PS = GPU killer?

[maddin]

I now have the third graphics card within two weeks while using PhotoScan intensively with OpenCL.

The first card (Nvidia GTX580) constantly locked up my system entirely while showing uniformly colored screens; the second card (Nvidia GTX 650 Ti Boost) gave me weird colored artifacts on the Windows desktop that would change their color with every change to the screen (e.g. mouse move).

Both problems would not go away with a driver update or by booting Windows into VGA mode, so I believe they are hardware related (and they both went away immediately, once I installed a new graphics card).

Both cards were still on warranty, so not too much harm done, but now I am curious to hear whether others have similar experience and what I could do about it.

Thanks
Martin

[Wishgranter]

can test the GPUs in other PC(s) ?? 

[maddin]

The defect cards are already on their way to the vendor, but removing them fixed the problems, and putting them back made the problem return. So I am fairly confident the cards were broken.

Now I wonder whether anyone else using PS regularly has seen similar problems?

Martin

[Exhale]

I wouldn't think so PS is directly effecting your gpu..
There must be other impact on your gpu..  It can be;  high temprature , unstable overclock, PSU can effect.....etc... 

Also, once you got your  GPU damaged on your own system, it will make problem on other systems...

I don't know your gpu's vendor but Make sure, enough cooling, check your overclock settings if there is, double check your PSU.

Every GPU has different idle temprature and working temprature as well as power  consumption..

By the way, do you check your system temprature when it is on load ?

[maddin]

I wouldn't think so PS is directly effecting your gpu..
There must be other impact on your gpu..  It can be;  high temprature , unstable overclock, PSU can effect.....etc... 

Sure. I wasn't trying to say that PS algorithms are the problem. More the intense use of OpenCL (which might happen with other applications, too).

I don't know your gpu's vendor but Make sure, enough cooling, check your overclock settings if there is, double check your PSU.

No overclocking used, PSU is good. The vendor was Zotac.

Every GPU has different idle temprature and working temprature as well as power  consumption..

By the way, do you check your system temprature when it is on load ?

Yes. That was the first thing I checked, using GPU-Z. The GPU temp never got higher than 55 degrees Celsius, which I consider safe. Also, I would expect the GPU to shut down if it got too hot, but not be damaged?

Anyway, I don't want to blame anyone, just understand whether that is something that happens from time to time, or whether I should be worried about GPU #3... ;-)

MartinB

[Exhale]

All right then,
Can you tell me about your computer specs. in details with model and vendor names please.
I will have a look deeply...  Somehow system would be bottlenecked.

[maddin]

My system is an HP Z400, six core Xeon 3.3 GHz, 24 GB RAM

[Exhale]

Hi Martin,

ok, I had a look at your system details from HP offical QuickSpecs that is http://h18000.www1.hp.com/products/quickspecs/13276_na/13276_na.PDF

Your system detail is ;
HP Z400 Workstation
Intel Xeon Processor W3680 3.33 GHz, 12MB cache, 1333 memory, 6.4 GT/s QPI, Six-Core, HT, Turbo
475 watts wide-ranging, active Power Factor Correction, 85% Efficient
RAMs , HDDs, .... etc

You have 80Plus power supply and it is 85% Efficient that means its total consumption is 475 watts , however it is using %15 of 475 watts for its self to make 220v AC  > 12v and 5v DC. So you have aprox 410 watts available left..

As I've seen your GPU consumed a total of 227W under load in tests.
You CPU takes 130 W ,  plus  RAMs , HDDs....etc..     so you are lacking of power wattage..
I also made sure depending to HP's offical QuickSpecs. on this link please have a look at page 17
 http://h18000.www1.hp.com/products/quickspecs/13276_na/13276_na.PDF
Your GPU has almost the same power consumption with tesla c2050.  And this offical Example Configuration shows you need 460.40 W where as your current power supply can provide 410 watts available power.   
This is a big problem on load. It seems you need to upgrade your PSU however your system has custom build PSU and it has different size... it is really unfortunate, I donno how you can solve this..

[maddin]

Thanks very much for the research! I'll open my case and double-check the PSU details.
Do you think too little power could actually damage the GPU? I was expecting crashes and hangs if power fails, but not image artifacts?

Martin

[Exhale]

Honestly, I never faced this kind of situation personally.
However I know insufficient PSU fails the system and even it causes permanently deffective components as a result.

Any news about your PSU ? upload a photo if it's possible.

[maddin]

OK, so I opened my machine (again) and checked the PSU. It is rated as 600W. Attached is a snapshot of it's label.

Martin

[Exhale]

This is getting interesting....  Further investigation needed

[Marcel]

In my experience Photoscan doesn't nearly draw as much power as the maximum rating of the components .  For example,  my system with an i7 (TPD 130 Watt)  and 2x Amd R9 290 does not exceed 450 watt when using Photoscan.  This is with full cpu and gpu load.

*Edit: it actually peaked 600 Watt according to my power meter (which tracks the highest value). But that seems to be very rare.

If you look at the official wattage of the components you would expect a power draw of well over 700 watt.

So the only way to be sure it to hook your system up to an energy meter,  and get the practical result.

[Wishgranter]

dont forget that PSU is aging, approx 10 % a year.... so every year you get 5-10% lower power output from it.....

[jedfrechette]

dont forget that PSU is aging, approx 10 % a year.... so every year you get 5-10% lower power output from it.....

Do you have a reference for this? If true I have a couple systems sitting around that probably shouldn't even be able to boot.