audio processing on a GPU

I saw this article show up on my RSS feed

thought it was pretty interesting, I've never heard of a GPU being used in this way before, and frankly I'm excited about the possibilities here.

any thoughts, comments?

  Quote

A Special Offer to Join ACM

who the fuck is ACM? fuck ACM.

actually, can i change my name to ACM plz.

Tesselated audio?

jack up the anti aliasing settings for that smooooooooth, rounded sound.

make some nice squelchy basses with the anisotropic filter

then let the z-buffer rip it

I was hoping for some serious commentary

Serious comment: Hook John Carmack up with Autechre.

  On 5/10/2013 at 11:20 PM, oscillik said:

I was hoping for some serious commentary

 

Heh sorry man.

What part of it has you excited? I guess I don't know enough about audio processing but I'm not sure how processing on a gpu would be different than anything else... Is it the possibilities of object oriented sound design?

it must be the analog warmth of a decent GPU's shaders

  On 5/10/2013 at 11:27 PM, StephenG said:

 

  On 5/10/2013 at 11:20 PM, oscillik said:

I was hoping for some serious commentary

 

 

Heh sorry man.

 

What part of it has you excited? I guess I don't know enough about audio processing but I'm not sure how processing on a gpu would be different than anything else... Is it the possibilities of object oriented sound design?

 

well graphics cards are many MANY times more capable of multi-threading than a standard consumer CPU.

the most cores you can get in a CPU at the moment is 8 cores.

the amount of CUDA cores in my GTX Titan is 2688.

There are a lot of potentially mind blowing applications in physical modeling, but with overall prosumer music making trending towards real, analog, and retro-cheesy instruments I'm not sure the consumer demand is there. The CPU has been "fast enough" for just about any everyday audio processing purpose for the last 10 years.

GPU accelerated VSTs is a pretty obvious idea that afaik nobody has been pursuing because there isn't a need yet like there is in graphics or scientific modeling. Any time you have shit load of floating point math to do and are you aren't using the GPU it's a bit silly.

You mean like this sort of thing -

http://gpuimpulsereverb.de/

  On 5/11/2013 at 7:24 PM, cult fiction said:

There are a lot of potentially mind blowing applications in physical modeling, but with overall prosumer music making trending towards real, analog, and retro-cheesy instruments I'm not sure the consumer demand is there. The CPU has been "fast enough" for just about any everyday audio processing purpose for the last 10 years.

 

GPU accelerated VSTs is a pretty obvious idea that afaik nobody has been pursuing because there isn't a need yet like there is in graphics or scientific modeling. Any time you have shit load of floating point math to do and are you aren't using the GPU it's a bit silly.

 

 

Well the area where I think someone should capitalise on is the whole 'offloading of workload' thing. you can get those Waves plug in cards that offload DSP from the CPU to their own proprietary card. Costs loads of money and is a closed architecture.

Doing it this way, with the GPU, could open up this kind of offloading to a lot of people without the need to buy expensive plug in cards (of course, this depends on if you already have an adequate graphics card...my GTX Titan should be enough!).

And if the technology is open source, even better.

  On 5/12/2013 at 1:30 PM, mcbpete said:

You mean like this sort of thing -

 

 

http://gpuimpulsereverb.de/

that's awesome, I didn't know the GPU was already being taken advantage of in this way. Perfect way to do it too, reverb response files and all that.

GPUs are crazy powerful. newer iPads have very capable GPUs built into them as well.

people who say that iPads aren't capable of running pro-level synths / fx / full DAWs should consider the GPU potential of those things.

I agree with the general sentiment that CPUs are powerful enough to do crazy things. The main problem is that the CPU in your computer is kind of a jack of all trades, there to run either Windows or OSX (or some nerdlinger OS if you're a nerd), and everything running in that OS has to navigate that OS and do things their way.

The reason the likes of Kyma and the Eventide Harmonizer did such good jobs on audio is that they are basically computers specifically designed for DSP. The CPUs in the Eventide H8000FW are ancient in computer years, but the whole thing is custom made just to process audio, so it's the perfect tool for that.

No matter how fast consumer computers get, you'll always have to deal with some buggy operating system that's there to try to do everything you want, rather than one specific thing you want.