Issues with your account? Bug us in the Discord!
Very interesting article about certain "feature" of P4/NetBurst
E.T
Quote-o-matic
in Zocalo v2.0
[url]http://www.xbitlabs.com/articles/cpu/display/netburst-2.html[/url]
If you're lazy interesting part starts at page 5.
Looks like Micro$oft doesn't have monopoly with "do nothing-loops". :D
If you're lazy interesting part starts at page 5.
Looks like Micro$oft doesn't have monopoly with "do nothing-loops". :D
Comments
Page 14 and 15 get to the meat of it, but you need to read pages 5-13 in order to get enough background to understand 14 and 15.
[B]The Prescott architecture is funky and does funky things to try and execute code as fast as possible. [/B][/QUOTE]Basic thing is this: commands are run even when required data isn't available, but result is just discarded and command is run again after delay loop. (which "runs parallel" with main pipeline)
Which can lead it to run same commands again and again if required data is far from processor, (&if required data is in RAM whole string of commands can make even hundred loops) and they can even go to wrong order which causes running every command at least twice even when data required by first command is available if commands depend on results of previous commands.
And after first command in original code has required data and gets out of delay loop that slot is filled immediately with new command which can mess order even more because again this command can be in loop much before originally previous command required for getting right data.
So it just doesn't only run those commands in delay loop many times but it automatically messes order of new commands when they're jammed to first available slot in string of commands running that delay loop.
PS. I didn't notice any talking about bits or other things like that, so it can't be that hard to read.
[B]so more loops is better?[/B][/QUOTE]Well, how do you think low clock frequency Pentium M (P3 based) is able to beat P4 "ticking" much faster?