Information générale
Catégorie:
Installation et configuration
Versions des environnements
>>Just be careful on double core cpu's: the queryperf counters can get out of synch there, logging miminmal time machine behaviour from your program if you don't apply necessary pathces...
>
>Great... luckily, I always try to keep my machine somewhat suboptimal (that's how "slow" is said in Microsoft newspeak), so my laptop is dual core, but the development box is not. And I'll probably keep it that way.
>
>And what do the patches do, force the counters to execute in the same neck of the processing woods as the caller?
AFAIR it guarantees synchronous micro-timestamps - skimmed it on the AMD site, as our first double cores were from them. I had seen a hightened variance in repeated measurements (not high enough to mask the effects I was after, but visible when comparing to single core) before some gamers were seriously PO to get to the bottom of timing problems. From there on I had argued with myself that for high definition measurement not only the process measured runs on one core only, but other "dangerous" tasks are put on another core (AV, Firewall and so on). Did not change that approach after reading about the synch problems, as it isolates the intersting process even better than the patch from error variance. One of the arguments I won...
regards
thomas
Précédent
Répondre
Voir le fil de ce thread
Voir le fil de ce thread à partir de ce message seulement
Voir tous les messages de ce thread
Voir tous les messages de ce thread à partir de ce message seulement