A complete SDK for getting high performance audio from your GPU.
Examples include the NAM player, which is defacto the worlds best guitar amp simulator. Support for lots of GPUs, including Apple’s M1-M4
1 Like
Attribution-NonCommercial-NoDerivatives 4.0 International
Are CC licenses even applicable to software/source code? I’m sure I’ve seen people saying they’re not at all suitable. But IANAL.
1 Like
I’m trying to grasp what that license even means 
can I use it without paying a license fee? (I have a commercial product )
They do not, other than CC0 (which means “public domain”)
CC actively recommends against using CC licenses for software: Frequently Asked Questions - Creative Commons
The Attribution-NonCommercial-NoDerivatives 4.0 International license they are using is most certainly not a FOSS license and I’d recommend anyone to stay away from using this SDK until they fix this ..
2 Likes
I have invited GPU Audio to come here and discuss.
Btw NoDerivatives clause for repos would be against the Github Terms of Service.
You have to be allowed to fork a repo and make modifications.
1 Like
Let’s hope GPU audio chimes in, there’s got to be a clear way for developers.
Here one can find their license agreement: Home | GPU Audio
Seems to be for internal testing only. No commercial or open-source distribution possible. I suppose they will offer a commercial license at some point.
1 Like
ok, it will be interesting to see how such commercial licensing would look like..
Judging by the “hello world” gain processor example, this seems more like a proof of concept than an SDK ready for commercial licensing. The gain example code is quite complex and requires you to deal with a lot of plumbing manually. Personally, I think I’ll be looking at this again maybe when there’s a v2.0.
4 Likes
As an aside, NAM is not a simulator, it is a snapshot modeler.
A lot of marketing messaging about decrease in CPU load.
However not a single word about the potentially detrimental overhead of copying audio from CPU to GPU and back.
IMHO GPU audio is only useful for extremely CPU intensive processing where the benefits outway the overhead of having to move audio data around… Interesting to see how this will evolve.
3 Likes
GPUs have a compute channel that’s separate from the graphics pipeline, I presume they’re using that. I’d like to know more about global compatibility, and fallbacks available.
I did try their convolution demo a few months back, but it didn’t work on an Nvidia 3090, which was slightly off putting. 
Doing things in mass parallel is the future though, brute force transforms are amazing.
yea, a million people have had this idea before. For this sort of thing, the idea doesn’t mean anything it’s all on the execution and proof that it actually works with high compatibility and scales.
A solution looking for a problem.
2 Likes
Not really. Even Macs have fairly good GPU power these days 
Also, it’s not just about the graphics pipeline any more.
I feel that, if I ignore it, it won’t go away!
1 Like
I contacted Alexander Talashov of GPU Audio: he told me he will reveal their licensing plan on their site soon.
About the pro/contra discussion here:
As a commercial developer, I can only say that there is a demand for GPU processing, because people want all the processing power they can lay hands on. On macos silicon there is unified memory, so the overhead with moving data around is gone, but even with the data being moved: neural networks cost so much power, it is profitable to do it.
If you as a dev can tailor to a market that values performance very much, adding GPU support will give you a sales edge. Ofcourse it is only profitable for big processing tasks like neural networks.
For me personally I also like to boost FFTs and sinewave generators: if GPU audio boosts them as succesfull as it does boost NAM, the initial overhead to adapting to the complexities of GPU Audio will be absolutely worth it.
The more elite nature of the GPU support can be a marketing advantage: we’re are all fishing in the same pool. If we can make plugins and apps for higher end exclusively, the pricing can be adapted to that as well.
As we grow as developers, we tend to get exhausted by new methods and frameworks. We have our own set ways to achieve anything with the existing systems, we become conservative. I admit that I am not looking forward of the added work of learning to work with GPU Audio. But I know it might very well be worth it, and that is what counts.
If it is succesful JUCE might get a wrapper for it, that will even out the programming overhead “plumbing” that is now visible in the GPU Audio SDK.
4 Likes
Well, I’m looking forward to trying it when I have the time, and remain positive for a good licence. It looks like a mess to install though.
As an aside, here’s some random examples of a DCT in the GLSL shader language (ALL code is on the right):
&
&
2 Likes
Hi,
Anyone tried the sdk ?
I’m looking for someone for an easy project using FIR which is already in there ?
have a great day.