That certainly sounds like a PSU issue, or at least, it sounds exactly like an issue I had back in the day with a 450w rosewill that would do the same thing under certain graphically intense situations in HL2. Went away when I upgraded the PSU.
You can get a lemon from any manufacturer even if its quality. If it's new you should rma it or replace it in the store asap. Though the only sure way to test if the faulty component is the PSU is to use another one and see if the problem persists.
Also, just in case you were thinking about buying a PSU tester I would save my money if I were you since from experience, PSU testers do not put any load on the PSU so you can't really be sure of anything unless the PSU is absolute crap and doesn't provide the appropriate voltage.
As far as lemons go, I have a Netgear N900 (WNDR4500) and given the rave reviews you would think it would be a quality router but oh man, was I unlucky! The damn thing would restart and crap out on the internet every few days and the lag in games was horrendous. Even after days of adjusting every setting on the router and re-reading the entire manual 2 to 3 times and playing with every settings on the QoS the thing would just suck. I now have the most expensive Access Point I've ever bought. One thing it was good at was with signal range.
Everything I write is Sarcasm.
Also check your CPU and GPU temperatures. Use an app (program) that either logs them or displays them in a 'real-time' chart so you can see how temps vary with usage.
Make sure your cooling fans are spinning at appropriate speeds (e.g. typically as CPU gets hotter, the fan should spin faster, etc). If they aren't already doing so, try forcing all the cooling fans to run at 100% RPMs for maximum cooling (e.g. direct connect to 12VDC or use fan control software to force 100% RPMs). See if the increased cooling changes the symptoms. Programs like OpenHardwareMonitor can help in both fan/temp monitoring and fan RPM control of system, CPU, and GPU fans. http://openhardwaremonitor.org/ http://www.majorgeeks.com/files/details/open_hardware_monitor.html
On the other hand, I've had 4 ~700W class OCZ 'Xtreme'-series PSUs exhibit power stability issues, including two that were intermittently 'glitching' and dropping +3.3V rail to ~1.2V and +12V rails to ~10V for milli-seconds at a time, which are definitely out of spec and could cause symptoms such as yours. These highly intermittent (and possibly load dependent) 'glitches' are NOT detected by the 'go/no-go' simple PSU testers one sees for about $15-$30. They report the PSU was 'OK'.
Also, depending on motherboard power regulator design and subsystem using the specific voltage rail, this may or may not show any symptoms during operation. Eventually 2 of the PSUs died completely, including 1 OCZ700SXS 'StealthXtreme' model that had been RMA'd and factory modified for a previous known design/manufacturing defect.
...Also, just in case you were thinking about buying a PSU tester I would save my money if I were you since from experience, PSU testers do not put any load on the PSU so you can't really be sure of anything unless the PSU is absolute crap and doesn't provide the appropriate voltage....
I completely agree the go/no-go PSU testers don't diagnose a number of PSU operational failures.
However, they CAN save diagnostic time and prevent major damage to hardware when all it takes is 30 seconds to check the PSU for gross failures (like out-of-spec DC levels, power-good signal timing, and so on). Using a dirt-cheap tester can save a lot of time in those situations.
Being careful and doing redundant testing HAS saved me money and time in time-critical situations.
Many here at HWA have probably figured out I'm very aggressive at minimizing exposing costly equipment to potential sources of damage; real-time and latent. Hence...
I also always test a PSU (and all its connectors) with one of these cheap testers before connecting it to a motherboard..even if it is a PSU I 'know' was good last time I used it. It's not a guarantee, but doing this has pre-screened-out many bad PSUs.
I also then connect to a cheap 'known-good' 'throw-away' motherboard with display and various 'throw-away' peripherals, and boot it up to do a second test to reduce the chance of destroying a very expensive motherboard. Only after these two tests (relatively cheap if something goes wrong) do I actually put a previously known//unknown/new PSU into a system that is important to me or others.
note: a 'throw-away' device is simply one that I don't care if it's damaged/destroyed. It's likely one that I've salvaged from somewhere, and costs me next to nothing if it's destroyed.