Sorry I'm yet to receive my very first 1660 ti graphic card parcel to experiment this out.. at the moment I'm on 3200g built-in graphic card which hopefully, when my all parcels arrived (together with 3600x), I then hopefully can be able to experiment this out (once done assemble my new pc upgrade).. hopefully..

And so, while then, I thought I share with yall this video on how to set things up with GPU and any Windows Apps.. Just wondering if we can make use GPU power to accelerate DAW (Can we?) Happy experimenting first.. toast and share your results #learning #experiment

 

Last edited: Jan 12, 2021

Reverberate LE plugin used nvidia cuda, but somehow it didn't go further, as i understand there are latency problems when data goes from cpu to gpu, and then back to cpu

 

there were attempts do do that since around 2004 but i dont think there is one real product (except the above)

it would be especially nice for complex decovolution processes, or to put it simply, for reverbs.

and i think it is correct to say that technically, the UAD cards had "GPU-like" FP processors.

 

Last edited: Jan 12, 2021

Audio related calculations are sequential.
GPU is great at handling parallel instruction.
But sadly, GPU can't handle sequential instructions better than CPU.

Just FYI, if you want to render a picture of beautiful flower and a butterfly on it, you can work on rendering the flower and butterfly discreetly. The rendering process of flower and butterfly are parallel. But say, in a special world, where the butterfly has the power to change the color of the flower using the stroke of it's wings, there, the color of the flower will be a function of the butterfly's will and so, rendering this picture will become sequential. So, if you have an EQ and then a Compressor and then a Reverb on a channel, they cannot be worked upon discreetly. Compressor will depend upon the signal coming from the EQ and the Reverb will depend upon the signal coming out of the compressor, which as mentioned above, will depend upon the signal coming out of the EQ. So, this interdependency will cause issue to GPU.

Now, we can use GPU for just the EQ or just the comp or just the Reverb but then, there will be huge latency and issues because of CPU-GPU communications.

 

I already had that technology in 2013, maybe you haven't heard about LucidLogix Virtu MVP?

 

I have a Nvidia Quadra K4200 which allows the use of the Cuda cores and have used LiquidSonics Reverberate and Filtrate on a few occasions, latency is really high and not really practical for real time use, so only ever used to process and bounce.
I think I'm right in saying that only the older LE version works with Cuda and only available in 32bit....results were fairly good.

Acustica Audio's Nebula 3 also have Cuda optimisations, party due to having a couple of UAD cards and because Nebula was taking too much disk space I uninstalled before having the K4200 - You reminded me that i should try Nebula again now that i have more space available!

Another is https://gpuimpulsereverb.de/
Get cuda-z http://cuda-z.sourceforge.net/

Im not sure how much benefit one will see from Cuda, ultimately the increase in CPU power over the last few years meant we were not as restricted with plugins as we once were, it was promising but poor real time performance meant that for DAW the idea was left in the past.
If that could be overcome somehow there could be some possibilities in the future...

I spent a lot of time going down the rabbit hole researching VST and Cuda, the potential appeared to be there but never fully realised.

That said, you got me googling to see if much has happened with Cuda over the last few years...
Not much by the looks of it other than these guys seem to be doing some interesting things: https://twitter.com/GpuAudio
https://mediamotoreurope.eu/meet-the-founders-gpu-audio-unleashing-sound-professionals-creativity/

 

Last edited: Jan 12, 2021

Very good answers here.
It's complicated for a number of reasons but yeah, the latency related to the parallel-oriented type of GPU processing is the main one.
Let's just say if you want to compute a short FFT many times per second as is the case with DAWs. That latency overhead due to back and forth communication between the CPU-GPU (and their RAMs) isn't worth it.
If it was a quite long FFT, or several ones, it'll be worth it.

So yes, the idea is very cool, but real-time audio is one of the worst use cases. As many of you know, for a lot of other number-crunching uses like servers, AI, Video FX, scientific research, cryptomining (get a hint Ethereum) is widely used.

As some fellows said, Nebula is the only plugin I've seen with GPU support. Makes sense because is one of the latency killers.

Now, I'd love to see how it'd work in Apple's M1 chips. Unified CPU/GPU/memory in the same chip. I can't know for sure but I bet is feasible on those new babies.

 

Last edited: Jan 12, 2021

Russians already making it possible. They promise low latency as possible without dropouts at highest sample rate
https://www.facebook.com/GpuAudio/

 

Now, we have Threadrippers (16 core or 24 or 32 or 64 Cores so for threads, double these values). So, at this moment, if you have money, lack of CPU power is evitable.

 

how much latency does that introduce ?

 

Here I found a PDF about, it says that needs a new Kernel for low latency, like a custom linux

https://www.swisst.net/fileadmin/swissTmeetings/ECC2018/Referate/1A2_Rosenthal_ZHAW.pdf

 

Last edited: Jan 12, 2021

how should that work, doesn't it also work with a normal GPU from intel so standard? I own (Intel (R) Core i9-10850K CPU 3.60GHz 3.60 GHz 64.0 GB Win 10 Pro). i don't have a graphics card

 

damn Coronalypse, I need more social life or won't get laid for the rest of my life

 

but it can't be just latency related, we use a lot of processes that induces latency.
for instance. All Zynaptic plugins would be suitable, and all acustica plugins would benefit from it (ZL instances aside)

I think the overall benefit in processing power is not as big as people believe it is

 

Don't know about the subject of your post but the prices of GPU's and CPU's are very f##king high right now. Why would you buy 1660ti if you can get the new 3060 for around the same price if you could find one because they get out of stock pretty fast!!

 

Ive been wondering if an external gpu could possibly help reduce stress on my macbooks cpu. I have an intergrated Intel Iris on this machine, and when i hook up an external monitor, the cpu is what is taxed.

 

Interesting and informative in thread here thanks to y'all pointing things out I was on random on YT and saw about crypto mining farm which makes me wonder if we can do other things using GPU card for crunching software.. I was thinking something like can it be more better result than CPU OC.. theoretically he he..

 

Because you won't find a 3060ti for less than 590-600 euros, at least not in Europe.

 

i would to harness the power of my 3090 in my daw but it only works for certain plugins like fab filter.
Oh and theres a 12gb 3060 on the way so go for that one if its for musicif there is a thing about plugins useing gpu memory.