a few (and incomplete) comments on memory configuration:
running tests for *raw voices* and using the demos zara & wondering why, we found memory latency, memory bus speed and number of memory channels to be the more important the more one puts stress on the CPU.
this is a non-linear relation between number of CPU cores, CPU speed, number of voices, number of instruments, (audio-)latency and number of output channels.
example #1 zara (22 instruments), stereo, latency 128 samples, 8 core 5550 @ 2.93 GHz, HT enabled
1066 = 39% CPU (crackles), 1333 = 34% CPU (no crackles)
example #2 zara (22 instruments), stereo, latency 256 samples, HT enabled
i7 950 (4 cores) @3.2 GHz, dual channel memory 56% CPU, X5550 (8 cores) @2.93 GHz, triple channel memory 25% CPU
in a nutshell: the closer it comes to the CPU limits the more important is memory performance.
we are still collecting data for a clear performance table and all figures you have for other configurations under controlled conditions are welcome.
christian