GPU Performance limit thermal.

Hi everyone,

I have an issue where I keep getting performance limit thermal yes. I have a water cooled strix 3090 with an ek water block and backplate. It has since come top my attention that ek provided the wrong instructions for applying the the right size thermal pads to the back of the card and they might not be actually making contact with the memory on the backplate. According to hwinfo my temps dont seem that high. Really I just want clarification that this might be the issue and not some bug. I dont really want to dismantle my loop if its not going to fix issue and im not entirely sure my thermal pads are not touching the backplate anyway. Hope someone can help thanks
 

Attachments

  • Screenshot (2).png
    Screenshot (2).png
    659.8 KB · Views: 72
Indeed, your temperatures look OK (including memory) and should not be tripping the thermal limit.
Those performance limit indicators come from NVIDIA drivers and they don't provide further explanation on criteria triggering them. So I'm afraid I cannot help with this, you should ask NVIDIA about this.
 
Ok Martin. Thanks very much for your reply. So you can you just clarify that you don’t think my rear memory modules could be overheating. Thanks
 
68 C is a very good temperature for GDDR6X, the throttling starts at ~ 110 C.
 
You're welcome. BTW, the GPU is able to report several other temperatures that are not shown to public. If you'd attach the HWiNFO Debug File with sensor data at the highest load/temperature, I could check if there's perhaps some other temperature going too high, but I won't be able provide you its meaning because it's sort of a hidden feature.
 
Last edited:
Modern CPUs/GPUs monitoring a vast amount of internal sensors for their own protection and functionality but as Martin said are hidden. From what I see, I think the GPU is totally healthy.
Memory junction temp is the highest among all memory modules if I'm not mistaken. Like a CPU/GPU junction temp (hotspot).
This GPU is exceeding its power limits, using all voltage it can and pretty much using all resources available. And thats because its watercooled. Even so, it has to stop to a certain point and thats why those limits exist. Its not something to be determined from first glance, but you can figure it out from other readings. I think its totally normal behaviour.
 
This version of HWiNFO provides the tool tip info on every sensor/reading. What is the description of those limits?
And It could be a lot easier for others to examine this screenshot if full name of each sensor was visible...
 
Ok that’s great. Here's another screenshot with full sensor names visible. How do I attach the debug file? As its a csv file.
 

Attachments

  • Screenshot (4).png
    Screenshot (4).png
    605.4 KB · Views: 29
Last edited:
I analyzed the file and there's another temperature sometimes reaching a bit higher value, but it's just a few degrees above the memory. The maximum I observed in the dump was 69 C.
But as I said, I don't know the meaning of this value, I just assume it's some hotspot temperature, but not sure if it's in the GPU chip or somewhere else.
It might be interesting to expose this value to users, but it could cause confusion if its meaning is not well known.
 
Ok. I think I will order some new thermal pads and see if that cures the throttling. Thanks ever so much for your help. Actually just checked the new ek installation manual for the backplate and it looks like it’s probably not just memory that is not got the right size thermal pad on it. So it probably relates to something on the back of the card.
 
Last edited:
Let's do some test. I will expose this temperature as HotSpot and you can check if the Thermal Limit correlates with this value.
Will provide a new test build in a few minutes...
 
So I’ve just tested with the new version with the hotspot but it doesn’t seem to be that triggering the thermal throttle. I have a max temp on that hotspot of 69.8 but thermal throttle yes triggered at about 67 after it had already reached the maximum of 69.8. Also I’ve been sat here 10 mins then the thermal throttle yes triggered and what is strange is that the average changed to yes at the same time as the maximum.
 
Last edited:
Ok that’s great. Here's another screenshot with full sensor names visible. How do I attach the debug file? As its a csv file.
I assume you’re using a modified BIOS because that 460W I think can’t happen on any stock one. At this level of power (120%) it’s normal that something will (and should) hit a wall.
 
I have a Asus strix oc (480w) It still triggers thermal throttle yes at stock settings. I was only concerned because of the mistake of ek supplying the wrong instructions/thermal pads for the backplate. I only spotted it on a forum I read and a lot of people don’t seem to be getting the thermal yes but like you say it’s likely they have a different brand of 3090.
 
Back
Top