determine cause of auto shutdown when playing games

Question

The problem

When playing a game, the computer suddenly shuts down, although not completely because some fans (maybe the GPU's) are still running and the light on the power button is still lit. I have to press it to shut down completely. Then again to start up, which then sometimes fails (no display or freezing) the first time.

I'd rather not unnecessarily replace either the graphics card or the PSU. What is most likely the issue, and is there a way to find out for sure without having a spare graphics card or PSU lying around?

Specs

• MSI X99S motherboard
• XFX Radeon R9 280X Double Dissipation graphics card - updated to latest drivers
• 650W Antec PSU
• Not overclocking anything as far as I know

Things I've noticed

• This never happens when the GPU is not under load. I can leave my PC on for days. So I was thinking either GPU temperature or power supply issues.
• When my computer has been off for a while, it takes longer before it shuts down. Once it starts happening, it will shut down immediately when starting to play a game using 3D graphics. This made me think it was GPU temperature related.
• I checked the temperature of CPU and GPU using HWMonitor. Last time it happened, the peak temperature had been 83°C but I checked just before the shutdown and at that point it was already back to 70°C (probably because I had switched windows so that the game had run minimized for some seconds). CPU temperatures were low (under 50°C).
• At one time, during the first try to power up after the shutdown, the computer appeared to be booting correctly but without display output. I pressed the power button to turn it off, then on again, and I had display output working as normal.
• PC is in a fairly dusty place, which may have a negative effect on some of the parts.

Answer 1

Sounds like you've pretty much figured it out - you have a classic case of "over-temperature or under-power".

As you suspect it might be dusty in there, that's your first point of attack - strip & clean the entire machine, paying special attention to the GPU; which may require, depending on design, that you remove the heat sink to get right inside. This will necessitate new thermal paste.

That's the cheap first play.
All it takes is your time plus 5 $£€ for some new paste.

If that sounds too daunting, pay a professional to do it, rather than get it wrong & break it.

If it still fails, second suspect is the PSU. They get 'weaker' over time, but if it's under 3 years old, that wouldn't be my first suspect. Measuring a PSU's output whilst under load is not a simple task. If you get that far without professional help, it's probably cheaper to just replace it than pay to have it diagnosed.