GPU Audio - goodbye latency?

[HughBanton]

I may be behind the times, I only spotted this this morning. Does anyone know more?

https://www.gpu.audio/

https://www.attackmagazine.com/features ... o-latency/

Dream come true for me if applicable to FS creations

H

[Tepeix]

Maybe with an external DLL ? But i really don't know if it's possible as i never used this.

[R&R]

Dream come true for me if applicable to FS creations

Even my schematics would run then

Running a Virus emulation (https://dsp56300.wordpress.com/) via GPU though... THAT would be something else wouldn't it

[tulamide]

No. GPUaudio is a company, and it is developing closed source code, in order to sell their products. It only uses a VST3 shell (probably because no DAW wanted to directly support it yet), but it is no VST tech. It just acts like a VST on the DAW side, but uses exclusive code to access the graphics card.

[tulamide]

I just remembered something, that might make the topic more transparent.

The thing with latency is that processors, even with multiple cores, are sequential by design. There is always a tokenizer active, who hands the token to a process, which then has full access for some time, until the token is passed to another process. Because of the Gigahertz speed of today's computers, we don't notice it and the impression of parallel computing takes over. But if you are scrolling through a webpage, while listening to a song, and running a video, they only seem to be parallel. In reality, they all get time allowed via tokens in a sequential order. Now, to compensate for the time, a process does NOT have processor time, all processes use buffers. It is normally quicker to render tasks ahead of time and store them to be used, when needed. Something like realtime doesn't exist.

But, if you have two or more independend "CPU"s, you can get to real parallel processing. This is handled again using buffers, by letting one "CPU" do one job, while the other does a second. For example, in games, the GPU does all calculations needed to create an image, while the CPU does all the user code, like checking the controllers, calculating physics, updating positions, etc. You only need to be one buffer ahead for the whole running time, so at the start the CPU does the first calculations, then hands over a buffer to the GPU, and while the GPU creates the image, the CPU does the calculations for the next image.

This concept is not restricted to CPU and GPU, though. Every processor can be used for it, given it can be accessed. Julian Storer, for example, the mastermind behind JUCE, started a project in 2016, called SOUL, which simply stands for SOUnd Language. SOUL tries to access each and every processor available in a given system. For example, all normal motherboards, even the cheapest ones, come with a DSP chip, but it is to date only used as ADC/DAC (the so-called soundchip). SOUL can access it and let the DAC do the realtime work. It also accesses graphic cards, network cards, whatever processor it finds and can handle. The fascinating part about SOUL is, that the programming is done with a totally secure script language with JIT compiler, which enables us to use an easy and streamlined language to program, while at the same time getting a speed faster than C++ compiled code. Most important, SOUL is not another sound plugin format! It doesn't replace any existing plugin format, but acts as a helper tool, which you can integrate into your programming environment, where it just takes over the critical realtime part.

Sounds fantastic, right? Now the downside. To finance the whole thing, in 2018 Julian made a deal with ROLI. In 2020, owning the rights to SOUL, ROLI went bankrupt and restarted with a smaller, more LUMI concentrated company layout. Julian Storer left ROLI, and only the open source part of SOUL is still accessible (which is not the whole SOUL). He immediately stopped working on SOUL and it is abandoned for over two years now. However, with NI bought by Soundwide back in April, it was revealed that a new branch for Julian Storer was created, the company Sound Stacks. NI just said about it "They’re working on innovative new audio platform technologies to benefit developers across our whole industry."

But Julian was a little more precise on the JUCE forums.
In 2021, before the Sound Stacks reveal: "Ces and I resigned from ROLI a year ago and that was the last time we had any contact with anyone at that company. The work we did on SOUL belongs to ROLI, and it was never our place to comment on what they choose to do (or not do) with it. Maybe they had other things to worry about last year…

However, if you liked the direction we were trying to take things in with SOUL, keep a look out for our new stuff, as we still believe those general principles are the future for DSP coding"

Regarding reviving SOUL: "But for our own plans: no, we’ll certainly stay well clear of that name."

It's to assume, he will repeat the same mistake, by programming for a company instead of open source it, which would keep it alive if something happens. Also, it is safe to assume a closed source product you may have to buy from one of Soundwide's brands.

But definitely in the same direction as SOUL went.

[R&R]

Interesting stuff... that they managed to utilize other almost peripheral systems to do work.

I wish there were more innovation like that...
- Too slow?... just throw in a [fill in blank] and the system adapts to manage task. In the spirit of decentralized mesh technologies.

The future we never get to see I guess?
Since the demand for large profit (often times just to keep companies afloat) usually outweighs any sustainable and truly longlived hardware technology?

[HughBanton]

Thanks for all this, it's most interesting.

Hey, Julian Storer's a clever guy! If you have an hour to spare, worth watching this ADC18 lecture he gave, back when SOUL was taking shape :
https://www.youtube.com/watch?v=-GhleKNaPdk&t=910s

Whether or not any of this ever has relevance to Flowstone itself, it can't fail to be an inspiration re the future of the audio & musical stuff that obsesses us all here.

H