Use memtest x86 overnight to see if it's bad RAM. Then use HCI Memtest again over the next night to check the RAM from within Windows - note that you need to read the instructions and run enough instances to test all your available RAM. I've had an instance where memtest didn't detect bad ram but HCI did, after RAM replacement everything was good again. Symptoms were nothing like this though.
Have been running Memtest since midday. 3 instances each testing 2gig of ram. Each test has over 600% coverage with no errors so probably not the ram. This has also pushed the cpu usage right up and the temps went up to around 50C. So it is also probably not temps.
Leaves gpu and psu nether of which I have a spare of. I guess unless it becomes a real problem i will have to live with the odd random crash (which doesn't sit well).