Quick Sign In:  

Forum: VirtualDJ Technical Support

Topic: Stems 2.0 are terribly slow on an RTX 3050 laptop
As the title says, the new stems are terribly slow even on hardware that is supposed to support real time stems separation. A 2 minutes file takes like 5 minutes to be fully analyzed when placed in the Stems folder, despite my GPU reporting ~100% 3D and dedicated memory usage in Task Manager.

I have already tried the "Don't use GPU" option and I have already verified that I am running the latest NVIDIA drivers.
 

geposted Sun 22 Jan 23 @ 2:59 pm
This has been covered several times now but the RTX3050Ti with 4gb VRAM does not support stems. It will be reverting back to CPU therefore causing the long loading times.

Atomix need to let people know this as I highlighted the issue on a brand new laptop several weeks ago.
 

geposted Sun 22 Jan 23 @ 3:01 pm
Should've done some reading on the forums prior to posting that. But still, Atomix has lied to us...
 

geposted Sun 22 Jan 23 @ 3:09 pm
They have said they're looking in to a fix but nothing as yet. The issue was only discovered after the stems 2.0 release.
 

geposted Sun 22 Jan 23 @ 3:19 pm
ASUS zenbook UX8402za windows 11 22h2 home 12th gen i9-12900h 2.50 Ghz 32 gb RAM, nvidia geforce RTX 3050ti. VDJ2023 b7388. Rane ONE ASIO 1.0.1 .

several optimization settings , different ASIO latency settings, and windows power settings.
STEMS 2.0 works fine , VDJ works fine, and no processor or RAM spikes. GPU on NVidia will run 95% on stems precompute for one minute depending on mp3 file size.
but i can do several hours of no glitches, then just erratic audio latency static.

i continue to receive random audio cracks distortion static.

has anyone reinstalled windows 10 ? over a preinstalled windows 11 ASUS laptop.

would this be a viable option to try?
 

geposted Wed 25 Jan 23 @ 5:59 pm
Check the latency. Nvidia video drivers cause bad latency when driven hard. That's why I returned my Zenbook.
 

geposted Wed 25 Jan 23 @ 6:22 pm
kradcliffe wrote :
Nvidia video drivers cause bad latency when driven hard


No such issues here, on desktop or laptop.
 

geposted Wed 25 Jan 23 @ 6:25 pm
Maybe just an Asus thing but major issues with nvlddmkm.sys and it's well documented too.
 

geposted Wed 25 Jan 23 @ 6:40 pm
thanks, i'm going to have to switch to AMD with ASUS.
meanwhile continue to search for a settings fix.

i can't seem to get around the dpc latency problem with ASUS board and the NVIDIA 3050ti i will do a few more test with no stems computing, because STEMS will run the NVIDIA 3d gpu up to 95% and 98% for one minute.

i was waiting for maybe an updated driver from RANE, or NVIDIA, or Microsoft that will help the glitch.

both of my windows 10 lappys work without a problem.
 

geposted Thu 26 Jan 23 @ 3:47 pm
I have observed the same issue with stems to be processed really slow, but was able to solve it by removing the graphic card in the device manager incl. removing the driver. After reboot it was reinstalled and virtualdj was able to use the card and a normal song took only 30 sec for stems to be analysed. In addition I deleted all *.dll´s in the Plugins64 folder in the VirtualDJ folder. It was more a pre caution, but it seems to work. In VirtualDJ I used the settings GPU - Use High Power and Acceleration - nvidia
 

geposted Mon 20 Feb 23 @ 10:23 am
castonHome userMember since 2020
this is the VDJ problem of the stems engine, because when you use serato stems on the same hardware it works great, you don't need a great graphics card. the VDJ stems engine is just not very efficient. bug.
 

geposted Wed 15 Mar 23 @ 7:28 am
user22137884 wrote :
this is the VDJ problem of the stems engine, because when you use serato stems on the same hardware it works great, you don't need a great graphics card. the VDJ stems engine is just not very efficient. bug.


100% false.

Stems 2.0 is leaps ahead of Serato stems.
Serato stems is the equivalent to the first iteration of Virtual DJ stems.

Stems 2.0 does require a high end GPU. It’s also been well documented that the 3050 with 4gb of VRAM is insufficient.

You could change stems to lower quality and then you will see it will work and sound very similar to Serato stems. But it will not sound as crisp and clear as Stems 2.0.
 

geposted Wed 15 Mar 23 @ 7:45 am
castonHome userMember since 2020
my opinion is constructive for dj who can use different software but not for virtualdj therefore you delete
 

geposted Wed 15 Mar 23 @ 8:52 am
bazzio68ControlleristMember since 2018
Dan (djtouchdan) wrote :
user22137884 wrote :
this is the VDJ problem of the stems engine, because when you use serato stems on the same hardware it works great, you don't need a great graphics card. the VDJ stems engine is just not very efficient. bug.


100% false.

Stems 2.0 is leaps ahead of Serato stems.
Serato stems is the equivalent to the first iteration of Virtual DJ stems.

Stems 2.0 does require a high end GPU. It’s also been well documented that the 3050 with 4gb of VRAM is insufficient.

You could change stems to lower quality and then you will see it will work and sound very similar to Serato stems. But it will not sound as crisp and clear as Stems 2.0.


But how do you declare that stems 2.0 works quite ok (50 seconds for a 5 minute track) with a AMD Radeon with low dedicated memory and a RTX3050 with 4GB doesn't (about 4 minutes for a 5 minte track)?
Btw als tried VDJ with no driver installed on the Nvidia and it computed a 5 minute track in about 40 seconds...
It's not as fast as it could (or should be) but i can live with that speed.

 

geposted Wed 15 Mar 23 @ 3:37 pm
AdionPRO InfinityCTOMember since 2006
It looks like NVidia is pretty bad at using system memory, so as soon as it uses more than 4GB memory and starts using system memory it becomes extremely slow.
The Radeon (perhaps because it was designed as integrated graphics) seems to be able to use system memory just fine, so doesn't suffer from this slowness.
 

geposted Wed 15 Mar 23 @ 3:41 pm
bazzio68ControlleristMember since 2018
From what i saw, VDJ is not using all the dedicated memory of the full 4GB it could.
When stretching the Nividia dedicated memory to the full 4 GB. by opening a program that also uses a lot of VRAM, and close it again in it's peak, the Nvidia could suddenly become fast enough (40 seconds for 5 minte track) for computing stems 2.0
 

geposted Wed 15 Mar 23 @ 4:01 pm