Is there a known bug with pent 4s with audio plugs?

david-ruby wrote on 5/15/2003, 8:10 PM
David Abraham Fenton ( if your here) brought up an interesting thing.
My Prob is as follows:
I had an interesting find. I had a huge video project and loaded it onto a new pent 4 2.5 cpu system and as I got half way throught the project on playback the cpu reached 100% and started clipping the audio. I was shocked!!
I then loadd same project into a dual pent3 1 gig cpus system and at same spot showed 50% cpu usage!!!
What the heck could this be?
David Fenton mentioned to me about a pent 4 denormal bug. What is this?
Is there a fix?
Is there life on mars? ; )


Baylo wrote on 5/15/2003, 10:48 PM
The Pentium IV Denormal problem relates to how the CPU processes very, very small numbers.

As I understand it, when the CPU has to perform operations on small numbers, it switches to a 'high precision' type of mode. As you might expect, this 'high precision' takes more processing power. Now, when you think about it, very quiet audio is really just a succession of very small numbers, and when you apply affects, etc to audio what you are really doing is performing calculations on these small numbers.

You can see where I'm heading with this. The problem is that when Intel designed the PIV, they designed it so that the 'high precision' mode kicks in at a relatively much higher level than was the case with the PIII. I say 'relatively', because in reality these numbers are still so small as to be inaudible. This means that, for software that does not take this into account, the PIV CPU actually has to work harder when there is a quieter signal (or silence) than when there is an audible signal.

Whether this is a bug or not is semantics - it was by design on Intel's part, but seems to have had undesirable and possibly unintended consequences. In any event, it was the chip that changed, not the software. That said, the only solution is to change the software.

The best fix is to get the software developer (it's typically plugins that exhibit this problem - especially free ones) to release a patch that changes the treatment of these small, inaudible numbers. There are also plugins around on the web that deliberately insert some noise into the signal path to boost the signal out of the problem range. I hear conflicting reports as to how successful they are in solving the problem.

As to life on Mars, I'm still working on that one...

Hope that helps,

david-ruby wrote on 5/16/2003, 9:15 AM
LOL. Thanx for the mars add.
I got it totally now. I was thinking there was a similiar prob with a few other software packages I had worked with before that had to be fixed.Hmmm..
Might be an ultrafunk prob. I will have to research more.
Let ya know of any thing strange.
Thank you
david-ruby wrote on 5/16/2003, 9:16 AM
Wonder if this is also related to win2000?
Still lookin.
Baylo wrote on 5/16/2003, 9:34 AM
As far as I know it has nothing to do with the OS you are using. It's all about the CPU.

BillyBoy wrote on 5/16/2003, 6:15 PM
Seems Intel is ALWAYS admitting to floating point math problems...why I switched AMD.