Question Gigabyte R9 280X Windforce is badly overheating

[Fardringle]

Granted, this card is old. But the 280X still works very well for some distributed computing projects that make use of its double precision calculations.

Unfortunately, this card has a problem and I don't know how to fix it.

Under full load from the MilkyWay@home distributed computing project, the card spikes up to 97C frequently, which causes a driver crash that reverts the card back to 501Mhz, making it very slow and inefficient, and requires a complete reinstall of the drivers to get it working again at anything more than 501Mhz. If I use MSI Afterburner to manually force the GPU to 700Mhz and RAM to 1300Mhz (stock is 1000/1500), then the card will run steadily, albeit a lot slower than it should. It overheats and crashes at anything higher than 700Mhz. It still gets quite hot at 700Mhz, but not hot enough to crash.

This is in a room with a very cool 20-22C ambient temperature, and a case with excellent cooling. The Ryzen 9 3900X in the same case, also under full load from DC projects, stays well below 70C. A single fan GTX 1060 also runs cool in the same case at 100% load. For testing purposes, I removed the side panel from the case and pointed a very high volume desk fan directly at the GPU and the temperatures only dropped by a couple of degrees at most and sometimes there was no difference at all, so it seems to be getting plenty of ventilation in the case.

All three of the GPUs fans seem to be spinning well and the heat sinks are clean. I've used a custom fan curve in MSI Afterburner to force the fans to 100% when the card gets over 65C, and I have replaced the thermal paste with good quality Noctua NT-H1. And I even pulled it out and replaced the good stuff again just to make sure it was applied well.

The only thing I've noticed that seems like a potential fix is that the thermal pads on the VRM and RAM chips are worn and dirty, but I don't know if they are actually overheating or not since the card doesn't appear to have separate temperature sensors for the GPU and the VRMs. I'd be happy to try replacing the pads if it will help, but don't know what product would be best to use.

I'd appreciate any suggestions you can give to help me get this card working properly again!

*edited to add a little more detail to the post*

 

[Fardringle]

Is it possible that the heat pipes in the cooler (or something else) have gone bad? If so, how would I verify that?

 

[EliteRetard]

Initially I was going to say remount/repaste if adding extra airflow isn't helping, thinking heat wasn't getting into the heatsink. But since you've already done that... Then I suspect it's a limit of the GPU cooler and "dry out" of the heatpipes. Basically there's to much thermal load and the liquid isn't making it back to base before evaporating again. That or the mounting surfaces are very uneven, I can recall a few newer GPUs that were such an issue that thermal pads were used on the core. Actually I think AMD is still using those pads even on the newest RX6000 cores.

I suspect you don't want to invest in an old card, like adding a water cooler. Have you tried to undervolt as much as possible?

There is a chance that the VRMs are overheating, and the hotter they get the less efficient they are, which might contribute to higher power and heat throughout the card? I guess thermal pads are fairly cheap, wouldn't hurt to try changing those. Measure the thickness of the ones you have, and add a bit (maybe 0.5mm) considering they are already used/squished. You don't want pads that are to short (no contact), nor to tall, as that will interfere with the mounting position/pressure on the core. Speaking of which, you could try some super thin washers (or like 1 to 2 sheets of paper) under the core screws to increase mounting pressure. Gotta be careful with that, to much can crack the die. Tighten the screws a little at a time each, in a criss-cross pattern.

Does your card have a backplate? You wont get a ton out of the back of a GPU, but if you're hitting thermal limits a couple C can help. I'd suggest putting a thermal pad over the back of the GPU core area. Either to conduct to a backplate, or zip tie on some metal (got any old chipset/MOBO heatsinks or anything you can salvage?).

Oh and as for which thermal pad, again look for the right height first. Then look for W/mK, higher is better but consider value as well (don't spend much on an old card). 6-8 W/mK is probably what's used normally for VRM/RAM, while I have seen pads as high as 60+ W/mK (spendy). Something like 10-20 W/mK would be pretty fancy, without being to exotic/expensive. Thermal pads come in almost any size/shape but even getting a bigger square of it and cutting what you need would work. You might need to measure the area of all the current pads if you are trying to get an exact fit or can't easily estimate by looking.

What the heck, here's a few random name brand examples (randomly picked size/thickness):

6 W/mK

https://www.amazon.com/ARCTIC-Compound-Efficient-Conductivity-Handling/dp/B00UYTTMNI/

Amazon.com: Fujipoly/mod/smart Premium G-M Thermal Pad - 60 x 50 x 1.0 - Thermal Conductivity 6.0 W/mK: Computers & Accessories

Amazon.com: Fujipoly/mod/smart Premium G-M Thermal Pad - 60 x 50 x 1.0 - Thermal Conductivity 6.0 W/mK: Computers & Accessories

www.amazon.com

8 W/mK

Amazon.com: Thermal Grizzly Minus Pad 8 Thermal Pad, 100 × 100 × 1.5 mm: Computers & Accessories

Amazon.com: Thermal Grizzly Minus Pad 8 Thermal Pad, 100 × 100 × 1.5 mm: Computers & Accessories

www.amazon.com

11 W/mK

Amazon.com

Amazon.com: Alphacool 12452 Eisschicht Thermal pad - 11W/mK 100x100x1mm Thermal Pads & Paste Thermal Pads: Computers & Accessories

Amazon.com: Alphacool 12452 Eisschicht Thermal pad - 11W/mK 100x100x1mm Thermal Pads & Paste Thermal Pads: Computers & Accessories

www.amazon.com

14 W/mK

Amazon.com: Alphacool 12458 Eisschicht Thermal pad - 14W/mK 100x100x0,5mm Thermal Pads & Paste Thermal Pads: Computers & Accessories

Amazon.com: Alphacool 12458 Eisschicht Thermal pad - 14W/mK 100x100x0,5mm Thermal Pads & Paste Thermal Pads: Computers & Accessories

www.amazon.com

17 W/mK

Amazon.com: Fujipoly Ultra Extreme System Builder Thermal Pad - 60 x 50 x 1.0 - Thermal Conductivity 17.0 W/mK: Computers & Accessories

Buy Fujipoly Ultra Extreme System Builder Thermal Pad - 60 x 50 x 1.0 - Thermal Conductivity 17.0 W/mK: Heatsinks - Amazon.com ✓ FREE DELIVERY possible on eligible purchases

www.amazon.com

After that quick glance, my brain is saying the Arctic 6 W/mK and the Fujipoly 11 W/mK looked like decent value. 'Corse that'll depend on what thickness and size you choose...so you'll just have to look around and compare. I don't know how well other brands compare, if there are budget brands that are good. I tend to stick with names I recognize and feel like I can trust to be accurate and decent quality.

 

[Fardringle]

Thank you for the very thorough answer! I'm going to answer your comments first, and then add my own comments at the bottom.

Initially I was going to say remount/repaste if adding extra airflow isn't helping, thinking heat wasn't getting into the heatsink. But since you've already done that... Then I suspect it's a limit of the GPU cooler and "dry out" of the heatpipes. Basically there's to much thermal load and the liquid isn't making it back to base before evaporating again. That or the mounting surfaces are very uneven, I can recall a few newer GPUs that were such an issue that thermal pads were used on the core. Actually I think AMD is still using those pads even on the newest RX6000 cores.

I just pulled the card out again and removed the heat sink assembly again to look at the thermal pads, and the thermal paste on the GPU core seemed to be very evenly pressed/spread on both the core and the heat sink so I think it's getting an even mount there...

I suspect you don't want to invest in an old card, like adding a water cooler. Have you tried to undervolt as much as possible?

I considered water cooling, but don't want to spend that much for a really old graphics card. The only way to undervolt this card is to use a custom BIOS, and I'm a bit hesitant to do that unless I don't have any other options. I don't have budget for water cooling, and definitely don't have budget to replace the card completely.

There is a chance that the VRMs are overheating, and the hotter they get the less efficient they are, which might contribute to higher power and heat throughout the card? I guess thermal pads are fairly cheap, wouldn't hurt to try changing those. Measure the thickness of the ones you have, and add a bit (maybe 0.5mm) considering they are already used/squished. You don't want pads that are to short (no contact), nor to tall, as that will interfere with the mounting position/pressure on the core. Speaking of which, you could try some super thin washers (or like 1 to 2 sheets of paper) under the core screws to increase mounting pressure. Gotta be careful with that, to much can crack the die. Tighten the screws a little at a time each, in a criss-cross pattern.

The thermal pads appear to be about the thickness of heavy printer paper. Not quite card stock, but thicker than normal paper.

The screws tighten down on springs so I'm not sure I could put anything else under them, unless I remove the springs. I suppose that's a possibility for something to try if needed.

Does your card have a backplate? You wont get a ton out of the back of a GPU, but if you're hitting thermal limits a couple C can help. I'd suggest putting a thermal pad over the back of the GPU core area. Either to conduct to a backplate, or zip tie on some metal (got any old chipset/MOBO heatsinks or anything you can salvage?).

The card does not have a backplate. There are actually dozens (maybe even a hundred) small components attached to the PCB directly behind the GPU core, so I'm not sure if putting thermal pad there is a good idea or not...

Oh and as for which thermal pad, again look for the right height first. Then look for W/mK, higher is better but consider value as well (don't spend much on an old card). 6-8 W/mK is probably what's used normally for VRM/RAM, while I have seen pads as high as 60+ W/mK (spendy). Something like 10-20 W/mK would be pretty fancy, without being to exotic/expensive. Thermal pads come in almost any size/shape but even getting a bigger square of it and cutting what you need would work. You might need to measure the area of all the current pads if you are trying to get an exact fit or can't easily estimate by looking.

Since the smashed pads are VERY thin, I assume that would mean I want the thin (0.5mm) pads..
.......

Now that I've gone through all of that, I have a very unexpected update to add.

When I pulled the heat sink off to check the pads, I of course had to put new thermal paste on the core. And since I did that, the computer has been running for about 30 minutes with the 280X under 100% load, and it hasn't ever gone above 67C!

If I hadn't ever replaced the thermal paste before, I'd blame old paste. But I've done it twice recently, and I carefully checked the alignment both times, so I'm sure it was installed correctly those times as well. I might have put a little bit more paste on this time than the previous times, but that shouldn't make a 25-30C temperature difference..

I'll let it keep running as is for an hour or two, and if it stays in the mid 60s I might try putting it back at the stock 1000mhz speed to see what happens.

I really don't know why it is suddenly behaving now, but I'll definitely take the results!

 

[Leeea]

If I hadn't ever replaced the thermal paste before, I'd blame old paste. But I've done it twice recently, and I carefully checked the alignment both times, so I'm sure it was installed correctly those times as well. I might have put a little bit more paste on this time than the previous times, but that shouldn't make a 25-30C temperature difference..

Always, 100% of time add far more thermal paste then you will ever need.


Gamer's Nexus did a video* up about this. Basically, if you dump the whole tube on, you will get within about 1C of ideal.

If you do it just right, you will get ideal.

If you mess up doing it just right, your dot does not quite spread out to the entire surface, things can get far less ideal. Which sounds like what happened to you?


*https://www.youtube.com/watch?v=EUWVVTY63hc

18:42** specifically mentions effects of not enough on GPUs:
**https://youtu.be/EUWVVTY63hc?t=1124

 

[Fardringle]

Always, 100% of time add far more thermal paste then you will ever need.


Gamer's Nexus did a video* up about this. Basically, if you dump the whole tube on, you will get within about 1C of ideal.

If you do it just right, you will get ideal.

If you mess up doing it just right, your dot does not quite spread out to the entire surface, things can get far less ideal. Which sounds like what happened to you?


*https://www.youtube.com/watch?v=EUWVVTY63hc

18:42** specifically mentions effects of not enough on GPUs:
**https://youtu.be/EUWVVTY63hc?t=1124

There was only a VERY small difference in the amounts that I used each time, but I suppose it's possible that I just didn't use enough before.

Regardless of reason, it ran steady at 66-68C for almost two hours, so I bumped the card up to the stock 1000Mhz core and 1500Mhz RAM and so far (about 20 minutes), it's been steady at 74-76C. I can definitely live with that!

Thank you both for your input!

 

[EliteRetard]

There was only a VERY small difference in the amounts that I used each time, but I suppose it's possible that I just didn't use enough before.

Regardless of reason, it ran steady at 66-68C for almost two hours, so I bumped the card up to the stock 1000Mhz core and 1500Mhz RAM and so far (about 20 minutes), it's been steady at 74-76C. I can definitely live with that!

Thank you both for your input!

Based on this info I do think this was a mounting issue, because it's direct die tiny changes matter.
Even if you are trying to be careful, it seems microscopic changes can make a noticeable difference.
This is why AMD started using core thermal pads, they couldn't design a reliable/repeatable mount.
I think by remounting several times you eventually got lucky and hit the sweet spot for even pressure.

GN did pressure testing when AMD started using thermal pads, and they found out why they switched.
Paste is better than thermal pads, but only when you have good even pressure across the entire die.
This is why increasing mounting pressure can help to a degree, trying to make a flatter tighter contact.
Conversely, a thermal pad can more easily be forced to tightly form a hard contact on uneven surfaces.

Vega GPU Mounting Pressure Variance & Quality Control | GamersNexus

stub Early reports surrounding Vega GPU packaging indicated minimally two different package processes, though later revealed a potential third. For the two primary forms of Vega GPU packaging, we’re looking at clear, obvious differences in assembly: The silicon (GPU + HBM) is either encased in...

www.gamersnexus.net

 

[Fardringle]

Update just for informational purposes:

It was warmer than normal here yesterday, but even then the GPU stayed well below 80C, so it definitely seems to be working much better!

 

[Fardringle]

Update again in case anyone comes across this thread searching for a similar problem.

So it turns out that a thick layer of good quality CPU thermal paste does work very well, for a week or two, then the card went right back to overheating again, and the GPU core had virtually no paste on it again.

I tried a .5mm thermal pad, but it didn't work at all (card went straight to 100C as soon as the computer was turned on, even without any load on the GPU.

However, I remembered that Dell sent us a bunch of tubes of really thick thermal paste (supposed to be used for warranty repairs that we do on in-house computers) that is virtually impossible to spread out well on a CPU die. It's about the consistency of thick modeling clay or wood putty. Since nobody wanted to use the junk, the tubes just got tossed in the back of a cabinet. I put a blob of that nasty garbage on the center R9 280X (didn't even try to spread it), and surprisingly, it is working EXTREMELY well. The card has been running at 100% load for almost a month now with that Dell paste on it and has not once gone above 77C!

I really don't know why the good quality CPU paste doesn't work well on the GPU, but it's good to know that the thick junk does work really well, at least on this AMD GPU.

 

[VirtualLarry]

Some Shin Etsu variant, perhaps? I think that's what is on Intel stock heatsinks. Some of that stuff is kind of like moulding clay, in consistency, rather than toothpaste, like a lot of various CPU thermal pastes are.

 

[Fardringle]

Some Shin Etsu variant, perhaps? I think that's what is on Intel stock heatsinks. Some of that stuff is kind of like moulding clay, in consistency, rather than toothpaste, like a lot of various CPU thermal pastes are.

Yep. It's Shin Etsu MicroSi.

 

[EliteRetard]

I suspect the thinner paste you used was being affected by "pump out".
The expand/contract cycle caused the paste to squeeze out over time.