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

[Xenakios]

and subsequently this...

The developer posted that here on KVR and then disappeared, no more posts since May 2018. I guess they realized GPUs are not going to bring anything useful and marketable for audio processing, like so many developers before them.

[mcbpete]

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

Just tried two simultaneous tracks with Eventide's Precision Time Delay, shifting one track by 2ms and aside from the obvious phasing issues it really is nothing - Hell if you've speakers 2.5ft away from you you'd get a larger delay than that !

[EvilDragon]

So why should we have a (much) larger delay than the one from speaker to you if we don't have to? As if it's not enough that audio buffers result in added round-trip latency, GPGPU plugins would add to that...

[PeterP_swe]

GPU processing isn't that different from what UAD is doing with their hardware and plugins, so I can see it work. But there are drawbacks in form of extra latency and you need specially written plugins for it work.

ROLI is doing some interesting experiments with "sound shaders" (see soul-lang.org), I can see that using the DSP from GPUs. But of course, the same drawbacks here again.

[logifuzz-vst-plugins]

Beyond audio and the usual graphics/video/crypto mining, are anyone putting their GPU cores or extra memory to some other use?

[THE INTRANCER]

Beyond audio and the usual graphics/video/crypto mining, are anyone putting their GPU cores or extra memory to some other use?

They were utilised in 2004 for the Spirit Rover on Mars and NASA's GPU cluster at the NASA Center for Climate Simulation (NCCS)

https://www.geek.com/games/nvidia-techn ... rs-555177/
https://sciences.gsfc.nasa.gov/600/high ... /gpus.html

[DuX]

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.

Young musicians and producers never learned how to conserve resources [mixer channels, FX, polyphony etc.] That's why their projects use much more resources [CPU, RAM] than mine, for the same, or even better result.

I see people loading new instance of Kontakt on every track and incredibly "funny" things like that. Never realising that one instance of Kontakt can contain hundreds of different sounds on 4x16 different MIDI channels, and 32 different audio outputs. Never thinking "I should make this bass monophonic" or "Maybe I should've used just one reverb on an AUX send instead of 24 insert reverbs". We who worked with a real studio know this stuff and how to conserve resources, but most importantly get the same or usually better sounding results. They don't. That's why there are so many people who are not satisfied with any computer, but they never question their abilities, or their cocked up workflow. Ego is a bitch, eh?

Having said that, I had high hopes for GPU DSP years ago, but then it turned out much more complicated than it seemed, and yes- the latency issues. So it won't happen any time soon. Shame, of course, but it just won't.

Learn how to conserve resources. Any computer manufactured in the last 5 years should be able to do anything you want, you just have to maybe do a bit of studying. Yeah, studying is so hard... no fun, eh?

Free newbie tip: keep the bass monophonic and mono. Same with the kick drum. All bass sounds should be mono, unless they go really high into the frequency spectrum, and even then you should make everything below 200Hz mono, because we can't hear bass in stereo. It's just a waste of resources to make it stereo, and the mix won't sound as good.

[pekbro]

I would think the key to using the GPU for audio is in asynchronous processing, not parallel.
The fact that we can compress audio should be evidence enough that it should be possible
if you hit it with enough computational capacity and some nifty predictive algorithms.

[Hanz Meyzer]

It is possible with OpenCL. There are working examples on github. But now Apple deprecated OpenCL, because they suck. It was not old at all.

[telecode]

Beyond audio and the usual graphics/video/crypto mining, are anyone putting their GPU cores or extra memory to some other use?

At my work we have a POC for trying to use GPU computing for statistical data analysis. Basically, some of the jobs take 2 or 3 days to complete, so we are trying to see if there is a way to use GPU computing to see if it can reduce it to less time to complete. It hasn't really taken off yet due to too much other stuff that needs to get done and not enough human resources available to devote a person to try to get it to work properly.

[Psuper]

If you plan on doing any audio recording with that DAW in your room, noise is an absolutely huge issue, especially as you increase the workload of the graphics card. And graphics cards aren't near as easy to deal with when it comes to silent cooling.

Just an angle it seems no one is thinking about, and is extremely important for the majority of home studios -- it was the #1 reason I went with a passive cooled card for the latest DAW I built -- and needed it because 4k output on motherboard chipsets is still barely workable in a windows environment if you need snappy performance moving things around.

[EvilDragon]

I see people loading new instance of Kontakt on every track and incredibly "funny" things like that. Never realising that one instance of Kontakt can contain hundreds of different sounds on 4x16 different MIDI channels, and 32 different audio outputs.

Hold on right there. While Kontakt does have its own multiprocessing support, at some point you get to the point of diminishing returns. There is definitely validity in spreading things across multiple Kontakt instances rather than filling one up fully with 64 instruments or whatnot. The tradeoff is RAM usage (60-80 MB per additional instance of Kontakt), but in most if not all DAWs (yes, Reaper included), CPU usage will actually be improved with multiple Kontakt instances used, for the same number of instruments loaded.

[DuX]

I see people loading new instance of Kontakt on every track and incredibly "funny" things like that. Never realising that one instance of Kontakt can contain hundreds of different sounds on 4x16 different MIDI channels, and 32 different audio outputs.

Hold on right there. While Kontakt does have its own multiprocessing support, at some point you get to the point of diminishing returns. There is definitely validity in spreading things across multiple Kontakt instances rather than filling one up fully with 64 instruments or whatnot. The tradeoff is RAM usage (60-80 MB per additional instance of Kontakt), but in most if not all DAWs (yes, Reaper included), CPU usage will actually be improved with multiple Kontakt instances used, for the same number of instruments loaded.

Thanks for clarification. That's true. But I was talking about the extreme cases when people do it automatically and I see it often. New sound = new Kontakt instance. Not so many people know about the advantages of either usage case.

[EvilDragon]

Yeah, that is also true. I can see that from workflow perspective just using one NKI per Kontakt instance is a lot clearer and simpler, no need to care about MIDI channels or outputs or whatnot. Much easier to just load something and go, and sometimes speed is more important than flexibility. Plus, as mentioned, in most hosts, using a single sound per Kontakt instance WILL yield in better CPU utilization, trading off a bit of RAM in the process.

[logifuzz-vst-plugins]

Beyond audio and the usual graphics/video/crypto mining, are anyone putting their GPU cores or extra memory to some other use?

At my work we have a POC for trying to use GPU computing for statistical data analysis. Basically, some of the jobs take 2 or 3 days to complete, so we are trying to see if there is a way to use GPU computing to see if it can reduce it to less time to complete. It hasn't really taken off yet due to too much other stuff that needs to get done and not enough human resources available to devote a person to try to get it to work properly.

Interesting project, mind me asking what programming language you guys are using to do the GPU parallelism? This would be a proper academic project if the statistical task can be benefited from multi process.