XFX RX590 bad cooler design and heat with crash issues

This will be a massive post as I’ve done a lot to try and cool this thing as best I could along with a lot of testing. Some cooling changes worked, some didn’t. I will focus on the 2 that made a measurable difference.

First a short summary so people can see that I did cover all the basics.

  1. All drivers and BIOS is up to date.
  2. It is NOT a software problem… it is very much a hardware issue. See more below
  3. I’ve been building and modding PCs since the early 2000’s and I’ve worked as a software dev/engineer for over 15 years now. I am quite capable when it comes to building and fixing machines… and not just PCs either.
  4. XFX has a no return policy… so I can’t get my money back and just swapping the card I have for another one won’t change the design problems…
  5. I have never ever had any overheating problems with any of the PCs I have ever had and I used to have a overclocked AMD Socket 939 x2 machine with a passively cooled motherboard and passively cooled GPU. Those reading this who are old enough to know, will know how hot these 939 systems could get…
  6. Posting ‘You should not have bought AMD’ is just a waste of server storage for the forum. Besides, XFX is the guilty party here, not really AMD.
  7. With that said, one would think AMD would keep a closer eye on what their partners are doing.

PC Specs:
AMD FX 8320 with 4.5GHz overclock (It’s not the overclock!)
ASUS Sabertooth 990fx r2.0
16Gb DDR3 HyperX Fury 1866 CL9
Superflower 550W PSU
120Gb ADATA boot SSD
250Gb Samsung SSD for some games
2x500Gb HDD in RAID 0

First, I am very capable when it comes to machines in general. In South Africa… you have to be. Buying new tech here is very expensive.

Custom CPU brackets...

This XFX RX590 was on a great special which is why I bought it, but even so, the cost amounted to about the amount I spend on fuel just to get to work and back for 4 months. In Africa, you fix things for as long as you can.

As an example, I got my hands on a massive phanteks cooler that came off a Intel system. I designed some simple mounting plates and now I run a 4.5GHz overclock.

The donor plates are used in server cabinets to hold switches that are smaller than the width of the cabinet. Anyone who has worked with these things will know that this is hardened steel and does not bend easily.

So what exactly is the issue. In short, heat.

African summers... keep this in mind when looking at the XFX card...

In South Africa, this is a common temperature during summer:

Air conditioners is just as expensive here as most other tech. Then there is the cost of running these things. Our power utility is a plaque on this country and the cost of electricity went up close to 500% since 2007 and they still don’t have enough money to prevent load shedding. So no, I can’t keep my room/house at a nice 21°C at all times…

I knew the card ran hot before I bought it. I knew 225 watts is a lot, but other cards pulling that much power, or more in some cases, seems fine with air coolers. No reason this card should not be fine either… right?

First 2 weeks or so the card was fine. Yes it constantly ran at 80c under load, but from the reviews online this was normal. One day while I was playing Atlas, the PC crashed… locked up and screen went black. I assumed it was the game since it was still new and bugs should be expected. So I hit reset on the case, but the display did not come back even though I could tel the machine was booting and loading windows.

Straight away I thought this was a little strange. If it was a software issue, hitting reset should have solved the lack of display output. So I hit the power button and after it shutdown I started up the machine again. Now the display output came back. Already this felt like a hardware problem.

A few more weeks go by and I had about 3-4 other random crashes. It seemed kinda random and so sending the card to the retailer would result in them not seeing the issue and telling me the thing is fine. I had to find a reliable way to get it to crash before wasting my time sending the card back for testing.

I tried Furmark and superposition for HOURS with no crash. I also set the CPU back to stock and still got a crash. It’s not the overclock on the CPU.

One day I decided to load up a game I have not played in a while, War Thunder. After about 1 hour of play the system crashed again. I restart the machine and go straight back into the game… took maybe less than 2min. Then after about 10min of game play it crashed again. This already suggests a heat soak issue…

Testing Testing Testing...

I did some more testing and sure enough, I was able to get it to crash somewhat reliable with War Thunder. I also saw in AIDA64 that the power draw would sit around 200-210w with the game, but only got to about 190w with Furmark and only 195w with Superposition.

I know this number in AIDA is not super accurate, but it never the less shows that there is a power draw difference.

While I was doing my testing, I got a warning I did not expect at all. The ASUS monitor software brought up a warning about my PCIe-1 slot that was overheating. It was going above 60°C. I knew these sensors were there, but it never occurred to me that the slot would overheat. I solved this with some airflow changes we will see later.

AGAIN, this crashing is not a software issue. By this point I was pretty sure it was heat related, BUT… the GPU it self never even got close to the mid 80’s. Also, the fact that after the first crash (after about 1 hour of play) it would then crash after 10-15min of play suggests there is a heat soak problem. Something on the card is getting saturated with heat and it can’t get rid of it fast enough.

So I took the side panel off and placed my trusty Kenwood fan right next to the case. I then played War Thunder for 3 hours with no crash. Max GPU temp was 76°C. Clearly we have a temperature problem, but the GPU never gets close to overheating.

I then ran more Furmark test and took manual control of the fans and lower the RPM so the card would warm up, but even then it barely got to 85 degrees as it started to thermal throttle… like it should… and still no crash.

I also noticed that when the card is left completely stock with complete stock fan settings, the fans would ramp up much earlier and much harder when gaming vs running a furmark test. Clearly the game is doing more to the card and the card knows this since it knows to ramp the fans up more even though the GPU it self will be at 80°C with the game or with furmark.

I don’t have thermal couplers to test exactly where the temp problem is, but below we will see where the problem most likely is…

While searching the web for any answers I found Gamers Nexus and the tear down video they did on the XFX RX590.

What XFX actually did... or did not do...

Couple of notes here first:

  1. Gamers Nexus ran their test in a room that is kept at about 21°C
  2. Any testing done by Gamers Nexus would have been done on a open test bench… as any tech reviewer would.
  3. I have the card in a closed case… like any person would in the real world.
  4. Yes, GN found that switching the bios switch made a huge difference to the temps, but that’s kind of obvious because of point 1 and 2 above.
  5. All the bios switch does is apply a very aggressive fan curve and in a open bench that would make a huge difference. In a closed case, it made almost no difference since max RPM for the card while gaming went from 2800rpm to 3000rpm… and it still crashed.
  6. Gamers Nexus did not experience any crashing issues which makes sens because of point 1 and 2 above…

Link to video: https://www.youtube.com/watch?v=wN0q-BiTve4&t=906s

In that video they found that XFX had used a steel plate to bridge the gap between the VRAM modules and the actual cooler. Steel is terrible as a heat sink. If you’ve ever used a grinder to cut steel you will know how hot that steel will get from the friction of the cutting disc and you will also know how long that steel will stay burning hot after you stop grinding it. If you want a graphics card to suffer from heat soak… use steel…

Straight away I was angry. XFX was clearly cheap here, but to make things worse they did not even put ANY thermal compound between the steel plate and the actual cooler. Furthermore, after I took a closer look at the card and cooler, I found that the cooler doesn’t even cover everything properly. It looks very much like XFX used a cooler design they had lying around and then used the steel place to bridge the gap. How could XFX be this cheap???

I have never had to return any PC component I ever bought. So I decided I will contact XFX and see what can be done to get my money back. This is nothing short of pathetic when it comes to cooler designs. But… this is the response from XFX…
image

Basically… it’s now my problem and my options are live with it or be a unethical douche and dump it on someone else.

Why don’t I just swap it for another one? Because it won’t change anything… unless they changed the design since I bought mine… which they haven’t.

The retailer I bought the card from did try and test the card, but by this time winter has started and they could not get it to crash. By this time the max temp during the day would be maybe 22-24°C and not the 35-40°C about 2 months earlier. So I started working on the airflow in my case.

Airflow stuffs

I tried a lot of things, but will focus on the 2 options that actually made a measurable difference.

First, let’s look at some of the issues. Pretty much all open air GPU coolers will blast hot air in every direction in the case. If you have enough airflow from the case fans, this is generally not a issue.

I started by connecting the case fans to the fan controller that came build in with the case. This way I can set the fans manually to low, medium, or max with a switch at the front of the case. When connected to the motherboard, the fans will only speed up if the CPU gets hot.

In this first image we see the air flow from the one intake fan at 100%. Note, I have 3 frontal intake fans.

In this image the GPU fans are running at 88% and basically canceling out the intake fan. This means the hot air is getting recycled A LOT…

I moved the card to my second PCIe slot and mounted a 92mm PWM fan capable of Tie Fighter levels of power and noise above the GPU outputs. The idea here was to pull as much hot air out as I could. I then made some crude ducting with cardboard to try and direct the blast from the GPU fans towards the rear and direct cool intake air to the bottom of the card. This actually made a difference. Temps under load went from about 80°C to about 75°C with synthetic tests like furmark.

By no means did I ever thought this would be a permanent solution. This is purely experimental.

I then decided to take out the cardboard and put the card back in slot 1, but this time I placed 2x140mm fans in the middle front of the case. There is some steel from the case that obstructs airflow. So the idea here is to try and overcome that without cutting my case up.

NOTE: I KNOW fan stacking does nothing in general except if you go with a counter rotating setup, but even then you will not increase airflow. However, fan stacking will help if there is an obstruction as can be seen in the images below.

This brought the temps down to about 70°C with synthetics tests like furmark. It would still get to about 76°C with War Thunder. Also, the 92mm fan is now mounted below the GPU outputs and it still pulls out a ton of hot air. Again, the card is blasting that air everywhere so this fan is making a decent difference pulling hot air out.

Even though I managed to reduce the temps and I now have a case that sounds like a group of Tie Fighters attacking a Rebel base, it is now mid winter here in South Africa. We see about 18°C max during the day, but easily see double that during summer. I will only know if what I have done to the airflow of the case is enough to stop it from crashing once summer hits, but by then my 1 year warranty will elapse… how convenient for XFX…

Some final stuff and images...

Here we see my brother’s old Sapphire R9 270 that is rated at 180 watts.

Here we see a cooler properly designed to cool all components as it should…

Here we see XFX waste a bunch of aluminium on the looks of the card instead of using that aluminium to cool the thing properly. Note, the back plate on this card is also aluminium… but makes NO contact with the card in any meaningful way. The back plate is just more aluminium wasted on the looks… and we all know when it comes to a graphics cards, how it looks is way more important than how it actually functions…

Final thoughts:
If you asked me if you should buy XFX I would say no for 2 main reasons.

  1. They have a no return policy. Maybe all other manufacturers have this, I don’t know. Would make a good video if some tech reviewers could look into after sales support with the industry connections they have.
  2. It’s very obvious XFX slapped this card together on the cheap and Gamers Nexus also point out the lack of efficiency in what XFX has done here. As such, how can I trust anything that comes from XFX and then recommend it? No.

I will never buy XFX again and everyone I know, or anyone who asks me, will not buy XFX either. They under estimate the influence us tech savvy people have and with the world of social media spreading info about companies behaving badly has never been easier.

Chances of me ever buying a AMD graphics card again is pretty much zero also and again, I will advise anyone I know or anyone who asks to rather buy team green. Why, like I said earlier… AMD should be keeping an eye on what their partners are doing… it’s called quality control.

So far I have not had any more crashes, but again, it’s mid winter here in South Africa now. However, I now have this worry in the back of my head at all times that the card might crash and that crash might be it’s final crash. How can you really enjoy your rig when you have such a worry plague you all the time?

I know people will have more questions, but I might not answer. At this point I have spend wwwayyyyy more time and effort on testing and config changes than any reasonable person would. Almost every Saturday for the last 2 months have been screwing around with XFX’s crap.

1 Like

VRAM chips don’t need cooling really, unless you are overclocking and overvolting them. Lots of video cards don’t fully cover the VRAM chips.
Steel plates, most coolers use them. It’s not like the steel plate is doing the actual cooling. The contact area of the cooler is either cooper or aluminum.
RX 480/580/590 do get hot. Not strange to have them near 80-85C during heavy gaming load, even with good aftermarket coolers (unless you have rather low ambient temps). They start throttling after that (around 95C for the core and 115C for the VRM).

So after some discussion with the retailer they arranged to send the card back to XFX for testing. Took a few weeks, but I got a call on Thursday (15 Aug 2019) from the retailer and while XFX was testing the card it crashed… I had a good laugh at that point…

The main reason for the call was that XFX wanted to know if they should send me a new card. Obviously I said no. I’m not sure what arrangement the retailer has with XFX, but I got store credit and got a ASUS GTX 1660ti instead even though I had to pay a little extra to make up the difference in cost. I should get the new card by Wednesday.

XFX did not want go into any real detail about what is actually wrong with the card… which seems odd to me, but whatever… after months of screwing around with setup changes and arguing with XFX, it’s over.

I know you can cool VRAM with air if you leave them stock and I knew very well what the temps looked like for the 590s before I bought the card, but in this case the steel plate will trap heat. Steel should not be used in this manner… if you use steel for structural reasons and not cooling, then it’s fine. Here we see XFX use steel to cool the VRAM chips with thermal pads between the chips and steel, but NOTHING between the steel plate and the main cooler… even though they should not be using steel like this in the first place.

How cheap do you have to be to:

  1. use steel like this and
  2. not bother with a thin thermal pad between the steel and main cooler.

What is really frustrating about this design is that XFX used a ton of aluminum for the back plate which does nothing for cooling… if they used just a little less aluminum for the looks, they could have had decent cooling for the memory.

Last thing I would like to point out. I found that the XFX rx580 is exactly the same as the XFX rx590. So XFX did a shitty design on the 580 which is 185watts and then used the exact same design on the 590 which is 225watts.

No matter which way you slice it, XFX screwed up here big time. My card was right on the edge and sometimes crashed and sometimes not.

Anyway, it’s now sorted as per my previous post.