As long as I had my XFX 6800xt, I had some temperature problems. The GPU temperature went up to 80°C and the hot spot temperature usually stood at 105°C, but occasionally went up to 110°C. So the fans were screaming as if they were being slaughtered.
So I removed the shroud, replaced the 90mm fans with 120mm fans, which I had lying around in my “tech box” and which are now hanging on the heatsink through zip-ties. And a 100x100 3mm thick thermal pad between the backplate and the back of the board.
Result: Furmark (as an extreme example) just manages to bring the respective temperatures to 57°C GPU temperature or 75°C hot spot temperature.
I would never have thought that an investment of €8 could make such a difference
I am using with my RX 6800 a different thermal paste: “GELID GC-Extreme” I read that this one has good level of viscosity. (I have a Waterblock on mine).
But sometimes I’m tempted in getting a good modern thermal pad and give it a shot to cover more GPU area.
There must have been a poblem from the start. Honestly if it reaches over 100 degree with high fan speed you should have just ask for a replacement. My 6800XT is mostly at 75 hot spot without changing anything on it.
My 6800XT is mostly at 75 hot spot without changing anything on it.
Even in GPU intensive games like Cyberpunk 2077 or even War Thunder?
I forgot to mention: I play with a 3400x1440 resolution. Thats’ twice the pixels of 1920x1080.
RMAs are such a copout. These are little issues that can be easily fixed with some tinkering.
My 6800xt has bad hotspot temps and I’ve wanted to give it new pads for awhile. It has only gotten bad in benchmarks though (100-110C where GPU temps are 70max).
That’s until recently when I started playing CS2 and this is the first game that has started to push my hotspot temps near 100C in actual use.
My question is do you or anyone know what kind of performance/stability improvements I can get by getting my hotspot temps under control?
If your card is reaching 110C, it will start to throttle down to not exceed those temps, the amount it throttle varies based on how much it needs to back down to keep it in the 110C max tjunc temp.
That large of variances is usually due to uneven contact between heatsink and the die. Some of the 7000 series had these issues as well which some people tried to correct with repasting etc, though it made little sense to do that as cards were new and should be RMA’d. GPU at 70C and hotspot at 110C is not normal.
If that 6800xt is out of warranty, you can repaste it and also inspect your cooler and the card itself to see if you’re getting any type of bowing (curves) in the PCB, this can cause uneven pressure and a poor mount. Other reasons can be just the cooler is not flat etc.
If you’re under 100C on hotspot, it’s not ideal but it shouldn’t reduce performance. I personally would not be comfortable with a 70C GPU and 100C hotspot, 30C delta is a bit on high side.
It’s unlikely that thermal pads alone with contribute such drastic drops, it’s likely that in the process, OP was able to better seat the cooler with even pressure distribution which is what you’re going for. Do some research but people have gotten pretty good results with ptm7950 sheets as opposed to thermal compounds on GPUs.
Repasting and remounting a GPU is far cheaper and far QUICKER than having to RMA it and wait weeks for replacements.
Personally, if you’re into PC building then you should already be used to tinkering with your hardware. Idk why people get so uptight when it comes to fiddling with GPUs.
My question is do you or anyone know what kind of performance/stability improvements I can get by getting my hotspot temps under control?
Besides Games:
- use fancontrol to control your… fans: https://getfancontrol.com/ and create a fan curve that is useable
- if possible, deshroud your GPU. The plastic covers a lot of hot air.
- when you have a backplate on your GPU, check if its thermically connected to the board via thermalpads. if not, get some thermalpads (not the cheap one I had but pads with a high heat transfer of at least 10 W/(m*k). Also be sure, that your backplate is metal not plastic
- cover at least the areas of the VRM and VMEM like in this picture (not mine, I just found it)
- repaste your GPU. Or use these: Thermal Grizzly Kryosheet. But they are a) expensive and b) electronically conductive, so make sure it sits on your gpu and does not touch anything else. But they have a great impact on heat transfer
- Or, if you have the money, get a Raijintek Morpheus 8069. Add some nice fans to it and it should help you
Ingames:
- don’t use TAA with high settings, that will heat up your mem
- use the chill feature that is included in the drivers. You limit the frames, therefore your GPU doesn’t have to work as hard to get high FPS.
- Shadows/RTX in games also use a lot of power… seriously.
Is it MERC or SWFT?
SWFT
I have the same card. It goes up to 110 °C at default settings but never reaches 90 °C when undervolted to 1050 mV.
I’m thinking of replacing the thermal paste with Honeywell PTM 7950 or Upsiren PCM-1. How did you disassemble it? Can you share some insights with me about that?
How did you disassemble it? Can you share some insights with me about that?
AFAIR you unscrew all screws on the backside of the card, that should lose the complete heatsink/shroud construction. From there, you’ll find more screws that connects the shroud to the heatsink.
i put pads at the back of my 6800 XT, initially it helps but after gaming for a while, the backplate gets heatsoaked and it goes back to the same terminal temperatures as before the pads, albeit just a slower climb to reach there.
so no, your “€8 investment” didnt do the heavy lifting, it was the deshrouding with the giant fans that did.
Fans did the thing.
Thermal pads on the backplate affect temperatures, but only to a degree. They certainly can´t affect core temperatures.
Pretty annoying how AMD GPUs are plagued with bad core to hotspot deltas. I got the Red Devil and still have 30 C delta 🥲. I do plan to open and play around with it. For now I just increase the fans. I do have pretty good silicon. Can do 2.5 GHz on the core and 2150 MHz on mem (fast timing) without increasing power limits and lowering voltage down to 1020mV. Usually get around 7-10% more fps.