A Question for George

[Hades_au]

Overclockers.com are saying that it's the 'older versions' of Prime95 that "may" be the cause of errors on their overclocked machines, rather than the machines themselves. The article is here http://www.overclockers.com/tips00238/.

Care to comment?

[cheesehead]

Your question is for George, but here's my comment anyway: Author Ed Stroligo seems to be referring to the suspicions of some overclockers without actually agreeing with those suspicions himself.

Originally Posted by Ed Stroligo

The problem might not be your machine, but an older version of Prime 95.

If you go to Prime 95's revision history,you'll see that a number of changes were made in versions 22.2-22.6 to correct bugs that created erroneous errors. In particular, versions 22.2 and 22.3 have revisions made because "version 21 was too aggressive in choosing FFT sizes for the P4."

This may/might be the reason why at least some machines are failing Prime 95 at speeds a good deal lower than other stability tests.

But his ultimate message is: get the latest version of Prime95 and test with that. I'd agree.

[Hades_au]

Originally Posted by cheesehead

But his ultimate message is: get the latest version of Prime95 and test with that. I'd agree.

It's nearly always best to have the latest version, however, with regards to Prime95, my understanding is that the revisions have little to do with the torture test. The torture test is simply comparing known results. I don't see how the different revision numbers change the torture test.

These people have simply jumped on George's 'error messages' and are using them to diminish the value of the program. See the thread at Overclockers.com.au http://forums.overclockers.com.au/sh...hreadid=126826 although you may need to sign up to view the site.

[xtreme2k]

hiya hades_au again

hahaa

I am so bloody tired of explanining over and over again.

[cheesehead]

Originally Posted by Hades_au

It's nearly always best to have the latest version, however, with regards to Prime95, my understanding is that the revisions have little to do with the torture test. The torture test is simply comparing known results.

Yes, and I think Ed Stroligo knows that, so he's just trying to shortcut by persuading the complaining OCers to go to the latest version.

Originally Posted by Hades_au

I don't see how the different revision numbers change the torture test.

_We_ know that these version changes aren't relevant. But the complaining OCers probably don't know enough details about Prime95's workings to realize there's no connection and are just jumping on that to explain away Prime95's superior instability-discovering properties.

So Stroligo sees that it's easier to simply persuade them to go to the latest version to eliminate that possibility of blame than to explain the details of how Prime95 works. When the complaining OCers see that the latest version finds the same problems in their systems as the earlier version, some of them will figure out that the revisions were irrelevant to the torture test.

[Prime95]

Originally Posted by Hades_au

Overclockers.com are saying that it's the 'older versions' of Prime95 that "may" be the cause of errors on their overclocked machines, rather than the machines themselves. The article is here http://www.overclockers.com/tips00238/.

Care to comment?

I emailed the fellow that the items he references in the whatsnew.txt file did not affect the torture test.

[pakaran]

Originally Posted by Prime95

I emailed the fellow that the items he references in the whatsnew.txt file did not affect the torture test.

Wouldn't they affect it in the other direction if anything?

By having more efficient use of prefetch, the CPU is in use rather than hanging on a load instruction, and thus might heat up, I'd guess? Forgive me hardware isn't my field...

Nathan

[Prime95]

Originally Posted by pakaran

By having more efficient use of prefetch, the CPU is in use rather than hanging on a load instruction, and thus might heat up, I'd guess?

You are correct, but that improvement happened in version 21. The only thing in version 22 that affected the torture test was using more memory.

[pakaran]

Thanks for the info George.

I've taken some classes in hardware, which is why I was curious.