At least his use of RealBench is a lot better than the synthetics like Prime95/OCCT/IBT.
http://www.overclock.net/t/1510388/haswell-e-overclock-leaderboard-owners-club/2390#post_22900116
I used to believe the old timers and veterans when I first joined this forum but my view is now 180*. A stable system for me is the maximum operating speed it retains in all real world applications
I run. It makes no difference to me if some
synthetic power virus loads my CPU/GPU to 100.00% because I will never run any program that will use my components like that.
Also, the newer versions of these programs place a completely unrealistic load on CPU/GPU. FurMark = useless junk.
What's more important to you, having your system running 200-300mhz slower but stable in synthetic tests that have no association with real world programs? Or, 100% rock solid stable in every game, every distributed computing, rendering, encoding, etc. application you use?
I am not suggesting that RealBench is the only valid way to stress a CPU but I'll prefer real world apps over synthetic apps for stability testing.