NVidia CUDA support for Melda Plugins?

[Svama]

Hey Vojtech,

i'm loving your plugins and like to use them on every channel, but my QuadCore CPU comes fast to an end in Cubase, especially when using Oversampling.

What's about NVidia Cuda support to unload the CPU? Is there something possible or is it to difficult?

Thanks and greetings

Svama

[MeldaProduction]

Hi Svama,

unfortunately it is not that simple and there's is much more important stuff to do first (not that I wouldn't be interested in this one ).

Anyway a recommendation I give everyone - use upsampling ONLY if you need it. In many cases I have heard people enable it "just in case". Not a good idea! Upsampling protects you from aliasing, but since it is a pretty complicated filtering, it also introduces error as any other processing, so enabling it "just in case" does more bad than good. Ideally you should work in 96k and then you need no upsampling at all (except for some extreme situations). Personally I rarely feel the need for upsampling even in 44k, but I admit that my ears suck .

Cheers!

[pc999]

Just in case, dont forget the ATI cards/openCL.


BTW things like Fusion shouldnt help making this thing easier (at least on the latency side)?

[Moonwrist]

Old topic but any news about it?

I don't have the latest and newest computer but there is processing power unused in a GPU card.
This isn't a problem in every projects but only in the biggest one.
Would be nice to avoid rendering groups as there is almost always something to fix and then the wait for a new rendering to complete.

I don't know how difficult this would be to program but have you reconsider this?

[MeldaProduction]

In the forthcoming major update graphics will be completely driven by GPU, so far it looks like about 10x speed improvement! But only for graphics, so in most hosts you won't be able to measure this using the VST performance meters, only system CPU meters, which are sometimes a little problematic.

Anyway for audio processing - there are severe limitations of GPU uses (no matter if CUDA or OpenCL...), main problem is latency and the fact that audio signals are causal, so true parallel processing the GPU is so good for isn't really usable here. From what I know it's useful only for buffer-based algorithms, such as FFT, which are not so used due to all the problems it carries, except for analysis and reverberation. And as far as I know there isn't such a big improvement in this either, so despite the big boom some time ago, when everyone was talking about GPU, it seems it's not really useful in most cases.

[Moonwrist]

Well that is also improvement.
So do I understand correct that MAnalyzer, MMultiAnalyzer, MStereoScope and all EQ plugins whit analyzer on and lots of graphic information up front the new code will make them a little easier for the CPU?

Well, if it would be easy everybody would do it.
I know only Liquidsonic that have a plugin with GPU support. Reverberate convolution plugin and linear phase eq and I use that at mixing, so haven't really checked what the latency is.

[MeldaProduction]

Yes, analyzers and overall drawing will take much less CPU, hence also faster & smoother respose.

And yes, FFT based algos (linear-phase eq & convolution reverberation) are the only ones GPU is useful for. That's because these algos themselves have latency, so the fact that GPU doesn't respond quickly doesn't matter that much. However I'm not sure how much does it makes sense even for this - I heard that the GPU improvement compared to fastest FFT implementations with advanced instruction set supports (which we use) was very low, but who knows...

[CableChannel]

In the settings dialog there are a couple of options that you can set accordingly to save a little CPU. Won't do much, but a little. Like automation protections, interpolation quality etc.