This summer I built a new PC:
Asus X570 TUF Gaming MB
AMD 3900X processor
32GB GSkill 3600Mhz RAM
Seagate Firecuda 2TB SSD
New Corsair RM850x PS
To start, I used my old 1070Ti video card. It was rock solid in all games. Even if you overclocked it to its limit, it would remain rock solid in all games. So, all the other components in this build present no issues. The system is stable.
Then, recently I decided to upgrade to a Gigabyte 5700XT OC card, and the weirdness begins.
At its default settings, you can run the 3DMARK Timespy stability test, over and over, and it will pass. But in games, there are issues. Forza 7 will crash or feeze at least once every two hours. Project cars will occasionally crash or just give you a black screen when the race is supposed to begin.
Trying to overclock gets even funkier. I can increase the power tuning +30%, undervolt to 1150mV, and again it'll hit ~2050Mhz consistently and loop the Timespy stability test forever with no problems. Project cars doesn't seem affected by these settings too much (apart from the random weirdness it always exhibits with the 5700XT), but if I load up Forza 7 with any power increase at all (even +10%), it'll freeze almost instantly - even in menus. Sometimes it'll work fine until I crash the car in the game, and then it freezes. The weird thing is, Forza 7 is not a demanding game, and runs 4K 60Hz Ultra settings and only uses the GPU at 60% load - yet it's crash city. Project cars 3 is very demanding but runs much more stable with overclock settings.
And Wattman seems to be another jumbo headache. It just randomly keeps resetting my voltage back to the default of 1200mV. What's up with this?
But even at default settings, this GPU doesn't seem entirely stable. I'm almost ready to buy a new Nvidia card and chuck this thing.
Anyone got any suggestion?
Sounds like a power issue. AMD recommends a 700 watt power supply for that card. You might want to exchange that power supply for a 750 gold. It will cost a bit more, but the power supply is literally the most important part of you system, and is no place to go with the minimum.
I applaud your choice of new power supply. I have the same one. It gets very good reviews. Mine is 2 years old now and going strong. Best power supply I have ever had. 8 more years on the 10 year warranty too!
If your gpu is not stable now it has nothing to do with the power supply. If you are in your return Window you might want to try another model card. If not you might want to look into an RMA.
You can try these.
+ Connecting two power cables from PSU to GPU.
+ Decreasing 100-150-200Mhz GPU core frequency.
+ Increasing the core voltage which is 1200mV to 1225-1232-1248mV.
Don't buy anything Corsair brand anymore...
Edit: I was also using a Corsair (a 750 Gold) a long time ago. It was used with FX9370 and HD7870 (approx 450w) with a system suitable for its capacity and it was burned.
Those who buy Corsair think they are getting a quality product and this has nothing to do with reality. For this reason, I just warned you.
Thanks for the suggestions.
I've never had a video card that was unstable out of the box!
How do you maintain the undervolt settings? I find Wattman keeps randomly setting it back to 1200mV, even though all my other changes stick. And, yes, I've tried the classic suggested solution of disabling Windows fast starup.
Pretty much all AMD cards need tinkering and often are not stable for the past 3 years now since wattman came on the scene. This was never an issue before that. This is a situation that should not exist. Cards should always work properly at default settings. If you have to lower your speed to get it stable then you are not getting what you paid for.
No stabilization problem for AMD-GPUs without WattMan! Your PSU may not provide GPU-12v required or your motherboard does not provide the required power. When many PSUs sold as good quality load processors, the GPUs-12v drops. GPUs may need higher core voltage. So undervolting won't solve your problem.
Another problem for AMD GPUs is that it doesn't contain quality VRM. AMD overcame this problem with the x570. The next GPU series is likely to come with a higher quality VRM like nVidia.
The rx 5000 series video cards were not planned at such high power at first.
I don't know if you heard, but shortly after the launch they changed the bios to be able to compete with NIVIDIA. The first video cards sold were weaker. Bios update pushed the chips to the limit.
Hence my theory:
because they pushed the chips to maximum they had to raise the voltage, pushed the voltage to the maximum, so it became very sensitive to PSU voltage noise. From here all the recommendations to change the PSU. Because, even if you have a good PSU that goes with other video cards without any problem, if it's not 80+ Gold or Silver it will have a little voltage noise, the rx 5000 chips being at the upper limit of the voltage any more intense noise will send the card in error.
You don't need a 700W PSU because the 5000 series video cards are not very power hungry. I'm sure it works fine with a 500W PSU (I also tested this) but it should be 80+ Gold and it will work.
But what I recommend to all RX5000 owners is to lower their chips from overvoltage.
I'm sure that undervolting will bring you all only benefits, stability, the fan will go slower (hence less noise), the board will go colder (that means longer Turbo) and so on.
Decreasing the voltage without decreasing the speed breaks the stabilization. The suggestion you write to fix GPU in AMD standards causes most users to have more problems probably. Every GPU has an operating voltage range. GPU crashes instantly when voltage is applied outside the operating range. So overvoltage won't hurt a GPU. However, excessive voltage may cause the GPU and VRM to overheat. But it makes more sense to wish something unstable to overcome this problem by heating.
Did you read what I wrote? I work as a mechatronist I know what I'm saying.
You can lower the voltage without lowering the frequency and for some GPUs even a lot, of course not for all cards. And not at all goes, some more others less.
And I tell you from experience I bought Gigabyte rx 5700xt. It was hot and many black screens, I bought two PSU 600W and 700W. The 700W one gave more black screen than the 500W or 600w one I had. And after undervolt it's much colder and benchmarks in all I gained + 12% performance, although I dropped the clock from 20 .. to 1999MHz.
If you knew how a Transistor works, you would understand why high voltage hurts, plus excessive heat. Yes each GPU has a voltage range, exactly what I said AMD has increased this range to compete whit Nividia. They raised the frequency and voltage beyond 100% certainty that all will work.
But not all chips are stable at high voltage. they put higher voltage even though not all of them need such high voltage. They didn't have to adjust each chip separately. Most will be stable and ok, but many will have this problem with the PSU. That's why everyone talks about PSU when it comes to rx5000 series. If you don't believe me, check what consumption the whole computer has and you will see that you don't get close to 500W with consumption, but many still recommend 700W. But too few understand that it is not about consumption but about voltage stability and voltage noise.
I'm sure everyone who had to buy a new PSU and solved the black screen problem could do a little Undervolt and not have to change the PSU.
Too few volts and too many volts is just as bad for the GPU. Overvoltage can have much more serious negative effects on your GPU than undervolt.
Undervoltage is always better. Find the voltage that is lowest and stable for your GPU and that is the best. Don't forget that temperature also plays a very important role in performance, Undervolt is always colder, which means longer turbo time.
Even the undervolt extreme is not good. But my opinion is that because of the duel with Nividia, AMD or hurried and put the range at volts too close to instability.
This is an area that I have always felt AMD is inferior to Nvidia. Currently Nvidia even though it has validation is having issues with its own and its AIB partners choice in caps and other coordinating components with their 3xxx series. Issues very similar to what AMD has had for several years not. Likely though it will get fixed, and fixed quickly by them, if you go by past experience. You are exactly correct in what your are saying about how the electronics work. AMD to my knowledge does not validate its board partners designs to the degree that the green team has historically. This has led to a lot of unstable cards on the market, especially so with Navi. I think too that AMD reference specs are very much at best case scenario. GPUs should be built with a good amount of headroom. This means caps that will control the noise, voltage and heat under peak boost conditions and not cause the driver to reset. Your are correct to that build should be run as designed lowering voltage in many cases can be more damaging than excess voltage. Because of component choices many AIB cards will boost and pull more voltage and go beyond spec causing the driver to reset. I saw similar with my RX 580 as well. It was a OC edition but the bios out of the box only supplies reference spec power. You had to raise the power limit to get the card stable and set a way more aggressive fan curve to deal with the additional heat. Those are issues that should have been default between the hardware chosen and the bios. The end user should not have to tweak to run the card out of the box at defaults.
Regardless of all the reasons Navi has issues and there are many, these cards should work at default settings. If they don't that is the very meaning of defective. There is not a lot of parity in what the AIB partners use on their cards, then you couple with it, that many are offering OC editions. Then you add that AMD gives a minimum and a recommended spec for the power supply when they really should only give the recommended as the minimum. Then doesn't validate the partner cards. This all spells instability for end users for a variety of reasons.
Polaris GPUs implement VBIOS information with 7 or 8 level vcore and frequency value. The level with the highest frequency has a low vcore, but for the next frequency is high. And you solved your problem by increasing the power limit.
You increased the power limit and GPU did not work with high vcore required frequency and your bugs were fixed. That's all. The problem you are experiencing is not in a way that covers every user.
I agree with you. ROG RX 480 was not working properly with AMD Bulldozer processor. Then I bought 8th generation Intel and all my problems are gone. The old system's PCI power was probably weak, or the RX 480 allegedly consumed more power from PCI. However, my situation is not generalized. I just wrote this. Goodbye...
My experience in the past few years is that the most unstable systems are all AMD systems. Navi for instance seems to have less stability issues on Intel than when combined with Intel. Vega was the same way. It defies logic. God help you if you pair an AMD APU and GPU. Good luck.
I should add that I do think that this instability in all AMD systems is not the hardware but again the drivers. For whatever reason AMD drivers just plain don't play nice with others. They can't seem to get them to work with Windows or their own chipset drivers.
I'm using nVidia for now. Doesn't exceed 65 degrees and 1300 RPM. And it is quite satisfactory for 27 inch 1080p. I do not leave my work to luck! The reasons for not choosing vary from person to person. And for me, failure to deliver the expected performance for DX9 and some DX11 games.
And as teoretic mentioned, they are releasing GPUs like OC.
Navi does seem to have vary poor backwards game compatibility. This is in stark contrast to how AMD had alway been in the past. They used to do better IMHO of supporting old games. Navi has changed that a lot. As a person with a ton of games that likes to revisit old titles I completely understand why you would go green.
I'm on the red side. I have used it for years. Now, if my Intel processor is not enough, I would buy a 3rd generation Ryzen. First 64-bit processors were not that successful when AMD introduced them. Anyway... We need to expect GPUs with low power consumption and a lot of cores from AMD like new Xbox GPUs, I think... Goodbye...
The original Athlon 64s were better than the Intel chip of their day. That was the last time the were truly on top. Although for a year there Phenom II was better than first generation i series.
Unfortunately you're right, Nvidia is more stable now (although they also had a problem with the drivers now in the 3000 series) they reacted better.
Although AMD doesn't have a very good reputation with the 5000 series, I'm impressed by how much power they managed to increase after a bios update. If you look at the performance of the first rx5700xt boards and the ones after the bios update, it's impressive. But that's why they can no longer react to problems like Nividia, because AMD knew that this would bring instability but otherwise no one would buy these GPUs. They could at least acknowledge or explain that to us but nah.
Now it's up to everyone to adjust their settings themselves until the cards are stable.
I do not encourage AMD what they did, but I remain in the Red team, I will always be in the weakest team to support them. We need competition, if AMD gives up making GPUs we will all have big problems. Then we will see high GPU prices and we will have no choice.
I am satisfied with the Rx5700xt that I have now, the price per performance is the best. It was frustrating at first with the black screens, but once it's over it's perfect.
a few days ago i bought a brand pc and i didnt realize psu or power supply is too important
after alot searching i got your point
I have a evga 600w plus 80 white and every time i wanted to run a game it crashed
i did all thing over the web such as update windows bios vbios adrenilen but didnt work until i saw your post
yes i agree with you cause of psu i wanst able to run
but what is solution
one get 700w power supply plus 80 silver toward
or just game simply under clock you gpu 1905 and increase voltage 1107
i dont know how voltage u need but u can find best voltage after a few time test
It should not reset itself. Do you have the latest version of Adrenaline 20.9.1? It resets itself only if you have a crash (black screen) or something wrong, that's normal. If it resets without a crash, then try a new Adrenaline installation with DDU.
After making the desired settings in Adrenaline saves a profile in case of crash you do not have to try to remember where the settings were. just load the profile.
Example of OC and Undervolt, look at this youtuber:
If you try this. please tell me if it worked for you and it's stable. I am curious if my theory is correct.
you can try even below 1099 voltages. if it works and does not crash then the card will run colder and turbo longer.
But don't forget undervolt in small steps: start with 1105, 1099,1090
This advice was spot on. The instability comes from temperature getting too high which in turn comes from too much power being sent towards the GPU.
I ended up lowering GPU max frequency to 1853MHz, and max voltage to 925mV, while overclocking VRAM to 1800MHz. Doing so keeps the card temperature at 65-70 degrees C even under heavy loads, and power consumption at about 130W, down from 200W.
Here's my stable settings:
Fixed to my issues when I got launch model ref 5700XT with stock settings used to get 110 Junction and 100 Temp Noise like a cheap vacuum, hadn't had a card that hot since ref GTX480.
Past year+ been running 1900mhz and 950v and 1850vram never crashed since doing that!!
Used to get black screen or lockups etc.
And Performance is higher as well than stock settings..
I could play around more and get more perf prob but it's quiet and stable now so left it like this for over a year.
Thank you very much, it's been months since I bought the gigabyte rx 5700 xt and see that there was no way to make it stable, all the undelvolts and none accepted it, you are a crack !!!!
This is the VRM for ROG version.
This is the VRM for Gigabyte version.
There is a huge difference in quality. And the ROG version will do much better for stabilization. The cause of this problem is AMD. AMD offers the manufacturer a low cost option.
Just picked up a 5700 XT myself. So far no issues to report. I picked up a Powercolor Devil branded 5700 XT. So far I'm still on the 20.4.2 drivers, but I'll likely try out the new ones in a little bit. Any common benchmarks that give issues? I'd like to try it out on my system as well.
I had a lot of issues with DX 9, 10 and some 11 games. DX 12 stuff ran pretty well. Benchmarks all ran pretty well though, heaven, superposition, timespy etc...
Had more issues with features not working right than game drive issues. I however have not had a navi card for 7 months now. So a lot has changed I am sure since then. You will be a good person to check all this stuff though and let us know what you find. Thanks!
So I did a DDU install of the latest drivers today, released two days ago, and it currently seems like all my problems have disappeared!
Holy cow, were the previous drivers that unstable?
Again, I don't want to be misunderstood but I will have to specify. 1-2 days of experience may not be enough to conclude that a GPU is stabilized with a display driver update. Goodbye...
But with the previous drivers I could not play Forza 7 with any amount of power boost without constant crashes, which is weird because that games doesn't use the power boost. It normally runs 60-100W.
Last evening I ran it with 20% power boost all evening, no problem.
Definitely looks though like the drivers really have been steadily improving in the past few updates. Which makes you feel a bit more hopeful about RDNA2 coming out of the gate on better ground than RDNA did.
I appreciate your feedback on your experience. It is easy to concentrate on the negative and is good to give credit when things improve.