[SOLVED] GPUs may want more power than you think?! problems with RX 580, RX 590 etc

gn842a

Honorable
Oct 10, 2016
666
47
11,140
[EDIT: this is a failed hypothesis, see next post. It may be that there are users with insufficient watts in the psu out there, but the problem I've been having was not solved by more power.]


toms hardware analysis of RX 590

radeon rx 590 power consumption on Anandtech

So....looking around on different fora, I am not alone in having epic problems with keeping the Nitro RX 590 stable. I was messing with drivers, going in to UEFI and trying to reduce voltage, messing around with AMD software and fiddling with things I did not understand and getting into trouble--driver installations up the wazoo. But it was a member of these fora who, looking at my video of a crash using Unigine Superposition run, commented: "If I didn't know you had a 850 watt Seasonic I'd say that was not enough watts in the psu." Bing bing bing! For me that was the best answer of the week. I better go back and award best answer to that man. I am grateful.

So I did a couple of things. in no particular order. First off, I did a careful, not a casual, assessment of my psu requirements. I have been in the habit, for example, of leaving out things that I didn't think were significant, including my anachronistic habit of keeping two DVD/RWs on the build. When I did that, it turns out that my "recommended" psu level from Corsair was 700 watts. Well, I have 850. No problem right?

Wrong? ! I believe that these power estimators are somewhat outdated. As graphics cards grow more hungry, their variance increases. So the RX 590 runs at about 250 watts but it spikes to 300 or even 360, depending on whose measurements you believe. By contrast, the R9-380, that I've been using without incident and which did not crash during any benchmark, caps at around 200 watts. I don't know its spike level, but I do know I can run Unigine Superposition on the R9-380 gpu all day and it never seems to mind. By contrast, my brand new RX 590 would seem to be fine for normal usage and even a couple of the older bench marks (UserBench), but it would crash often during Passmark's transition from 2d to 3d and almost always as soon as Unigine Superposition would launch and "as often as not" with Unigine Heaven.

Apparently the psu needs to be sized to the variance, that is to the worst spikes. Higher performing psus put out a lot more watts than they did a few years ago, and I'm not sure that the "how to build your pc" utilities on the net have kept up. They are reasonable for average types of uses and demand, but you have gpus now that are working 0 to 400 watts and 0 to 500 watts rather than 0 to 200 and less. And though their average watt output may be much lower, the spikes will kill performance. The gpu sends a spike-y demand for power, the psu says "sorry I can't do that," and the gpu responds by sending you into a green soup, a black hangup, or some other behavior.

One thing I've learned from a different hobby (astronomy gear) is that computers with insufficient power never quite fail in the same way. They act like drunks, and do many unpredictable things. Green screen, red screen, black screen, it's the luck of the draw. Maybe you'll get a freeze. Anyhow I think many users (like me) have been often been chasing things like UEFI adjustments, updates, repeated driver downloads--all good stuff to do. But it might be the case that the new "important thing to consider" is the power maximum of the psu. Really, I can't say when one should pursue a software fix (new drivers, update bios), it's certainly cheaper to do, though some activities (uninstalling gpu drivers and flashing BIOS come to mind) are a bit stomach churning for some of us. There might be a certain appeal to pursuing the cost free solutions first, though it's always dicey if you can brick the system. The alternative of buying a new psu may not have much appeal either. But I must say I was really startled to see this Nitro RX 590, which had been giving me nothing but grief, snap to attention and perform on a 1050 watt psu. So I'm putting it out there that you can be underpowered at 850 watts.

Greg N
 
Last edited:
Solution
No no. I was just asking if you made sure to limit variables during your testing. Some driver versions can be buggier than others just as game/software patches can fix bugs. Also not all bugs get fixed in the very next patch/ update. I wasn't implying that you need a new GPU driver for a new PSU.
Depending on how long you fought these issues before the PSU swap (and how diligent you are about updates) I just wanted to explore the possibility that you may have inadvertently fixed the issue with an update right around the same time you swapped the PSU.
I put a header in my previous post to warn the TLDR crowd that this idea did not work out. It is true that my gpu card worked very well on the upstairs build with a 1050 watt power supply.

But when I installed a 1300 watt Seasonic psu on my downstairs build the system continued to crash during benchmarks.

And so on to more experiments, to be shared in another thread. --Greg N
 
This is why I only buy PSUs rated at 1000W and are at least 80plus Gold Certified. You can generally get good ones on sale and they keep your options open. I'm running two R9 Furies in Crossfire. They pull 700W+ under extreme load but my EVGA 1000W G2 Supernova handles them like a champ. LOL
 
  • Like
Reactions: gn842a
This is why I only buy PSUs rated at 1000W and are at least 80plus Gold Certified. You can generally get good ones on sale and they keep your options open. I'm running two R9 Furies in Crossfire. They pull 700W+ under extreme load but my EVGA 1000W G2 Supernova handles them like a champ. LOL

Well now that I have a 1050 watt psu upstairs on the secondary computer and a 1300 watt downstairs I certainly don't have any regrets about it, that's for sure.

And my son is happy because he will get the 850 watt which I pulled out of the downstairs build. That's a whole lotta psus.....

I gave serious thought to EVGA but my personal experience has been with Seasonic so right now sticking with that (and the Thermaltake upstairs).

GN
 
  • Like
Reactions: Avro Arrow
Not sure I'd completely agree with these statements....but, to each their own. I certainly wouldn't discourage anyone from buying a little more PSU than they absolutely need. But, even with the intermittent 300W spike from a full-auto RX590, you'd still have 550W extra on your 850W PSU to use for CPU spikes. That's unlikely.
Also, are you dual-ripping optical media while you're gaming?
Were you running the same AMD driver version on both PSUs? Any updates to the games/ programs being used to stress test?


FYI, those anand graphs are SYSTEM power consumption of 380W with a 4.3GHz i7-7820X (140W TDP), not GPU-only.
 
Last edited:
  • Like
Reactions: gn842a
Not sure I'd completely agree with these statements....but, to each their own. I certainly wouldn't discourage anyone from buying a little more PSU than they absolutely need. But, even with the intermittent 300W spike from a full-auto RX590, you'd still have 550W extra on your 850W PSU to use for CPU spikes. That's unlikely.
Also, are you dual-ripping optical media while you're gaming?
Were you running the same AMD driver version on both PSUs? Any updates to the games/ programs being used to stress test?

FYI, those anand graphs are SYSTEM power consumption of 380W with a 4.3GHz i7-7820X (140W TDP), not GPU-only.

I'm not sure I'd agree with my own statements at all. I can just tell you that it is hard to chase something like this down if you're the kind of person who does a build every four or five years instead of four or five times a week. In the fora one gets a variety of types of information and it is hard to sort the relevance. But it is all that is out there. In the home one runs experiments and tries to do what one can to figure out what is going on. I was 99% sure that the ability of the RX 590 to run upstairs was due to its being on a build with a 1050 watt psu.

This certainty was in part because of the near instantaneous nature of the crashes that would occur at specific moments: When I initiated superposition, for example, and when Passmark went from 2d to 3d mode. It seemed to me that there was an increased power draw and that the upstairs build could provide it and the downstairs build could not.

I disproved my own hypothesis. I was wrong. This is unfortunate, because it cost me a heavy duty psu to do so. But I've spent money on worse things. And my son will get to collect the unemployed psu. I already promised him a good psu in the $150 range so I was committed to a third psu, he gets the 850 watt Seasonic, I keep the 1300 watt, and I'm out $100 more than planned (not $250 more than planned). And I have the knowledge I was barking up a wrong tree.

My crashing downstairs build may be due to some combination of (a) multiple screens (b) obsolete or inadequate cables and (c) too much distance from pc to the "far" screen--which is where I do 95% of my usage. Experimenting with one screen on, then the other, then both has led to some results that are reasonably replicable. The cable is a plausible hypothesis. It's a good thing that I'm used to my astronomy hobby expenses getting out of control or this would be driving me crazy. So anyhow, now I'm in $45 for a displayport cable.

And thanks for pointing that out about the Anand article's data. And good lord! I'm supposed to be running a different AMD driver because I have a different psu? I don't understand that. All the drivers on this build are a few days to two or three weeks old.

thanks
Greg N
 
Last edited:
No no. I was just asking if you made sure to limit variables during your testing. Some driver versions can be buggier than others just as game/software patches can fix bugs. Also not all bugs get fixed in the very next patch/ update. I wasn't implying that you need a new GPU driver for a new PSU.
Depending on how long you fought these issues before the PSU swap (and how diligent you are about updates) I just wanted to explore the possibility that you may have inadvertently fixed the issue with an update right around the same time you swapped the PSU.
 
  • Like
Reactions: gn842a
Solution
Make sure for each 8 pin connector you come with a new cable from the PSU, for example dont use cables that have a 2x8 pins splitter at the end, use separated cables for each 8 pin, if your videocard has 3x8 pins connectors, come from the PSU with 3 independent cables. There are some warning on seasonic PSU (for more then 300W use independent cables if i remember correctly).
 
  • Like
Reactions: gn842a
No no. I was just asking if you made sure to limit variables during your testing. Some driver versions can be buggier than others just as game/software patches can fix bugs. Also not all bugs get fixed in the very next patch/ update. I wasn't implying that you need a new GPU driver for a new PSU.
Depending on how long you fought these issues before the PSU swap (and how diligent you are about updates) I just wanted to explore the possibility that you may have inadvertently fixed the issue with an update right around the same time you swapped the PSU.

The fact of the matter is that when you change more than one variable at a time you cannot establish causality for any one variable. I actually read this in a philosophy class. That was decades ago, it made an impression and it seems to be true. In any case I raise the point because--well, I think we often change more than one variable at a time because we go insane if we do it any other way. The process of arriving at a fix becomes too slow if it is too methodical.

Nonetheless any time I made major changes I ran some tests before going on to the next step. And I consider changing drivers a major step, which I tested after installation, and then separate test after psu. Because I am also trying to follow the rule of least effort and least expense. Drivers are high aggro but low cost so they were first in the queue. Swapping the psu was fairly high cost so it came later. Very, very high aggro and very high cost is swapping mobo or OS. I think I would probably just continue to use my old R9-380 rather than do either.

Right now I'm in a holding period, the 590 went back for RMA and I'm getting some new cables. I think I'm going to switch to displayport possibly for both screens.

Thanks for helping me think through this,
Greg N
 
Make sure for each 8 pin connector you come with a new cable from the PSU, for example dont use cables that have a 2x8 pins splitter at the end, use separated cables for each 8 pin, if your videocard has 3x8 pins connectors, come from the PSU with 3 independent cables. There are some warning on seasonic PSU (for more then 300W use independent cables if i remember correctly).

Dragos thanks for taking the time to write. I did learn that lesson somewhere during my experiments: use two separate cables, one for each port on the gpu, and using two separate power source ports on the psu. That strikes me as a good idea.

I would add, however, that on my upstairs build the Nitro RX 590 works fine with one power cable with a splitter for two 8 pins. I don't know why. I just went ahead and tried it that way a few days ago. The nitro works upstairs in any configuration, near as I can see.

The main thing going on upstairs seems to be one screen. Other things that are different up there are Win 8.1 and the mobo. But the screen issue is really the only thing I've identified where I can replicate this crash problem fairly consistently.

thanks again,
Greg N

p.s. I don't have a GPU with three 8 pin connections, thank goodness!
 
What do you guys think about monitor drivers? I didn't even know monitors had drivers.....

Edit: I was able to find drivers for the Dell monitor that has been crashing. The TV I'm using (as a giant screen) is Panasonic TC-L50EM60. It's just using Windows default PnP drivers. -Greg N
 
Last edited:
Not all PSU cables are identical, some might be a little bit thicker some may be thinner, thats why upstairs works (but i think it is at limit) and downstairs doesnt, also the modular cnectors inside the PSU might be a little different, some might allow and amp or two extra but again it is at borderline. Safest way for a high power graphics card is to use a cable for each connector. From my point of view monitor drivers are useless, windows detects very well and i did not have trouble with that on any monitor or tv that i used or my friends or family.
 
Not all PSU cables are identical, some might be a little bit thicker some may be thinner, thats why upstairs works (but i think it is at limit) and downstairs doesnt, also the modular cnectors inside the PSU might be a little different, some might allow and amp or two extra but again it is at borderline. Safest way for a high power graphics card is to use a cable for each connector. From my point of view monitor drivers are useless, windows detects very well and i did not have trouble with that on any monitor or tv that i used or my friends or family.

If I do one screen, close to pc, on the unstable downstairs computer, it is stable. It is two screens and one screen at distance that seems to be the issue. I've had two different psus and each with their own set of cables on the downstairs build and nothing has changed. Both were Seasonics, not cheap stuff.

There are a large number of posts on the net for things like "RX 580 crashing" or "RX 590 defects" and you can read a lot of people advancing a lot of different theories and trying things out. Many of the solutions I've tried, some I have not. I have a suspicion that if I get an NVIDIA I won't be having these problems.

So when the RMA comes back I'll give it a try on the downstairs/unstable computer but if, as I think is almost certain, it also crashes, I will put it up on the more stable upstairs computer, and look around for an NVIDIA on sale. Or just continue to use the R9-380 which has proven itself to be a reliable work horse.

Meantime I am switching to displayport on my unstable/distant monitor on the unstable/downstairs computer so that might change a thing or two as well.

One good thing that came out of all of this, after years of having a display on the downstairs computer (TV screen) that never fit quite right, I was able to use the Radeon HDMI setting to get the TV screen to finally scale the desktop to actually be completely visible. It was always annoying having icons and the windows tab at lower left off the edge of the screen.

It is worth bearing in mind that I'm doing all this correspondence on the unstable/downstairs computer and I use the remote screen and the near screen and can do benchmarks with either one or both on simultaneously with no crashes at all -- on the Asus Strix R9-380. The R9-380 runs more reliably on the RX 590 Nitro's drivers than the Nitro does.

The R9-380's success makes me skeptical of theories that blame some aspect of the PC (drivers etc) or the hardware. But I don't mind trying a few tricks like using displayport instead of DVI-D for my remote monitor. Displayport is supposed to have more capability and it isn't too expensive.

thanks,

Greg N
 
Well we can put this long saga to bed. I tried one Asus RX 580, and two RX 590s on my build and they simply could not make it through a benchmark without crashing. After weeks of trying various solutions about the best I could do was: the Nitro RX 590 would crash coming out of sleep mode but not most benchmarks, except it had what I call a "white soap crash" where the colors would be washed out but the screen was usable.

Now I have an MSI 1660 TI and it is just as functional, so far, as the R9-380 it replaced. Because all I wanted was a new graphics card that worked as reliably as that R9-380.

I won't say one thing or another about whether I'll buy another radeon but for sure I'll be looking at crash reports more seriously no matter what I buy. The 1660 TI had racked up several dozen positive reviews on Newegg and reading through both positive and negative (which were very few) there no reports of crashes. I'm glad to get through a dozen hours without a crash....

Now what to do with the RX 590...

Between buying two graphics cards to do the job of one and testing every permutation I could think of, new psu, new cables....I think my $1200 build is getting to be more like $1800. I guess I'm lucky I have the resources to keep at it till I found something that works (let's hope it works for the long haul).

Greg N