I\'ve just finished a P4 (2.4ghz 533) test on a friends un-optimized Windows 2000 PC.
2 instances of Kontakt in Logic 5.3 ....WAIT FOR IT!! 320 stereo voices [img]images/icons/shocked.gif[/img] on a creative Audigy card.
This is more like it!!!
As i said before all without Filters and FX, just ADSR envelopes.
I set up a realistic Multi in each instance of Kontakt with voice limiting set to 24 on 10 channels,...they were all hitting 24 and going over slightly. More were added to the 2nd instance to make up a total of 320 stereo voices, all playing without a glitch!!
You get poorer performance when setting up a single channel to use more voices e.g I had 3 intruments set at the default 32 voices and the cpu meter peaked very quickly. 3x32=96 voices.
To get best performance limit each channels polyphony to what you actually need.
I just ordered a P4 2.4 GHz with the ASUS P4PE motherboard w/raid. It was like $10 more than the one without the raid. Ordered 1 GB of DDR 333 memory. Only ordered one 80 GB WD 8 MB buffer drive. I will add another drive as soon as I can scrape up more money.
What sound cards are you running on your machines? I may have to pick up another card for my older Giga machine.
I tried your trick of limiting the voice count, but the result was the same on the PC. I didn\'t try it on my mac. I am still getting better results on my mac than the PC, so I am a little hesitant to go build another one. I wonder why you would get peaks with a 32 note count per patch. I would think that if you were using a piano with the damper down that 32 wouldn\'t be enough.
I was able to get 93 voices (non DFD) from my g4 800 in standalone mode (as compared to 80 with DP running).
320 stereo voices is nice and all, but really, the only thing that will matter for pretty much anything but drums, is how many voices you can get while streaming - and preferably with as small a buffer setting as possible. I am not sure but I think GS uses a 32k buffer. Imagine, changing that preload buffer to 64k, would instantly make you load up only half as many samples as you can with 32k.