GPU Driven | The Future Of DAW Processing Or A Dream ?

[THE INTRANCER]

I was just thinking, having recently upgraded from a cheap old graphics card I had been using for a whole 9 years to a new kickass new one that was just released a year and half ago. I've been keen to discover and exploit the benefits that it has the potential of providing...and needless to say, it has in many ways. The most significant in relation to DAWs being screen resolution and being able to display the DAW itself, far beyond that of the 1080P native spec resolution upto a ridiculous 4K if one wanted to..

This had me thinking about the untapped potential of GPU driven processing of audio within DAWs... Is GPU audio mixing more of a complication of unknown entities in this area or an opportunity to take a brave stride in a new direction for DAW developers to set themselves ? Apart from the crowded oldskool ways we have been use to for the past 30 years ?

Before we could take the computational restrictions out of the equation simply by using external hardware connected by midi cables. Sure there was basic trackers and such like in the 1990's but if you really wanted to do some serious computer sequencing with limited computer hardware, external hardware was really how most serious producers wanted to work. Today though and with the Midi 2.0 protocol around the corner, things are continually moving forward in processing midi and audio... and not going back...maybe a side step with Midi 2.0 but anyway... back to the topic..GPU processing in DAWs.... who's using it, what audio software is using it..and is there a future for it....?

And if you're wondering from what GPU I went from and to.. well here's a little video for ye to all enjoy...if you've not been tracking my posts in the past few weeks.

[EvilDragon]

There have been threads about this already. Audio processing is in 99% of cases NOT a highly parallelizable process (which is what GPUs excel at), it's a serial thing, plus there is the issue of added latency because of data transfer between CPU and GPU.

In short, it's not going to happen in the way everyone wants. GPU processing is ok for stuff like FFT and convolution, but it doesn't lend itself to, say, calculating complex serial effects chains or synths.

[THE INTRANCER]

I found an interesting article relating to GPU's and music software ...

https://www.pro-tools-expert.com/produc ... e-pictures

and subsequently this...

[EvilDragon]

So 1 ms latency at 96k. That's 2 ms at 48k. I wouldn't find that acceptable

The article you linked to doesn't mention any of the well-known pitfalls of GPGPU...

[chk071]

There have been threads about this already.

Yep, loads. Seems like one of those things which just can't be killed...

[telecode]

I wonder if GPU will be useful down the road in 10 years when DAWs become more powerful and do most integrated stuff. Maybe a DAW that does 10000 tracks and loads all VSTs on the system at once and does auto mixing as you make the project?

I never thought you would ever need that many tracks,.but.. I have been watching the workflow of some of the new young chilllstep artists , and they use DAWs completely differently than older generation musicans that were weened on tapes and physical recording studios. Their projects are comprised of 100 tracks of little 10 to 20 sec snips of music. I can totally understand why they need the pro version of DAWs with unlimited tracks functionality.

[Apratim]

I had read some old article and all the things i could found that gpu audio generation was great but latency was the only issue... Don't know whats the scenery now in gpu audio scene

[sleepcircle]

i would like it if people used gpus for plugin interfaces more. some of them are very pretty, to be sure, but good GRIEF having the interface open takes a lot of processing power.

[Apratim]

This may help

[THE INTRANCER]

Digging up more on this in relation to the AMD side and audio processing where they seem to have had a more open source vibe to them over the years than what Nvidia perhaps have had with their propriety libraries until they made PhysX 4.0 SDK fully open source...

We have... https://gpuopen.com/gaming-product/true-audio-next/

AMD TrueAudio Next is a software development kit for GPU accelerated and multi-core high-performance audio signal processing. TrueAudio Next provides pre-optimized library functions for computationally expensive algorithms such as time-varying audio convolution, FFT/FHT, and audio-oriented vector math. Sample applications and examples are included to facilitate integration into audio applications. The SDK also provides GPU utility functions that support AMD GPU Resource Reservation, a technology that allows audio to share resources on the GPU with graphics while avoiding impacts to quality-of-service.

TAN SDK Features

Fast OpenCL™ parallel batch convolution engine
Multi-threaded, multi-queue, real-time architecture
Extendable open source API
CPU and GPU implementations for all functions
Sample applications for GPU acceleration of high-performance audio for gaming and professional applications

Benefits...

The TAN SDK provides real time audio applications the ability to leverage the OpenCL™ toolchain and to co-exist with graphics.

Accelerating game audio on the GPU frees CPU bandwidth for game physics and AI, while enabling advanced interactive acoustics modeling on many audio streams. Advanced audio processing has greatly increased compute requirements:

Physics-derived audio benefits from real-time, time-varying convolution performed on every audio source. TAN Convolution leverages the power of the GPU to provide reliable time-varying convolution for many audio sources, without blocking the graphics compute queue.

Higher-order Ambisonics rendering (3rd order or higher) adds the most environmental realism, but can require 16, 25 or 36 time-varying convolution filters per audio source. TAN makes these higher filter counts practical and simple to implement.

Most advanced audio algorithms depend on high-performance FFT or FHT (Fast Hartley Transform). The TAN library enables these algorithms to leverage the GPU.

A mixing application is included so that multiple audio sources can be mixed locally on the GPU to minimize data transfer overhead.

The TAN library may be used in conjunction with Radeon Rays as foundational accelerated algorithms for a complete physics-based audio solution.

[Tj Shredder]

The future will give us more of everything anyway. I don‘t care how. If its GPU, why not, if its ZPU, PPU or poops fine as well. What I would care about is that I don‘t need to know those details, that I don‘t need to decide between A or B. I am afine to technology as long I am able to control it and do the developement myself. But as long as I do not intend to program my DAW myself I don‘t care at all, let the devs decide...
Whenever my setup is too slow, wait a year and its faster than you could imagine the time you bought the fastest machine available...
And it will break what worked before as well... (which in turn will cost you time as well...)
I remeber well the time the first affordable 8-track hard disc recording appeared, it was heaven compared to my Fostex 8-track, all what I ever wanted and ever needed... Now the kids want 100 and more tracks but the music doesn‘t get better that way, maybe more lazy at least...

[mcbpete]

So 1 ms latency at 96k. That's 2 ms at 48k. I wouldn't find that acceptable

Sounds OK to me - Apparently that's the same time it takes from a drummer hitting their snare to the sound hitting their eardrums

[bibz1st]

If you take a look at Fathom forum, Everett, the developer, is quite sure he can use the GPU for his synth, whether that is something different to using it for a DAW, I dont know, is it ?

[EvilDragon]

So 1 ms latency at 96k. That's 2 ms at 48k. I wouldn't find that acceptable

Sounds OK to me - Apparently that's the same time it takes from a drummer hitting their snare to the sound hitting their eardrums

Not OK if you want to monitor with FX while recording live.

[Guenon]

The problem with this demonstration is, the CPU load is actually quite high, not somehow miraculously low (like, "look at this, it's this low, the GPU is doing all the work" low).

Just tested it on an almost seven year old system, threw in ten instances of the the first delay plugin I spotted that was installed on that machine (Unfiltered Audio Instant Delay), then five instances of the first flanger I spotted (U-he Uhbik-F), all in series on one track, set the buffers and sample rate to correspond with the video. Reaper CPU fluctuates between 00 and 02 in the Windows task manager exactly in the same way they highlight in the video

If one is doing a proof of concept video where the CPU load is showcased, why not run something that shows some different figures than a regular system from years ago?