Hello everyone,
 
 Problem: a new, out-of-box GTX 580 SC DS was tested in EVGA OC utility, and then in S.T.A.L.K.E.R. CS. The first one was no issue (running time ~10 minutes).
After the second one I found out that the temps were exceeding 100°C !!!
My specs below, but here I'll just say that my Cosmos has enough space and fans to keep normal temps.
 
After this observation, the OC stress test holds for 30 seconds. Once the temp reaches 70°, the card crashes (system keeps on running, but the card goes int "safe mode": fans 100%, no video signal), in some cases it halts the whole system.
 
Tried the following:
- Since my PSU (see below) has only 6-pin PCI-E connectors, had to use the supplied in box adapter (two 6-pin-> one 8-pin). Tried switching rails. NO EFFECT
- Installed beta NVIDIA driver, rolled back. NO EFFECT
- Unmounted card, checked contacts, mounted back. NO EFFECT
 
Have spoken with EVGA support, but no recommendation except "test in shop and RMA".
 
I believe, many of you, enthusiasts, have seen similar behavior or hit such problems.
 
Config:
ASUS P5Q-Deluxe, E8600 (stock), Dominator 8500C 2X2GB @1066 Mhz, PSU ThermalTake 850 AP, Vista 64 bit, WHQL NVIDIA 285.62, 3X SATA VelicoRaptor's. Case: Cosmos with 5 Noctua Fans (including CPU fan)
 
Thanks in advance!

The problem has been solved!
 
The cause appeared to be very strange.
It's ZALMAN ZM-MFC2 block, which is a controller of fans and has an additional option of monitoring the power, consumed by the system.
So, if the cable of the power meter plugged in - by reaching 380 W consumption, the card locks itself...
 
mystery