even fit in my dual mobo custom build. Water-cooling the 4090 (bottom right below) dropped > 30 degrees on average. If everything is seated correctly from the factory (there can be issues), the Giga-G-OC air-cooler works reasonably well, but given hot summer temps (it is 29 C ambient here right now) and the 600W envelope, I would consider water-cooling if you can.

View attachment 2621429

I never water cool and no I do not use the 600 watt limit of my card. I keep it at default and it never goes much more than 400 wats and usually high 300s sometimes low 400s with a rare spike to maybe 450 watts during gaming. Never gets close to 600 watts and I think it only does if you overclock and increase power limit above 100% which I never do.

My ambient temp is 21.6C (71F) as I have an AC set around that in my whole house so much cooler than your ambient temp.

What did your hotspot temp reach BTW?
 
I never water cool and no I do not use the 600 watt limit of my card. I keep it at default and it never goes much more than 400 wats and usually high 300s sometimes low 400s with a rare spike to maybe 450 watts during gaming. Never gets close to 600 watts and I think it only does if you overclock and increase power limit above 100% which I never do.

My ambient temp is 21.6C (71F) as I have an AC set around that in my whole house so much cooler than your ambient temp.

What did your hotspot temp reach BTW?
pre-water-cooling, around 107 C
 
The best thing I've found is simply keeping my 4090 around 450w or so. Zero difference in gaming experience and only a few fps here and there gain by running it unhinged.

I never even look at my GPU settings any more and simply keep it at a reasonable overclock and with the power limit set to 80%.


No practical reason to run any of the 600w capable cards near that power output level. One could actually make an argument for "dumb" but I won't start into that one ...

Smile Happy Font Circle Symbol
 
The best thing I've found is simply keeping my 4090 around 450w or so. Zero difference in gaming experience and only a few fps here and there gain by running it unhinged.

I never even look at my GPU settings any more and simply keep it at a reasonable overclock and with the power limit set to 80%.


No practical reason to run any of the 600w capable cards near that power output level. One could actually make an argument for "dumb" but I won't start into that one ...

View attachment 2621466
I agree, but don't mind hitting max power for the odd benches. Still, most of the time for gaming apps, I set it to around 430 W range. 4090s can have quite substantial transient power spikes, though. Also, my typical settings for 430W in the majority of games I play can still get to well past 500W in others, such as CP2077 max w/path tracing. At the end of the day, it is just about a cool-running and quiet system that lasts.
 
The max temp for any RTX 3000/4000 tight now are very similar or exactly the same:

GPU core around 90ÂşC but thermal throttling start at 83ÂşC
VRAM 120ÂşC but thermal throttling start at 106-110ÂşC depending on the AIB
HOTSPOT very similar to VRAM, and thermal throttling will start around of 110ÂşC

GPU Fans are linked to all this sensors, so any of this sensors will get to thermal limit fans will start spinning at 100% speed
 
  • Rep+
Reactions: Papusan
I agree, but don't mind hitting max power for the odd benches. Still, most of the time for gaming apps, I set it to around 430 W range. 4090s can have quite substantial transient power spikes, though. Also, my typical settings for 430W in the majority of games I play can still get to well past 500W in others, such as CP2077 max w/path tracing. At the end of the day, it is just about a cool-running and quiet system that lasts.
True. And I too enjoy running unhinged for occasional isolated benchmarks. But 450w is ideal and great because I can just pretty much forget my GPU as it handles everything at that wattage limit very well. But as you said, without a proper and doted over voltage curve, simply using the power slider does produce different results game to game.

But it works for me and generally keeps my core around 60c and hotspot no higher than 75C to 80C at most.

600w benching is fun too, though. ;-)
 
With my 4090 Strix I overclocked and undervolted.
I set the curve to 2850@0.975mv and added 1000mhz to the memories.
Nothing extreme but enough to improve the situation from all points of view.
Excellent temperatures, decreased consumption and increased performance.
I have a single loop with 13900k, 4090 and two EK XE 240 radiators in an O11 Air mini.
In a bigger case with more cooling possibilities they might be even better.
I made a Timespy on the fly to show some measurements:
Product Font Computer Screenshot Operating system


This is without undervolt and with a boost at 3150mhz:
Light Font Screenshot Software Electronic device
 
1 - 11 of 11 Posts