Hi everyone,
Whatever the AI processing device I set in the System parameters (CPU or RTX 4060Ti), unmixing process takes the same amount of time. (about 2 min).
The same unmix process in SL10 takes about 10 seconds (using GPU).
I’m only unmixing voice, using the “Fast” setting,
Anything I should try before uninstall/reinstall ?
All normal - there’s a limit regarding GPU acceleration on Unmix Song (due to new models/new libraries) that will likely be improved with the next patch, if all goes well. If you’re using the High or Extreme mode, you can already expect some GPU acceleration - and hopefully more with the next patch if all goes well.
I think it’s more than just a limit perhaps? I just logged with GPU Profiler SL V11 (vocal, bass, drum and other took 1:52 on a 3080 using Fast settings). As you can see the GPU use is almost none compared to V10 which did the same Fast settings in 0:10
V11 - 1:52s
V10 - 0:10s
It seems V11 starts using GPU for a few seconds, then stops and all processing done on CPU. V10 is barely using the CPU and using 60% of the GPU.
I was about to post about this
SL 11, for me is utilizing 100% CPU nearly 0% GPU ( using nvidia 3080 )
I really hope if we can get benifit from GPU acceleration as it is faster and less CPU intensive
Good to see I am not the only one who’s noticed this.
I was terrified when I tried to unmix a song using my GPU and it took an eternity.
I hope this will get fixed soon.
Hopefully a bug, it looks like it starts using CUDA then switched to CPU to me.
Heh… and it crashed as soon as I posted this…
Got stuck at 99% for quite a bit and crashed.
Yep mine reaches 98% then everything hangs for maybe 20-30 seconds and then does the last 2% and puts the stems into the window. Lot of crashes too with GPU enabled or disabled.
Would it not be an idea to still include the GPU-enabled models in V11 and let the user decide? To be going from 10 seconds to 1:52 minutes to separate a 3min song is quite a different workflow needed, perhaps not actually usable yet.
Is there any time-frame for when the GPU acceleration patches will be arriving? I am very eager to upgrade to 11 (from 10), taking advantage of the 15% discount. But my old i7 6700K pc will not take well to the unmix cpu load - which is why I put a RTX 3060 in my machine.
Optionally, is there a way to make Cubase 13 Pro see Spectralayers 10 instead of 11, as an extension? (I.e., when both 10 and 11 are installed.)
Best,
Magnus
I hope to have a patch next week, but doesn’t mean it’ll be published next week, it can take longer unfortunately…
If you prefer to have SL10 as an ARA plugin instead of SL11, just install SL10 again after installing SL11.
Thanks for the very quick reply, Robin. So hopefully we are talking weeks, not months, for the gpu acceleration? If I remember correctly, gpu acc was implemented in the first few patches of SLP 10 as well.
Best,
Magnus
When I tested it in my environment, there was a difference.
usage environment
CPU Intel Core i9 13900K
GPU RTX 4060 GDDR6 8GB
Analyze a 4 : 08 song (UNMIX) and select the highest quality
8 : 54 using GPU
14 : 35 using CPU
Yep, you get a bit of GPU assistance with High and Extreme modes.
Have they fixed this yet? Is there a way to set it to use GPU?
Or have they still not worked out how to use the GPU yet?
Unmix doesn’t touch my RTX 4090 at all…
RipX is way way way (did I mention way?) faster due to GPU usage.
@Darcey404 have you activated your GPU in Edit > Preferences > System ?
Ah, I overlooked the setting… Thanks. 
Shame they can’t auto detect that you have a GPU…
I just turned it on and gave it a run and:
CPU
12 to 15%
1 core is close to maxed out, and another spikey at 45%
GPU
7 to 22%
System
i9 9900k @4.7ghz on all cores, 128GB Ram, RTX 4090, Scarlett 8i6 Gen 3
Well at least it’s using some of the GPU now, and is quite a bit faster than it was while using the CPU.
As an aside, RipX only works with certain NVIDIA cards, and requires a specific version of the CUDA SDK to be installed. SpectraLayers, on the other hand, works with all modern GPUs and is not restricted to one manufacturer.
True it does pref’ PC Nvidia but will use CPU if no decent nvidia graphics card is available and the correct cuda toolkit version is not installed (free download).
RipX will use CPU if you don’t have the following NVIDIA graphics cards :
1070, 1080, 1080Ti, 2070, 2080, 2080Ti, 3070, 3080, 3090, and 40-series cards.
NVIDIA CUDA Toolkit 11.0 is free for download.
Note: When using ML (aka fake AI) via python with Pinokio, LLM Studio, ComfyUI, a1111, forgeUI and countless other AI stuff it always works best on Linux and nvidia gpu’s, then windows and nvidia gpu’s, as for mac, no idea, but I wouldn’t try it.
Well yes, I tried RipX today and it is very good, indeed.
