Crashing on heavy load: GPU faulty or PSU giving troubs?

[captain cactus]

Scenario: stress-testing CPU and GPU causes the AMD driver to crash. When it recovers, the GPU stress test (any, FurMark, Superposition, Valley, etc) stopped entirely but the CPU happily continues crunching numbers.

 

Setup: I have a Ryzen 1600 at 3.8 GHz, 1.25V that's survived a 4 hour pass of LinPack via OCCT aka it's stable at that speed. RAM's at 2933 1.4 V that's also stable, all on an ASRock AB350 Gaming ITX/ac with the latest P4.50 BIOS. My MSI reference Vega 56 though, that's a different story. As soon as that gets thrown into the mix, even at stock profile in Wattman, the system will pull along for a couple of minutes at most, if I'm not moving the mouse that is (but even if I don't it will generally crash within a couple of minutes). As soon as I even look at the mouse, the screen freezes, goes black for a couple of seconds and when it comes back the driver has crapped the bed. But as soon as I keep the CPU at rest and stress the GPU alone the system stays intact, mostly. I say mostly because it really depends on the mood my system's in. Sometimes it crashes as soon as load is put onto the GPU but most of the time it keeps running for hours and hours without a single issue.

 

My PSU is a Corsair SF600. I've been wanting to swap that out for something quieter as the fan doesn't return to zero RPM when there's no more load on the PSU making it quite load in the process, but I only wanna do that if it also means it'll solve my GPU issues. I don't wanna RMA my GPU that much because a: it's an MSI model and I can't find any RMA info on their site other than "contact retailer" but more importantly b: I bought the Vega 56 from LDLC.com, a French site, because it was the only place selling MSRP Vega 56 in the EU (I'm Dutch) at launch. Sending it back would mean risky over-the-border shipments with a we-no-speak-english company with no guarantee they'll even have a replacement unit in stock meaning I'd be stuck with no GPU at all. I also don't have a 2nd GPU at hand that I could test with which also isn't nice.

 

So what do I do here? 

[Princess Luna]

1 minute ago, lots of unexplainable lag said:

So what do I do here? 

Stop using FurMark is a good start.

[Paddi01]

2 minutes ago, Princess Cadence said:

Stop using FurMark is a good start.

Because it's unrealistic to real scenarios?

[Princess Luna]

6 minutes ago, Paddi01 said:

Because it's unrealistic to real scenarios?

Yes FurMark is pretty useless truth be told... Not only because it still kills GPUs by today but because what is the point to use it as a stability test for an overclock when it already crashes most cards on stock to begin with?

 

If you want a gaming card the gaming demand will never reach any where near the FurMark demand meaning you could probably be running it at a higher clock and gaining a few extra fps with it opposed to what FurMark would make it seem and so on... There's no reason why to use it...

 

And oh i have so many tales of Fermi, Kepler and Maxwell cards that died because of it.

 

Just stick to Superposition and Heaven and I also dislike Wattman software I'd just use MSi Afterburner instead... either ways a Vega graphics card is known to love sucking power from the wall... a SF PSU certainly is not the best alternative in the market and could be the cause of issues indeed but to be sure we'd have to go through a greater troubleshooting.

[captain cactus]

4 minutes ago, Princess Cadence said:

Yes FurMark is pretty useless truth be told... Not only because it still kills GPUs by today but because what is the point to use it as a stability test for an overclock when it already crashes most cards on stock to begin with?

 

If you want a gaming card the gaming demand will never reach any where near the FurMark demand meaning you could probably be running it at a higher clock and gaining a few extra fps with it opposed to what FurMark would make it seem and so on... There's no reason why to use it...

 

And oh i have so many tales of Fermi, Kepler and Maxwell cards that died because of it.

 

Just stick to Superposition and Heaven and I also dislike Wattman software I'd just use MSi Afterburner instead... either ways a Vega graphics card is known to love sucking power from the wall... a SF PSU certainly is not the best alternative in the market and could be the cause of issues indeed but to be sure we'd have to go through a greater troubleshooting.

Ok, stop using FurMark, got it. To the uninstall button we go.

 

I'll try running Heaven with a CPU load as well trying to trigger the problem. If it does, I'll update.

[captain cactus]

55 minutes ago, Princess Cadence said:

Yes FurMark is pretty useless truth be told... Not only because it still kills GPUs by today but because what is the point to use it as a stability test for an overclock when it already crashes most cards on stock to begin with?

 

If you want a gaming card the gaming demand will never reach any where near the FurMark demand meaning you could probably be running it at a higher clock and gaining a few extra fps with it opposed to what FurMark would make it seem and so on... There's no reason why to use it...

 

And oh i have so many tales of Fermi, Kepler and Maxwell cards that died because of it.

 

Just stick to Superposition and Heaven and I also dislike Wattman software I'd just use MSi Afterburner instead... either ways a Vega graphics card is known to love sucking power from the wall... a SF PSU certainly is not the best alternative in the market and could be the cause of issues indeed but to be sure we'd have to go through a greater troubleshooting.

Ok so I ran Heaven with a static cam on lots of moving things, then ran OCCT on all threads. The entire CPU, so cores and SOC, is pulling around 150W. Put the Vega on a +50 power budget to get it pulling 205W (on powersaving BIOS). FPS was stable at 105, OCCT didn't crash, system stayed responsive the entire time.

 

Then again, I'll try again tomorrow when it could be a completely different story.