CUDA or OpenCL engine

That might be interesting, but AFAIK the current implementation of CUDA is not yet viable for audio processing because of added latency? At least that's the case with currently available CUDA convolution reverb engines...

Also, if this is tackled with Pianoteq, I wouldn't really be concerned about polyphony. Max of 256 is perfectly fine for even the most elaborate work. I would rather concentrate on making the core sound calculating engine more complex, allowing more piano intricacies to be modelled to the most faithful degree (the attacks, "the woodiness", better soundboard, more complex resonances etc.).

Would be still cool to have a more detailed sound model by enabling CUDA processing, along with increasing poly.

Thanks for your input, I'll vote positively!

Answering briefly, it is standard coming from nVIDIA for driving computing made on GPU (graphic card processor).

GPUs have slightly different architecture than regular CPU (central processor, usually coming from Intel or AMD). Modern GPUs are tens or hundreds floating point processing units that can do no more than floating point numbers calculations, while CPUs are just few cores made for general purpose such as device handling, floating point calculation but also integer calculation, operating system and process/application handling etc.

While single GPU processing unit is weaker in terms of performance than single CPU core, GPU units simply outnumber CPU cores so much so altogether they have much more computing power than CPU cores.
That's why for all algorithms or processes that can be divided into separate calculation chunks GPU computing seems to be good alternative. And IMHO this is the case of audio synthesis here in Pianoteq.

You would need to have a CUDA-enabled graphics card (I think?), and I believe DirectX 10.2.

Good idea! I also voted positively!

I'd like to point out, though, that I agree with EvilDragon that the main advantage of this would be the opportunity to use a more sophisticated model rather than the higher polyphony

Then simply vote positively.

Yep. Because there are a lot more parallel computations done in a graphics chip than in CPU. Your CPU is 2, 4, 6-core. GPU is like a multi-million-core, with each core doing floating-point calculations on each pixel that gets displayed on the screen (for example, 1920x1200 resolution = 2304000!)

I presume this would greatly lift the heavy load that goes to main CPU, and yes, allow even more computations to be done in realtime. More computations = more precise model = better and more nuanced sound!

You didn't hear that Wallander Instruments is secretly working on WIVI Strings?

Just so you know, currently WIVI wind instrument player can manage about a 100 separate players on a normal 2-2.5 GHz dualcore CPU! How's that for using the available CPU?

Also, CPUs don't really need to evolve much in terms of speed - speed is pretty much enough as it is now, the only evolution will be in parallelization (going for more and more processing cores). And actual software support from certain programs for multicores is weak! Majority of VST instruments use only single core, for example! It's not CPU's fault - it's the programmer's fault for not utilizing the available multicores correctly! Thankfully, Modartt does this!

CPUs will still continue to evolve rapidly. However, the whole idea of putting a GPU in work aside from graphics is simply great!
You won't have an idle set of processing units waiting to shine only at the video tasks of your PC, instead you can utilize them to their full potential!