I am soon to be recording a symphony orchestra that will probably end up being mixed with some VSL samples. I haven't yet determined the sample rate I will record the orchestra at so thought I'd do some tests to see the CPU hit incurred when playing back VSL samples at their non-native sample rate. The test wasn't scientific but I did strive to obtain meaningfull results.
Set-up:- MacPro 3.2 8 core with 32GB of RAM, OSX 10.5.4, Logic 8.0.2, Vienna Instruments 3208, System Drive 2 disc SATA raid 0, Sample Drive 4 disc SATA raid 0. Logic's buffer was set to 1024 mainly to replicate the best available performance during a mixing session where latency is not an issue.
48 instances of Vienna Instruments were loaded into a Logic Session. A 6 note chord sequence with overlaps was inserted and set up by transpositions and different velocities to ensure no sample was "double played" by different tracks.
Results below:
At 44.1KHz playback was smooth without any artifacts this equates to a remarkable 576 simultaneous stereo voices (1152 mono)
At 48kHz playback was reduced to 45 tracks, 540 simultaneous stereo voices.
for higher sample rates I used 24 tracks as my basis
At 88.2KHz 24 Tracks played back smoothly with an average cpu loadacross the cores of 50%
At 96KHz 24 Tracks played back smoothly with an average cpu loadacross the cores of 60%
At 176KHz between 20-22 tracks could be sucessfully played back
At 192Khz between 18-20 tracks could be sucessfully played back
To mute tracks I used the bypass button on the Vienna Instrument plug-in (just muting the tracks for some reason does not drop the cpu load even if the tracks are not playing)
I was pleasantly surpised at how little overhead is required to sample rate convert (about 20% for 48k and 50% for 96k)
Higher rates were more demanding but when you consider 240 voices were still available at 192k it's pretty good.
Mind you the overriding impression is what an amazing machine a single MacPro 8 core is!
Julian