Revealing flaws in a GPU?

[mamamia88]

Trying to determine if I got a dud gpu or maybe a bad PSU or god forbid a motherboard.  My Sapphire 480 has been giving me problems where suddenly the screen goes black while under load and I have no choice but to reboot it. I've rechecked the power cables and reseated it. I've undervolted it based on a guide and managed to get furmark to crash but at least the screen didn't go black it just locked up on furmark. 8x msaa to be fair wanted to put it under worst case scenario.  I'm now running furmark for the last 25 minutes after setting a 25mv undervolted 20% increase in max power setting 1750 for ram and setting clock of 1266 as the cap for clock speed. If thats stable I can live without the factory overclock at least for now. PSU a cx550m from Corsair and motherboard is h110m-a2 from Asus. Current temps are 74 c on CPU and 28C across all 4 cores on CPU so that seems pretty much withing limits.  Under 100% load afterburner is saying GPU is drawing 94w right now.   Anyway tldr did Sapphire just get over ambitious with their factory overclock? I dont particularly feel like playing videogames all day to try and force it to crash. That seems like a chore. How long would you consider furmark running for and the system not crashing stable? And are there any more realistic tests I can run to simulate gaming? Because quite frankly I play games for fun and sitting in front of my PC trying to crash it isn't fun. 

[Godlygamer23]

For any kind of testing, avoid Furmark. I hate it when reviewers use it because it stresses the hardware beyond anything that consumers will use it for, and the stress is so bad both AMD and NVIDIA drivers force the GPU to throttle automatically when it detects Furmark or a similar program running. On top of that is that people have had stability in Furmark, but instability when actually playing a game, so bear these in mind when using it.

[Techicolors]

well it's a factory overclock, so it should be guaranteed that the card will be able to run its overclock 24/7. 

 

perhaps you can do an RMA? 

[Cyberspirit]

Do you have the latest drivers?

[mamamia88]

I have the Corsair cx550m which is just as old as the GPU. Anyway 45 minutes going on worse case scenario on furmark with no crashes. I have heaven installed will give that a go

1 hour ago, Cyberspirit said:

Do you have the latest drivers?

Yep even the newest release from the 8th

 

Well heaven is going strong on ultimate settings between 35 and 80 fps completely gorgeous for at least a few cycles now. If stupid thing crashes in me in solitaire or something I'll be completely flabbergasted

 

Well Heaven ran for an hour straight with no problems right after running furmark for 45 minutes straight. These are the settings I've used to achieve at least the illusion of stability. Anyone have a clue as to why lowering ram frequency while undervolted the GPU by 24 and setting a clock speed only 40mhz lower than default would do this? I'm too stupid

[Mick Naughty]

If the system is staying on it isn't the psu.

 

[mamamia88]

13 minutes ago, Mick Naughty said:

If the system is staying on it isn't the psu.

 

Its not possible to just lose power to the GPU and essentially be forced to reboot manually because you can't do anything?  Or couldn't lack of adequate power crash the display driver or something? I'm just throwing random stuff out there because I'm way too stupid to understand this all. I was under impression that display driver crashes would be blue screens though. 

[Mick Naughty]

2 minutes ago, mamamia88 said:

Its not possible to just lose power to the GPU and essentially be forced to reboot manually because you can't do anything?  Or couldn't lack of adequate power crash the display driver or something? I'm just throwing random stuff out there because I'm way too stupid to understand this all. I was under impression that display driver crashes would be blue screens though. 

Seems to be mostly software unless its a clock issue. If it was the psu, the system would shut off if it couldn't keep up.

Reinstalling the drivers with ddu and testing the gpu in valley or heaven and monitor all the specs to narrow down the problem.
When di you get the card? Was it bought new or used?

[mamamia88]

6 minutes ago, Mick Naughty said:

Seems to be mostly software unless its a clock issue. If it was the psu, the system would shut off if it couldn't keep up.

Reinstalling the drivers with ddu and testing the gpu in valley or heaven and monitor all the specs to narrow down the problem.
When di you get the card? Was it bought new or used?

I bought it new a few weeks ago. I just ran heaven @ 1080p the highest settings it let me for an hour straight and it didn't crash. I did the same with furmark prior. Those are supposed to simulate worst case scenario situations. When I ran furmark before I changed settings I was able to replicate the problem in less than 5 mnutes. I'm hoping it's stable now but I guess I need to play actual games to find out

[scottyseng]

1 hour ago, mamamia88 said:

-snip-

Also note that some GPU cards have poor VRM designs that can't handle the current needed by Furmark. It places an immense load on the VRM and some manufacturers cheap out on VRM and will void warranty if you use Furmark.

 

Valley to me is plenty as a stress test.

[mamamia88]

10 minutes ago, scottyseng said:

Also note that some GPU cards have poor VRM designs that can't handle the current needed by Furmark. It places an immense load on the VRM and some manufacturers cheap out on VRM and will void warranty if you use Furmark.

 

Valley to me is plenty as a stress test.

How could they tell? Plus it seems like they would have to disclose that somewhere.  Like warranties are usually void if used in a way that normally wouldn't happen. Pretty sure running stress tests to diagnose a problem to avoid sending it in for warranty in the first place wouldn't count

[scottyseng]

1 minute ago, mamamia88 said:

How could they tell? Plus it seems like they would have to disclose that somewhere.  Like warranties are usually void if used in a way that normally wouldn't happen. Pretty sure running stress tests to diagnose a problem to avoid sending it in for warranty in the first place wouldn't count

That's true, but there are some GPUs on the market that are known to have weak VRM (The components can't deliver the required amperage at the temperature they operate at). If one component fails, your GPU may die.

 

Yeah, it's a grey area for sure...on one hand, yeah, furmark is definitely not a normal full stress load, but on the other, we're kind of needing a good way to stress GPUs.

 

I think there's a 480 (forgot the model) that will not handle furmark because the VRM can't deliver the amperage needed when it gets hot (Which furmark does). It sucks because you need to take the GPU apart, look at the VRM components and know how a VRM works to be able to judge the safety of the VRM circuit of that GPU. Most people don't take apart GPUs for no reason..

[mamamia88]

Well i've been gaming until my hands hurt and it hasn't crashed so i would consider that stable. What could have caused it though? If i'm undervolting and increasing power limit could that point to psu? I also underclocked ram to 1750 instead of 2000 by default