View Single Post
Old 26th November 2012
Then you look at the ProTools Thunderbolt Native, which is using the computer's CPU to run all the plugins etc.

In all honesty, dedicated DSP hardware is the wrong route. It would make more sense to be able to slave up a bunch of computer workstations.

If I was designing a DAW from scratch at this point, I'd first start off with the Linux kernel, the type they use for multinodal supercomputers, and build out from there. It wouldn't be an application running on an OS such as Windows, with all that overhead and unreliability, it'd be an embedded system more like RADAR which runs on its own motherboard(s), and hooks into a desktop PC platform. Run just the user interface on one computer, probably running a desktop OS, and do the heavy lifting on a bunch of dedicated motherboards, say 8 tracks per motherboard. The latency could get very low indeed if it was implemented properly, it's the memory management and process switching that tends to produce latency on CPUs, whereas it's possible to tune an OS to be very low latency, as iZ have done in the case of BeOS. One could also tap into the SIMD systems on graphics cards for doing processing tasks.