PC turn off when run CUDA samples

Accelerated Computing CUDA CUDA Programming and Performance
1

I just install the latest CUDA2.0 driver, tool-kit and SDK from NVIDIA website. when I run the some samples such as N-body and Image denoising(when select 3), my computer turn off suddenly just like I pull out the power. Sometime, this problem happened instantly when I run these samples. Sometime it will last a while before turn off.

I have to say that the windows system is stable and I have updated the driver of video card and motherboard. Additionally, I have updated the BIOS of motherboard.

I will appreciate it if you could give me some suggestions.

my computer specification

Intel Core 2 Quad Q6600 System (2.40GHz / 1066FSB / 2x4mb)
XFX NFORCE 780i SLI MOTHERBOARD
Thermaltake Armor + Black Steel Case with 23cm Fan on Side Panel (No PSU)
CORSAIR TWIN2X4096-PC8500C5DF 4GB 1066MHz Memory (2 x 2GB)
2 x XFX GTX280 Standard 670M PCIe/1GB/D-DVI Video Card
Seagate Barracuda 1TB SATA-II Hard-Drive (32mb)
Silverstone 1000W Power Supply (Black / SLI ready)Samsung 2243BW 22”

2

This sounds like either an overloaded power supply or some kind of overtemp protection killing the power to prevent hardware damage. Two GTX 280 cards draw a lot of power, though one would think that a 1000W PSU should be able to handle it.

Do you have any way to monitor the temperature inside the case? Can you run the samples with only one GTX280 installed?

3

firstly, thank you.

I have used the software to monitor the temperature of GPU, but it happens so quickly that the software can’t update instantly. Additionally, I have try one card, but the problem still exist. So, I am confused.

Just before, I try CUDA1.1, and it seens better than CUDA2.0, the PC only turn off when run the Image denoising(when select 3). In CUDA2.0, the PC will turn off when run the N-body and Image denoising(when select 3).

4

It’s almost certainly a bad PSU.

5

thank you,

maybe, the supplier just came and took the computer back to fix.

I wondered what will happen if the PSU is not powerful?

  1. the GPU will not got enough power, the GPU indicator light will become red;

  2. the GPU will not work normally;

  3. the windows will detect the status and report it to user.

why the PC just turn off suddenly without any error messages?

6

Could be a bad PSU with a serious voltage ripple, or it could be that there’s a short in the system build. Either one is really annoying to debug.

7

my computer specification

Intel Core 2 Quad Q6600 System (2.40GHz / 1066FSB / 2x4mb)
Asus P5E64 WS Evolution
Silverstone Kublai KL01
GOODRAM 2 GB 1066MHz Memory (2 x 1GB)
EVGA 9500GT 512Mb DDR2 for display
EVGA GTX280 602/2214MHz PCIe/1GB/D-DVI Video Card for CUDA
Samsung F1 640Gb Hard-Drive (16mb)
Silverstone 1000W Decathlon Power Supply

I got similar problem with my computer. After 5 minutes of running “nbody.exe --device=1” winxp ask
me to reboot, with message “nv4_disp error”.
This problem causes GNOME and KDE4 to fail as well.
For me it looks like driver problem, because changing driver from 177.73 to 177.80 made my system works much better.
of course it may be something really bad with Silverstone 1000W Decathlon Power Supply.

8

Yep, smells like a power issue.
It could be a bad PSU, since the 1000W rating and brand name show its design should handle it.

But if it’s not a bad PSU, it could easily be BAD CABLING. Especially if you have the GPU hooked up to only one rail of the PSU. Did you play games with adapters and such to get both the 6 pin and the 8 pin power? Do those plugs come from different rails of the PSU to avoid overloading just one? Some Silverstones have just a single beefy rail, which would be fine, but not if you used splitters or whatever to get cables working.

The ideal cabling situation uses just two unique cables, one 6 pin, one 8 pin, both running straight from the GPU to the PSU without anything else involved.

9

Thanks for all above.

I don’t check the cabling, and the supplier will send back tomorrow. I wish they will fix the problem. Just like nancy^ says, I think there are some problems of the CUDA driver.

10

I’m confused how one could deduce that there was a CUDA driver problem if GNOME and KDE4 are failing.

11

sorry, it is the problem of PSU. BAD PSU.

12

First of all I would like to apologize CUDA driver development team for bleaming them about driver problem. Of cource there is no problem with driver at all. As well as there is no problem with Silverstone PSU. After tests the PSU was o.k. Silverstone exchange this good PSU to more powerfull unit ZEUS 1200W, when I asked them to do this. As you probably can imagine the problem still egsist.

 The tool I use for diagnostic of temperature was reporting 72 degrees C all the time (or lower temperatures). When I took EVGA GTX280 to my home computer and used EVGA Precision suite for diagnostic the GPU, I find out that real temperature in the time of system halt is more then 104 degrees C.