Quote:
Originally Posted by
norbury brook
➡️
audio processing is not well suited for parallel processing. And even if you somehow had 1000 audio tracks that could theoretically be processed in parallel, your GPU won't be able to do that, because GPU cores are not as independent from each other as CPU cores. Essentially, the GPU can only compute in parallel if the hundreds or thousands of tasks that are being computed in parallel are more or less the same, just with slightly different parameters. This is true for crypto mining, where you just run the same computation with a different input parameter for each task, or also for deep learning. It isn't true for audio, unless all of your 1000 tracks are doing essentially very similar things and for example have the same effect chain (but then, you would probably just sum them up to one channel and put the effect chain on this channel). Sure, there are some audio processing tasks that can computed in parallel efficiently like convolution reverbs or additive synthesis, but how often are you really hitting the CPU cap just because you have too much of this kind of stuff?
M
Thanks for that norbury brook, makes sense, and the analogous example of crypto mining helps frame it in your explanation.
Yeah, I'm not having cpu cap hits myself at the moment, and I was asking as different devs and whatnot were talking about this a long time ago as perhaps where certain things might go, and I will look more up on the nuance,
as yeah, I don't use things like Acoustica myself now, but I do know their products are cpu intensive, and I don't know enough about their product line to slot it in the sheer "convolution" category, as I imagine it isn't... (and I don't know how much additional coding resources would be required to handle that kind of project with so many moving targets etc... if it's a lot, or not... I guess I have my answer, in that if it was a simple thing, it would probably have been done already with the rise in proliferation of people employing stronger gpu cards/resources)
...or if cpu intensive products along those lines would benefit offloading some to gpu's (as obviously having an m chipped max computer and doing audio primarily, ... my video needs are a little overkill for 'me' in having 30 gpu cores (and I know there is a difference in the underlying architecture with that, and say, an Nvidia gpu) , but went with that chip for the needed p over e differential on my machine...,
(more in regards to the last iteration, as I don't have latest m4 or whatever, and I believe the distribution, let alone the ability of the next generation chip, on that has changed).
But appreciate your insight.