Originally Posted by ThrashZone
Is there at least one of your replies addressing the op directly instead of all this back and forth about p95 is great crap ?
Okay I give up you are the great blameless give it a rest
Yes, my first post in this thread addressed the OP's temperature question. In worst case loads, a 1GHz OC on a 5820K (3.3GHz base to 4.3GHz, or thereabouts) is probably too hot for an NH-D15. Of course, this being a forum thread anyone can read, any of these replies may be of relevance to anyone viewing them.
My issue was with your presumption that the OP wouldn't be doing anything that would justify more than Blender as a CPU stress test. This could be true, but we do not know. That it's an HEDT platform certainly doesn't imply, to me, that it's going to be used for less intensive work than a mainstream part.
"p95 is great" is a misinterpretation of my statements. Prime95 is one of many tools that have a place in identifying instabilities and identifying instabilities is ultimately the purpose of stress testing. A system passing Prime95 may not be stable, but one that fails Prime95 (in a manner not due to a bug in Prime95...which is easy to check by seeing if the same test passes on any CPU of the same type) certainly has objectively identifiable instabilities.
I'd also be remiss if I didn't point out that using the term "power virus" when referring to stress tests is a misnomer. Such a use that was popularized back in late 2009 when the insufficient cooling on GTX 480s, as well as the overly hot VRM on the Radeon HD 5800/5900 series parts, were experiencing excessive failure rates. FurMark and similar programs became (quite unfairly, IMO) scapegoats for what were essentially the first generation of parts that could not be run to maximum theoretical loads without damaging stock-clocked reference models. Later parts addressed this with various limiters and throttling/boost mechanisms, to maximize performance in games, without having to actually engineer boards and coolers for such peak loads. Only relatively recently have CPUs started to go down this same path. Anyway, 'virus' implies something covert, illicit, or out of user control...stress tests are not these.
Ultimately, any harm that could potentially be caused by demanding stress tests is generally less than that which may be caused by too cavalier an attitude toward stability. Crashes and throttling will generally be ample forewarning of potentially dangerous power or temperature levels, but instability can do damage entirely silently, and if there is any serious labor involved (for business or pleasure), data is quite likely to be worth more than the hardware itself.
Originally Posted by doyll
Bottom line here is best load test is one user will be using in normal use.
I strongly disagree.
You could easily have a system that has an unacceptably high chance of failing or corrupting a batch encode (say one in ten, for the sake of argument) that still has a very good chance of passing an equally long encode (90%, in this case).
If that system is instead running a few hundred MHz slower and several degrees cooler at the same voltages because one took into account what was required to be stable in more demanding tests, the odds of failing that encode decrease geometrically.
I do a fair bit of video editing and transcoding myself. The time saved by encoding 5% faster is not worth the risk of failing a long encode, or spitting out unusable files, because I used my encoder as the primary stress test for my encoding.