New version OCCT stability test

[virtualrain]

We do share some code, yes. i'm using the same libraries (gwnum) and some code for the CPU test was taken out of Prime95 ( which reminds me i completly forgot to mention that in the thanks.txt file). That's why you see a huge *bump* in temperature in the cpu test from 1.1.1b and 2.0.0.

However, the calculations are radically different, using different home-made algorithm which uses a part of the routines of Prime95 (they have proved very effective).

You could call that a "mix" between proven routines and home-made code

Can you explain how you feel that OCCT is better at stress testing than PRIME95? I'm just trying to understand the differences and how they translate into what people perceive as finding instability sooner. I'm somewhat skeptical of this claim.

[Tetedeiench]

Can you explain how you feel that OCCT is better at stress testing than PRIME95? I'm just trying to understand the differences and how they translate into what people perceive as finding instability sooner. I'm somewhat skeptical of this claim.

Really tough question, and pertinent as well

OCCT uses some, not all, routines of P95. That does not mean OCCT and P95 are the same tools. Think of those routines as an engine : you won't say that every car with the same engine is the same, right ? Some will still be faster, more comfortable, etc.

Well, to begin with, OCCT has 2 engines, Prime95 has only one. RAM and CPU tests are drastically different in OCCT, P95's tests are alike, except datum. OCCT and P95 do not share any data either : we're not working on the same number sets, and the routines working with those sets are a bit different in OCCT CPU, and really different in OCCT RAM.

OCCT comes with alot of features that P95 doesn't have (graphs, monitoring,...). But that's not the topic, just mentioning it.

Problem is i can't answer you with a mathematical proof showing you OCCT is faster.

The errors generated in a CPU are random. You don't have any rule governing that, or rather the rules are so complicated, random and impossible to design at the moment. So we can't predict when or what will make a computation error happen. What P95 and OCCT do is make those errors likely to happen, that's about it.

And how do we do that ? Well, by the old empirical method : try, try, and try, and sometimes, you'll stumble on something good. For example, a number you're working on that has a higher probability to raise errors, for some reason. Do not ask me why ! It's just a fact.

And thus i'll be using this number.

And so on.

OCCT has been around for... em, i don't even remember. I'd say 5 years now. In those 5 years, i've worked on gathering those data, taking beta testers results and "feelings", and improved OCCT.

So yes, i can't prove you that OCCT is better. There will always be cases where a failure will be detected by one program and not by the other. Prime95 is still a great tool, and i don't deny its value here. I'm just saying that, in general, OCCT tends to be faster at making errors happen. And i can't prove it. It's just reports of people who say that OCCT is faster, and there have been alot of them.

OCCT and Prime95, even if they share some code, are completly different tools, and i always advise people to use several tools, as to get a better idea at how stable their computer is.

I won't say my tool is the best, i'd just say that right now OCCT seems to be a bit faster according to user experience, and comes with more features.

The more tools the merrier !

My goal is not to steal anything from prime95... i'm just working on another tool for people to get a better idea of how stable their overclock is.

PS : dunno if i said it here, but i'm french, so if you see some spelling mistakes and such, just ignore them, english is not my native language