Do you trust real world tests more than synthetic ones?

[dantestyle]

Hey guys,

A few weeks ago, I put up a thread asking about overclocking my 4770k. Lots of stress tests perform well on it when overclocked, but the small fft test in Prime95 always caused it to jump into the mid-80s.

So I've been running my computer at 4.2ghz with 1.18V. And I've compiled video using both handbrake and Final Cut X (which is my primary use for this computer in the first place). I've played Bioshock infinite at max graphics for hours at a time and it's all stable. Also, the temperatures go to about 65 maximum. But when I run Prime95 small fft, it very rapidly makes it up to 85. Should I care about this Prime 95 test, if all the things I do with my computer on a daily basis don't push the temperature very far anyway?

Just for reference, I'm using an h80i cooler, and a Z87X-UD5H mother board.

Is there any great need to do a stress test for days at a time that I'm not aware of? If it gets me through my full days consistently sollid, do I need to worry otherwise?

 

[sahilmalhotra17]

It depends on what your real world application is.
The applications that you describe load your system very differently than p95. P95 now uses avx instructions which cause the processor to heat up faster and higher. Some simulation engines for high end aerospace robotics etc also use avx to optimize performance and crunch more numbers. If you are running any apps that run avx in your real world application you need to watch for your p95 nos. otherwise you would be fine with something like realbench, aida64 etc.

 

[seanmcd72]

If you are going to overclock, most people live by the rule of - it's not stable unless it can pass a stress test. Prime95 is one of the ultimate stress tests. Yes, it's going to drive your temperature up. 85 although high, is a safe temp. Haswell's run hotter than ever. You can run that for days at 85 and it should be just fine.


I do not overclock any longer. I don't see the point anymore. But, if I was overclocking (I did it for many many years) - I would loop prime95 AND 3Dmark (both at the same time) overnight for my stress test. If it could do that without crashing or stopping, it was good to go. Any sign of trouble, then the overclock is unstable and not worth the risk.

 

[justr]

Looping stress tests overnight only tests for-- drum roll-- looping stress tests overnight.

The overclocker's creed is that there is no such thing as perfectly stable. It literally does not exist. So you test for what you personally consider stable, or what the guidelines are for entering your build's results in a list of top-performers, or your real world use.

This can vary dramatically. For one person, 30 minutes of intel burn test, or 2 hrs of Prime is enough, followed by pragmatic stability when doing the sorts of day-to-day computing you'll use your computer for.

I think the idea of running tests for 24hrs has passed these days. The reason: 24hrs is just as arbitrary as 48 hrs, 72 hrs, or 20 minutes. If you're going to be running BF4 or FCPX/Compressor, why stress the hell out of your build for 48 hrs with a task your computer will never be doing? Better to get it stable-enough for you, then "test" it out in the real world by carrying out your work in ever more high-consequence situations (don't go from getting synthetic stable-enough to working on a big project that's do or die for you, for instance; better to work on a play project to see how it goes first).

My personal tests include IBT and Prime, running enough loops to find where the temperatures stabilize, and then a few more loops. This tends to keep the tests themselves short, and gives me a huge window for thermal safety. Nothing I do in the real world-- FCPX, Compressor, Aperture, Games-- gets close to what Prime or IBT does for heat. In my experience, I get stable enough from this route... then I go into doing the things I do and rarely, perhaps never, have had stability issues. So that;s my measure of stable.

The one test I've run overnight is a memtest when I first get new memory-- at stock clocks-- to make sure every DIMM is healthy. Boot directly into the test (not OS), and run.