I used to have a Tesla K40 which I borrowed from the Gfx dpt from the company I work in. I tested it during a weekend, and, after a support from the IT dpt (from the same company), it worked perfecly.
After this test, I planned to buy one ... But the investment/return is not perfect as you can use CUDA for reverbs only due to kernel's size.
If Acustica found a way to make Acqua automatically benefit from such a card, in an eq/compressor scenario, that would be a game changer. I seriously doubt so, because I recently ask G if Acqua would run over Reaper Remote to overcome the limitation of one computer and the response was negative. He would had probably told me at that moment that cuda could solve this.
Kels, that's over eight grand that card. Wow! Did you try it with any Nebula reverbs? What exactly did you test it on? One of the cheaper Tesla cards might be enough, if it was just to use for reverbs.
yeah, a lot of money but, if CUDA was the response for an infinite amount of Nebula instances, I would buy a Tesla asap (considering the good investment it can be)
what follows might be a little disapointing : CUDA helps a bit in realtime, a lot better at render time. I've intensively tested with reverbs only. All kernels have to be the same lenght and they must have a certain minimum lenght too. It does not work with every reverb though, whatever the previous considerations, and, at one point, I sampled my own reverbs to test it.
I've done little testing with eq. For compression or limiter, it's nearly or completely impossible (time mode does not work w cuda)
so, my point of view is, unless you are mixing with 10 different reverbs that eat your cpu, cuda can't help. And even in this mode, some cpu power is still taxed.
Thanks kels for the response. You're right of course, that is very disappointing. It may be that multiple computers are the way to go. It would still be interesting to see what the G-man has to say about it.