So I upgrade to v12 from v11 just last week due to the sale, big mistake it seems. I saw the other post talk about this too.
An mp3 that is 3:52 long as an example from my test, on a Win 10 (latest updates), AMD 5950X, 64GB RAM, NVME drives, configured to use NVidia 4070 for unmixing at High in v12 or Extreme in v11 settings:
So more than double the time to unmix the exact same song!
The quality in v11 is better too, more artefacts and separation isn’t as good in v12! Gone backwards!!
that is weird, it’s a major downgrade for me from v11. Same PC I have, did another test and it’s the same results, takes much longer with v12 to unmix a song than v11, and it’s definitely utilising the GPU as it goes to 99%!
The unmix models are different. Sometimes these models even change after in-version updates. They also use massively more VRAM, more than 17 GB if you want all stems at once. So there was definitely a change on how much computational power is needed.
well, the new models are worse, will look at maybe posting a video showing examples. Cleaner unmixing in v11 than v12, and much much faster too as mentioned multiple times!
@MattiasNYC That’s a good idea. But it will likely bloat the program size a lot. I wouldn’t be surprised if the making all previous models available will take up additional 30 GB or even more.
Fair enough if they’re really big. Though I will say that storage is relatively cheap these days and that an alternative would be to download on demand and allow the user to just get rid of the latest one instead.
I mean, if things are really this way, that a new “better” model is actually not only taking twice as long to execute but is also apparently not better, worse even, then the option should really be there.
In general I found 2025 to be a bad year as far as Steinberg software and company practices are concerned. Quite disappointed.
@SoundOf Of course it depends on how many different previous model files need to be included for it to work. Maybe it’s not that much at all. Or maybe you need a similar sized model file for each previous iteration. That would add up quick.
The SL12 Unmix Noisy Speech module is definitely better than SL11. I had presumed SL11 was better when SLP12 first came out, then I did a side-by-side…SL12 definitely better UNS than SL11. Yes, Robin had put a LPF in there that users asked to be lifted; which he he did address.
There are 13 different unmix modules in SL12. Instead of stating absolutes across the board, provide some substantiation of your claims. If you don’t take time to do that, then it’s impossible to take your claims seriously.
So, first port of call to better address your claims:
1_ which unmix modules do you think are “worse”?
2_Describe the material you were running unmix modules on
