I am shure that the 512 voice machine is based on a dual P2 machine powered by win NT. GS standard performs 64 voices even on a old P266/60MHz FSB and even on such an \"old\" machine performance headroom is left. With fast drives so 128 or even 256 voices should be available on a fast P450 system, which has a lot of performance headroom. If I increase the sample cache size (which means of course more RAM consumption) I am shure you could reach 512 voices on a PIII 550 with 512 Mb RAM today. It it´s just a question of the will to do it.
According to informed sources, the 512 Voice unit shown at the Winter NAMM was based on a design using four single board computers (SBC) plugged into an Active Backplane mother board. Each SBC runs a 128 note version of GS in Win98. The unit is somewhat expensive because SBCs are more costly to make and have a much smaller market than standard PCs.
Holger poses an interesting question about the upper limits of polyphony. It may be that a dual or quad CPU design may raise the limit. Likewise, BeOS (the fast new AV oriented operating system) may be a big help.
Multiprocessors are not supported by Win98 and at present, GS does not run under NT or Be (which due support Symetric Multiprocessing). NemeSys does plan to support Win2000 (ne. NT 5.0) and BeOS in the future. It may be that bandwidth restrictions in the current PCI architecture place an upper limit of 128 voices.
jphardy is right on target. The current high end system is a passive backplane system which is fairly expensive. It is 4 units and rather deep. You can create a similar setup with 4 separate computers networked together and with a switchbox but it would take quite a bit more space. We are taking orders for the system (NCS-44\\ NemeSys Configuration Service 4 available 4 installed) for anyone who needs one. Just give us a call. Also, if you want a good rackmount system, Sound Chaser can configure a real nice one for you. They can preinstall the software and hardware and everything.
Dave of NemeSys