[SOLVED] Junction temp immediately spikes to 110C - Radeon 6800xt

Status
Not open for further replies.

adsoyo

Distinguished
Mar 26, 2012
68
4
18,565
I recently bought and installed an Eisblock Aurora GPX-A water block for my MSI Gaming X Trio 6800 XT. When I benchmark or play a game, the junction temp immediately climbs to 110C and the GPU throttles itself. Core temp is 77-80C. I used the included thermal paste and applied it according to the instructions. When I saw temps were high, I disassembled, cleaned and liberally applied a different brand of thermal paste: Arctic Silver 5 but am still getting max junction temps. I'm using Gelid GP-Extreme thermal pads that are 1.5mm for the front (instructions say use 1mm) and 3mm for the back. All the pads have imprints showing good contact. Idle temps are about 30C. I'm stumped. It's like one of the chips doesn't have a thermal pad on it at all. Can thermal pads be "dead"?
 
Solution
Yes, gpu hotspot is the reading coming from the hottest sensor on the die.

If pads have too much give, or are too short, then the gpu core should be fine, but memory probably isn't.
If pads don't have enough give, or are too firm, then the opposite happens, as the PCB is bending/warping at points it normally didn't. The die likely isn't sitting as flush against the block like it's supposed to.

watzupken

Reputable
Mar 16, 2020
1,020
516
6,070
I may be wrong, but something don't sound right. If you are using custom water cooling, the core temp looks very high. The Sapphire RX 6800 XT Nitro+ that I used to own, rarely hit 80s, and this is on air with an ambient temp of around 27 degrees C. The junction temp is usually between 15 to 20 degrees higher than the skin/ core temp. So I think it is worthwhile checking if there is some protective film on the water block that you have not removed? It sounds like contact is very poor, which is hindering the heat transfer. Otherwise, you need to check if the water pump and fans are working properly on the cooler. The junction temp you mentioned should be pertaining to the core right? Because if that is so, then I don't think it has anything to do with the thermal pads that you used on the VRM and VRAM.
 

adsoyo

Distinguished
Mar 26, 2012
68
4
18,565
I've been reading up on junction temp since I posted this. Looks like the other name for it is "hot spot" and it is in the core. I thought junction temp was the highest temp anywhere on the PCB but that's not the case. So it sounds like the thermal pads are most likely fine.

The threads I'm finding elsewhere are saying 110C isn't dangerous but it is the point at which the GPU will thermal throttle. Part of the reason I water cooled is to never worry about thermal throttling so I really want to figure this out. I emailed Alphacool, we'll see if they respond.

I took a picture of the block off the card so you can see what kind of contact it's making but it looks like Toms Hardware won't let us upload pictures to the site. It seems to be making good contact. And no, there isn't a piece of film on the block ;)

When I stress the card and it's thermal throttling, the voltage is at ~1.015, watts are ~220, clock is ~2250
 

Phaaze88

Titan
Ambassador
Yes, gpu hotspot is the reading coming from the hottest sensor on the die.

If pads have too much give, or are too short, then the gpu core should be fine, but memory probably isn't.
If pads don't have enough give, or are too firm, then the opposite happens, as the PCB is bending/warping at points it normally didn't. The die likely isn't sitting as flush against the block like it's supposed to.
 
Solution

adsoyo

Distinguished
Mar 26, 2012
68
4
18,565
I should probably ask for clarification on this one: instructions say to hand-tighten the screws crosswise one turn at a time. They don't actually mean finger tight do they? I'm using a screw driver and tightening the screws until they stop.
 

Phaaze88

Titan
Ambassador
Yes, they do. The Eisbaer Aurora I'm using has the same instructions to hand tighten.
I did ignore this at first and followed the traditional method, but soon found that my PC wouldn't boot. Went back and let off some pressure... it boots now.
Whoa, Alphacool was serious about that...
 

adsoyo

Distinguished
Mar 26, 2012
68
4
18,565
Problem solved. It looks like the extra 0.5mm on the thermal pads was indeed too much. I just finished tear down and reassemble #5 this time using 1mm pads on the front. I fired up heaven benchmark to stress the gpu and to my delight the core only hit 51C while the hotspot plateaued at 66C. Sweet! I ran Time Spy since temps were under control and landed in the 96th percentile, splendid! Time to game all day, thanks to everyone who commented!
 
  • Like
Reactions: Phaaze88
Status
Not open for further replies.