IIDEEJAYII
Are there any optimizations on can make, with the Nvidia Control Panel to improve/boost VDJ stems 2.0 performance?
I have only changed the Power management mode for VDJ use.. to "Maximum performance", all others are the default settings.
Asking if there would be benefits from other optimizations?
Don't know which would improve Stems, or which aspects are being used during stems separation.
Screenshots of VDJ specific, settings for GPU
vid capture: https://gyazo.com/c94cd6b76c8f5123c2e0b0f530f42b87
VDJ: v2023 b7356
CPU: AMD Ryzen 7 2700X Eight-Core Processor (3700 MHz)
Memory: 65458 MB
Concurrency: 16
OS Version: Microsoft Windows 10 64-bit (Build 19045.2251)
Graphics Card Vendor: NVIDIA Corporation
Graphics Card: NVIDIA GeForce GTX 1080 Ti/PCIe/SSE2
Graphics Card Memory: 11264 MB
-----------------
Windows Graphics Driver Version: 31.0.15.2698
OpenGL Version: 4.6.0 NVIDIA 526.98
I have only changed the Power management mode for VDJ use.. to "Maximum performance", all others are the default settings.
Asking if there would be benefits from other optimizations?
Don't know which would improve Stems, or which aspects are being used during stems separation.
Screenshots of VDJ specific, settings for GPU
vid capture: https://gyazo.com/c94cd6b76c8f5123c2e0b0f530f42b87
VDJ: v2023 b7356
CPU: AMD Ryzen 7 2700X Eight-Core Processor (3700 MHz)
Memory: 65458 MB
Concurrency: 16
OS Version: Microsoft Windows 10 64-bit (Build 19045.2251)
Graphics Card Vendor: NVIDIA Corporation
Graphics Card: NVIDIA GeForce GTX 1080 Ti/PCIe/SSE2
Graphics Card Memory: 11264 MB
-----------------
Windows Graphics Driver Version: 31.0.15.2698
OpenGL Version: 4.6.0 NVIDIA 526.98
Posted Tue 29 Nov 22 @ 2:14 pm
angelicodin
Not related to this in general, but also reporting here that when computing stems 2.0, my gpu pegs to 100% and stutters the whole system on a RTX 3090. I have to pre compute or I don't bother using them.
Posted Wed 30 Nov 22 @ 3:03 am
Mister Tea
I also get stuttering when computing on the fly, but my 3080 never hits 100 percent.
Posted Wed 30 Nov 22 @ 1:02 pm
IIDEEJAYII
1) While attempting to further get better Stems 2.0, are there any resources available that are known to improve the Stems 2.0 performance?
2) Has the Virtual DJ PC Optimize webpage been updated for Stems 2.0, optimization?
https://www.virtualdj.com/wiki/PC%20Optimize.html
3) Are any of the Serato's PC optimizations, recommended or on the flipside NOT recommended or NOT useful for VDJ Stems 2.0 performance?
https://support.serato.com/hc/en-us/articles/203057850-Optimization-Guide-for-Windows
4) Are there any optimizations that can be done within the NVIDIA control panel to specifically improve VDJ use of the GPU for stems calculations? Or to configure other competing apps from using the resources required for best performance with VDJ, by turning down their access to the GPU?
Note: It has been mentioned in another thread:
Since VDJ is not using the GPU in the traditional way, why would the default settings for GPU use be best? Just wanting to better tune the Video card to possibly optimize VDJ during Stems 2.0 computations.
--------------
and sort of related:
4) Is the VDJ settings Option: videoMaxMemory have units of MB?
So for my graphic card if wanted to use all VRAM, (see specs above) one would enter 11264?
5) And what does the videoMaxMemory, default of "0" use for video max memory?
6) Is there any setting to force other graphic hogging apps to use the "Shared" GPU memory, (32GB in my case), leaving the Dedicated GPU memory available for VDJ?
2) Has the Virtual DJ PC Optimize webpage been updated for Stems 2.0, optimization?
https://www.virtualdj.com/wiki/PC%20Optimize.html
3) Are any of the Serato's PC optimizations, recommended or on the flipside NOT recommended or NOT useful for VDJ Stems 2.0 performance?
https://support.serato.com/hc/en-us/articles/203057850-Optimization-Guide-for-Windows
4) Are there any optimizations that can be done within the NVIDIA control panel to specifically improve VDJ use of the GPU for stems calculations? Or to configure other competing apps from using the resources required for best performance with VDJ, by turning down their access to the GPU?
Note: It has been mentioned in another thread:
PhantomDeejay wrote :
AI does not load your GPU the "traditional" way.
Since VDJ is not using the GPU in the traditional way, why would the default settings for GPU use be best? Just wanting to better tune the Video card to possibly optimize VDJ during Stems 2.0 computations.
--------------
and sort of related:
4) Is the VDJ settings Option: videoMaxMemory have units of MB?
So for my graphic card if wanted to use all VRAM, (see specs above) one would enter 11264?
5) And what does the videoMaxMemory, default of "0" use for video max memory?
6) Is there any setting to force other graphic hogging apps to use the "Shared" GPU memory, (32GB in my case), leaving the Dedicated GPU memory available for VDJ?
Posted Wed 14 Dec 22 @ 10:44 am
Adion
I don't think there are much tweaks needed/available except for making sure to update your video drivers.
I wouldn't recommend changing videoMaxMemory unless you have a specific issue either.
I don't think there's a reason to manually try to change which memory is used either.
As for the serato article, I don't see anything that should typically be needed there either, unless you have a specific problem.
I would definitely *not* recommend to change the minimum processor state. That just risks that your cpu is running faster than needed all the time, which will make it hotter than needed, and could actually prevent it (or the gpu) to run faster when it is needed because the computer is already too hot.
I wouldn't recommend changing videoMaxMemory unless you have a specific issue either.
I don't think there's a reason to manually try to change which memory is used either.
As for the serato article, I don't see anything that should typically be needed there either, unless you have a specific problem.
I would definitely *not* recommend to change the minimum processor state. That just risks that your cpu is running faster than needed all the time, which will make it hotter than needed, and could actually prevent it (or the gpu) to run faster when it is needed because the computer is already too hot.
Posted Wed 14 Dec 22 @ 11:11 am
IIDEEJAYII
Thank you Adion for your always, "on point" solutions :)
I have been tweaking things to get 2 GPU intensive apps to play well together. And in the past I feel I was running out of VRAM. Since after a reboot the VRAM is 1.6GB then when VDJ is on it increases depending on mode. (here is graph created a few weeks ago)
Red Stems OFF
orange stems 1.0 ON
yellow load track to deck stems 1.0 compute <<< No noticeable impact on GPU, which puts in perspective, the GPU load, needed for Stems 2.0 separation
green stems 2.0 turned ON
magenta load track to deck.. 2.0 compute
-------------------
And have at times, another app, that would like 6.5GB of VRAM. and would love to have this app use the "shared" GPU memory, if that is possible.
I have since tamed this apps use of VRAM, a bit .. and now allows VDJ to NOT switch to "Don't use GPU" mode, or to create, Stems engine errors. So all is good, at the moment, and always wishing to improve if possible.
I have been tweaking things to get 2 GPU intensive apps to play well together. And in the past I feel I was running out of VRAM. Since after a reboot the VRAM is 1.6GB then when VDJ is on it increases depending on mode. (here is graph created a few weeks ago)
Red Stems OFF
orange stems 1.0 ON
yellow load track to deck stems 1.0 compute <<< No noticeable impact on GPU, which puts in perspective, the GPU load, needed for Stems 2.0 separation
green stems 2.0 turned ON
magenta load track to deck.. 2.0 compute
-------------------
And have at times, another app, that would like 6.5GB of VRAM. and would love to have this app use the "shared" GPU memory, if that is possible.
I have since tamed this apps use of VRAM, a bit .. and now allows VDJ to NOT switch to "Don't use GPU" mode, or to create, Stems engine errors. So all is good, at the moment, and always wishing to improve if possible.
Posted Wed 14 Dec 22 @ 12:39 pm
theo.d02290
Hey !
VDJ seem to use specific ressources of graphic card
I have an RTX 4090 and computing stems take long time, but little faster as my previous RTX 3070
But analyzing my main library of ~4k tracks take more than 24h...
Taking 10 minutes for 50 tracks is so long
VDJ seem to use specific ressources of graphic card
I have an RTX 4090 and computing stems take long time, but little faster as my previous RTX 3070
But analyzing my main library of ~4k tracks take more than 24h...
Taking 10 minutes for 50 tracks is so long
Posted Thu 26 Jan 23 @ 12:48 pm
IIDEEJAYII
I assume that is a plot of GPU usage (% of max) vs time in seconds.
And if that is true, then it appears that your systems is computing about 8 tracks every 10 seconds, or about 48 songs every min..
(and this even includes the read and then the write to storage, times)
which is amazingly fast, my opinion, especially since tracks cued to be prepared, are a bit slower then then when they are loaded to a deck.
(assume that is done intentionally so that the GPU is not over worked for hours at a time.)
Why is it taking so long to do 4K tracks, at the above rate it should take less then 2 hours.
(if my math is right)
-----------------
1) How much VRAM is being used by VDJ, with drivers for the RTX 4090, during stems 2.0 calculations?
2) Also what is the time to prepare the stems for say a 3 min track when loaded to a deck?
looking at your graph I would guess it maybe about 1 second, right ?
And if that is true, then it appears that your systems is computing about 8 tracks every 10 seconds, or about 48 songs every min..
(and this even includes the read and then the write to storage, times)
which is amazingly fast, my opinion, especially since tracks cued to be prepared, are a bit slower then then when they are loaded to a deck.
(assume that is done intentionally so that the GPU is not over worked for hours at a time.)
Why is it taking so long to do 4K tracks, at the above rate it should take less then 2 hours.
(if my math is right)
-----------------
1) How much VRAM is being used by VDJ, with drivers for the RTX 4090, during stems 2.0 calculations?
2) Also what is the time to prepare the stems for say a 3 min track when loaded to a deck?
looking at your graph I would guess it maybe about 1 second, right ?
Posted Thu 26 Jan 23 @ 1:15 pm
locoDog
IIDEEJAYII wrote :
computing about 8 tracks every 10 seconds,
gpu peaks and troughs don't have to mean a new track, he said 50 tracks in 10 mins, 12 sec per track
Posted Thu 26 Jan 23 @ 2:05 pm
IIDEEJAYII
Yes that is what he said, and was attempting to interpret the graph.
(on my system, each peak is a new track being processed by the GPU, see the graph posted above in this thread, Dec 14th)
Expecting the currently most powerful NVIDIA GPU, the RTX 4090, to outperform my GTX 1080 Ti, which calculates 2.0 stems at 2.8s/min of track. (so about 8.4 sec for a 3 min track, 11.2 sec for 4 min track)
(ie was hoping the RTX 4090 would be 8-10X faster then the 1080 Ti)
And if the the RTX 4090 takes 12 seconds to prepare a single track, then that indeed, is not great for the RTX 4090 GPU.
-----------------------
Would love there to be a thread where times are posted for each of the typical GPUs/VRAM combos, so everyone could have better idea of what to expect from their current or future systems.
(on my system, each peak is a new track being processed by the GPU, see the graph posted above in this thread, Dec 14th)
Expecting the currently most powerful NVIDIA GPU, the RTX 4090, to outperform my GTX 1080 Ti, which calculates 2.0 stems at 2.8s/min of track. (so about 8.4 sec for a 3 min track, 11.2 sec for 4 min track)
(ie was hoping the RTX 4090 would be 8-10X faster then the 1080 Ti)
And if the the RTX 4090 takes 12 seconds to prepare a single track, then that indeed, is not great for the RTX 4090 GPU.
-----------------------
Would love there to be a thread where times are posted for each of the typical GPUs/VRAM combos, so everyone could have better idea of what to expect from their current or future systems.
Posted Thu 26 Jan 23 @ 4:00 pm
Adion
1080 can be up to 20x realtime
2080 27x
3080 48x
4080 62x
At these speeds, it's of course not very useful to save precomputed stems in the first place, but if you want to do it, the time it takes is probably limited by saving the precomputed stems itself, which are encoded using cpu.
2080 27x
3080 48x
4080 62x
At these speeds, it's of course not very useful to save precomputed stems in the first place, but if you want to do it, the time it takes is probably limited by saving the precomputed stems itself, which are encoded using cpu.
Posted Thu 26 Jan 23 @ 4:09 pm
kradcliffe
When I tested that Asus RTX3080 laptop I had it was nowhere near that quick, and even tweaking to full power etc. made no difference. I would have said it was maybe 15x but that's it.
Posted Thu 26 Jan 23 @ 6:49 pm
Adion
Since the same chip is used for desktops and laptops, for laptops the power is usually limited and performance will indeed be a bit less, but I don't have specifics on how much difference this makes
Posted Thu 26 Jan 23 @ 7:18 pm
kradcliffe
Maybe there should be a portal where people can report on the speeds they get with different cards and this can be compiled in to a sheet that users can refer to?
Posted Thu 26 Jan 23 @ 7:22 pm
Adion
Not sure why exact speeds per model would be interesting.
With any 3000 series or above you're in the 20x territory, or less than 15 seconds for the entire song. (and catching up to any position you jump to in the song in 1 second or less)
With any 3000 series or above you're in the 20x territory, or less than 15 seconds for the entire song. (and catching up to any position you jump to in the song in 1 second or less)
Posted Thu 26 Jan 23 @ 7:29 pm
theo.d02290
IIDEEJAYII wrote :
I assume that is a plot of GPU usage (% of max) vs time in seconds.
And if that is true, then it appears that your systems is computing about 8 tracks every 10 seconds, or about 48 songs every min..
(and this even includes the read and then the write to storage, times)
which is amazingly fast, my opinion, especially since tracks cued to be prepared, are a bit slower then then when they are loaded to a deck.
(assume that is done intentionally so that the GPU is not over worked for hours at a time.)
Why is it taking so long to do 4K tracks, at the above rate it should take less then 2 hours.
(if my math is right)
-----------------
1) How much VRAM is being used by VDJ, with drivers for the RTX 4090, during stems 2.0 calculations?
2) Also what is the time to prepare the stems for say a 3 min track when loaded to a deck?
looking at your graph I would guess it maybe about 1 second, right ?
And if that is true, then it appears that your systems is computing about 8 tracks every 10 seconds, or about 48 songs every min..
(and this even includes the read and then the write to storage, times)
which is amazingly fast, my opinion, especially since tracks cued to be prepared, are a bit slower then then when they are loaded to a deck.
(assume that is done intentionally so that the GPU is not over worked for hours at a time.)
Why is it taking so long to do 4K tracks, at the above rate it should take less then 2 hours.
(if my math is right)
-----------------
1) How much VRAM is being used by VDJ, with drivers for the RTX 4090, during stems 2.0 calculations?
2) Also what is the time to prepare the stems for say a 3 min track when loaded to a deck?
looking at your graph I would guess it maybe about 1 second, right ?
Today, when start VDJ
1.3 Go of 39.9 Go of memory processor GPU
1.2 Go of 24 Go of dedicated memory
0.1 Go on shared GPU memory
Just after start VDJ and start processing stems 2.0
6.1 Go of 39.9 Go of memory processor GPU
5.4 Go of 24 Go of dedicated memory
0.7 Go on shared GPU memory
With variant use of 10-22% of CPU
With the latest version of VDJ (v2023 b7443) average track time is 3:30 and make 9-10sec per track in general
Posted Fri 10 Feb 23 @ 6:10 pm
theo.d02290
Quote :
With variant use of 10-22% of CPU
With variant use of 10-22% of CPU
Sorry not CPU but GPU
Additional information :
I have 2 PC (1 desktop with RTX 4090 and the another one the laptop for VDJ for the event/dj)
This 2 PC have a sync for music, stems and visual scripts (DMX)
Laptop is cool and performant in general (i7 7700HQ with an GTX1070) but stems is SO LONG and script my visual is long but can do the job on laptop.
This why I set up sync between those PCs.
Desktop will perform the hard job for all 2.0 stems and visual. This is sent directly to the cloud and laptop have just to download to be synced with Desktop
Posted Sat 11 Feb 23 @ 3:40 am
as
Just in case somebody wants to know -> RTX 4070 TI Speed
OS: Win 11 Enterprise latest patch level
VirtualDJ: Early Access Build 7474
- Core i9-13900
- Z790 Mainboard
- 2 x 16 GB 6000Mhz RAM
- 2 TB PCIe4 NVME SSD
- ASUS ROG STRIX RTX 4070 TI OC
Song:
Robbie Williams - Angels / Greatest Hits / 320Kbit MP3 / L: 4m32s
roundabout 5 Sec -> 55x
Great!
OS: Win 11 Enterprise latest patch level
VirtualDJ: Early Access Build 7474
- Core i9-13900
- Z790 Mainboard
- 2 x 16 GB 6000Mhz RAM
- 2 TB PCIe4 NVME SSD
- ASUS ROG STRIX RTX 4070 TI OC
Song:
Robbie Williams - Angels / Greatest Hits / 320Kbit MP3 / L: 4m32s
roundabout 5 Sec -> 55x
Great!
Posted Wed 01 Mar 23 @ 5:46 am
IIDEEJAYII
Thanks for posting your 2.0 stems speed.
That is almost 1 sec per min of the track .. awesome.
and assume GPU has 12 GB of VRAM.
That is almost 1 sec per min of the track .. awesome.
and assume GPU has 12 GB of VRAM.
Posted Wed 01 Mar 23 @ 7:37 am
bazzio68
I have a RTX3050ti and tried all settings the Nvidia 3D config but none makes it quicker, a 5 minute songs take about 4 to 5 minutes compute the 2.0 stems. If i use only CPU it is quicker, about 2-3 minutes to compute.
I have the 7474 version wich should be using a combo of CPU and GPU? (hence the "Allow partial GPU acceleration with RTX3000 with 4GB memory") but it doesn't do that, it uses standard all GPU. To bad it doesn't use the memory available to the max (2,9 dedicated of 4 GB and 1,6 of 7,6 shared)
Btw forcing VDJ to use Nvidia card via the Nvidiaa config doesn't make a difference in performance.
All in all pretty disappointed with spending €117 for a licence and not able to use stems 2.0 the way it should be. Imo an AMD 7 Ryzen gaming laptop with 16G memory and RTX3050ti should be able to work fast with VDJ,
I have the 7474 version wich should be using a combo of CPU and GPU? (hence the "Allow partial GPU acceleration with RTX3000 with 4GB memory") but it doesn't do that, it uses standard all GPU. To bad it doesn't use the memory available to the max (2,9 dedicated of 4 GB and 1,6 of 7,6 shared)
Btw forcing VDJ to use Nvidia card via the Nvidiaa config doesn't make a difference in performance.
All in all pretty disappointed with spending €117 for a licence and not able to use stems 2.0 the way it should be. Imo an AMD 7 Ryzen gaming laptop with 16G memory and RTX3050ti should be able to work fast with VDJ,
Posted Wed 01 Mar 23 @ 12:00 pm
