Kevin Cox

I'm using vibraphone~ (one of PeRColate's physical models). Actually, a whole bunch of them (I've tested 20-80), inside a poly~.

I've found that if I leave the patch sitting with no input for a few minutes, it will double (or more) in CPU usage, and the audio won't work for awhile. Then if I send it a bunch of events, it will come back after making some crazy sounds, and CPU usage goes back down to normal.

I'm testing this on a macbook pro, latest versions of max/msp and percolate. I've tried changing vector size, no luck.

Does this happen to you? Any ideas? I'm wondering about the less-than-documented noteoff message inside the vibraphone~ help patch. Does anybody have experience with PeRColate's physical models?

Any help would be greatly appreciated - it's for a rapidly approaching installation. If I have to drop the vibraphone~ and write my own synth I want to start sooner than later! (hey - anybody have a good vibe patch?)

(this will require that you have PeRColate installed, get it at 

http://www.music.columbia.edu/PeRColate/)

(this will require that you have PeRColate installed, get it at http://www.music.columbia.edu/PeRColate/)
Main patch:

problems-with-vibraphone-from-percolate

On Apr 25, 2007, at 3:27 PM, Kevin Cox wrote:

> I've found that if I leave the patch sitting with no input for a 

> few minutes, it will double (or more) in CPU usage, and the audio 

> won't work for awhile. Then if I send it a bunch of events, it will 

> come back after making some crazy sounds, and CPU usage goes back 

Sounds like the developers need to account for denormalized floating 

point numbers on x86. As a hack you could possibly just keep some 

amount of activity happening so that it won't trail down to these 

super small numbers which have adverse performance characteristics on 

Windows FAQ entry, but also applies to MacTel:

https://cycling74.com/twiki/bin/view/FAQs/

WindowsMaxMSP#What_is_the_denormal_problem_and

https://cycling74.com/cgi-bin/searchsite.pl?SearchText=denormal

http://en.wikipedia.org/wiki/Denormal_number

>
> I've found that if I leave the patch sitting with no input for a  
> few minutes, it will double (or more) in CPU usage, and the audio  
> won't work for awhile. Then if I send it a bunch of events, it will  
> come back after making some crazy sounds, and CPU usage goes back  
> down to normal.

Sounds like the developers need to account for denormalized floating  
point numbers on x86. As a hack you could possibly just  keep some  
amount of activity happening so that it won't trail down to these  
super small numbers which have adverse performance characteristics on  
x86.

Windows FAQ entry, but also applies to MacTel:
https://cycling74.com/twiki/bin/view/FAQs/ 
WindowsMaxMSP#What_is_the_denormal_problem_and

Forum search:
https://cycling74.com/cgi-bin/searchsite.pl?SearchText=denormal

More info online:
http://en.wikipedia.org/wiki/Denormal_number
http://ardour.org/node/139

Wow - I remember these being an issue years ago when the pentium 4 came out - people called it the "pentium 4 bug." I assumed it had been ameliorated since.

I just tried it on Windows, and the same problem happens, so it must be the vibraphone~ external's fault. PeRColate comes with source... guess I'm going to try to fix it. Uh-oh.

I tried periodically sending events, but I don't think that's going to work. Lots of crazy sounds happen.

I suppose my last resort would be to go through and make samples of every sound vibraphone~ might be making in real-time, and build a quick sample player. ugh. These sounds are just perfect for what I'm doing!

Ah, good ol' PowerPC. No problems there. Guess I'll be running a separate computer as an audio server... oy vey

problems with vibraphone~ from percolate