登录:     


Forum: General Discussion

Topic: New mini pc for stems 2.0
Good morning,
I would like to change my current notebook (11th Gen Intel(R) Core(TM) i7-1165G7 @ 2.80GHz) and get a mini PC.
I'm looking at solutions from Geekom and Minisforum and Beelink that have processors with integrated GPU.
I saw the ryzen 7 8845HS with 780M GPU or even better (even if it costs more) the ryzen Ai 9 370HX cpn 890M GPU.
Can these machines process stems 2.0 in real time?
The Atomix specifications are not updated to the new GPUs and I would like to understand how to do a calculation to estimate the quality of a GPU.
Is there a calculation that can be done? Or take into account some benchmark?
I know that Mac mini M4 would be perfect, but I wanted to stay in the Windows world.
Thanks
 

发表时间 3 days ago @ 6:52 am
According to our stats, it looks like the 780M should be able to do stems around 8x so that should be usable.
We don't have any personal experience with these cpu's though, and I think it's only a few users currently using them so you'd have to see if you can try to be sure.
 

The Hx370 on board graphics does stems real-time around 10.5x
 

Adion wrote :
According to our stats, it looks like the 780M should be able to do stems around 8x so that should be usable.
We don't have any personal experience with these cpu's though, and I think it's only a few users currently using them so you'd have to see if you can try to be sure.

I hope to find some users who use it
 

kradcliffe wrote :
The Hx370 on board graphics does stems real-time around 10.5x


10.5 X means that on a 3 minute song it takes 18 seconds to calculate it? If so it should work, but what is the minimum multiplication value? and how is it calculated?
In these forums there is no one from Atomix who responds with precise data?
 

Yes basically dividing the song length by 10.5x

The software calculates the speed and displays it in the CPU tool tip.

BTW Adion is the lead developer.
 

You nee more than 1x for real time use.
Stems are also calculated in blocks of 12 seconds, so 1x would mean you have to wait 12 seconds after loading the song before you can use stems.
At 4x it would be only 3 seconds, so in most cases that would probably be fine
 

[quoteBTW Adion is the lead developer. [/quote]
here I didn't know, thanks

 

Adion wrote :
You nee more than 1x for real time use.
Stems are also calculated in blocks of 12 seconds, so 1x would mean you have to wait 12 seconds after loading the song before you can use stems.
At 4x it would be only 3 seconds, so in most cases that would probably be fine


on my laptop there is also an nvidia geforce MX330 and virtual tells me that I can't use glòi stems 2.0, so it doesn't reach 1.0x?

another thing is that in some time the 3.0 version of the stems will be released which requires much more power because I would be back to square one :)
 

Instead, what multiplier does Apple's M4 have?
 

 

For reference, I have a Geekom Mini IT13 which has an Intel i9-13900H with Iris Xe graphics. The GPU processes stems at around 1.7x speed. I don't use real time stems so couldn't tell you how well it would work.
 

DJPhallus wrote :
For reference, I have a Geekom Mini IT13 which has an Intel i9-13900H with Iris Xe graphics. The GPU processes stems at around 1.7x speed. I don't use real time stems so couldn't tell you how well it would work.

thanks for the feedback, hopefully someone with the Ryzen processor I indicated exists :)
 


Quote :

on my laptop there is also an nvidia geforce MX330 and virtual tells me that I can't use glòi stems 2.0, so it doesn't reach 1.0x?


I did a test with the MX330 and between the time to convert it and save it it took 2 minutes and 20 seconds for a 3 minute and 3 second song, so that means 1.37X if I did the right calculations
 

I use LEGION GO GPU-780M and it is fully usable. The GPU acceleration is 7.7 -9 times, and there will be no problem. However, for real-time STEMS2.0, the GPU will reach 100%, which will increase the DPC delay to a certain extent. However, in actual use, as long as you play music normally and don't scratch it, there will be no problem, because if it is scratched, ASIO may cause the sound quality to tear. I suggest that you pre-process the commonly used music with STEMS 2.0 first, and set it to On-demand in the settings. I have used this computer for 1 year and there have been no problems in the performances. However, if you want to use this kind of MINI PC, you must do deep optimization to reduce the DPC delay to use ASIO devices more perfectly. Another friend of mine uses a MINI PC with HX370, but he uses SERATO. At the beginning, he only did simple optimization, but there were some minor lags and sound problems. So he did the same deep DPC delay optimization as me, and used it for nearly 2 months without any problems. I didn’t feel that his 890M was faster than my 780M. When analyzing STEMS2.0, the GPU was also 100%. But the same thing is that only VDJ was installed in the computer, and other system services or other things not related to VDJ were disabled or deleted. I couldn’t open STEMS with SURFACE before, and if I opened it, it would overheat and shut down. Now AMD’s CPU+GPU has solved my troubles!
This is the photo of the MINI PC when I first got it!
 

Works flawless with Nvidia
 

tonyshan007 wrote :
I use LEGION GO GPU-780M and it is fully usable. The GPU acceleration is 7.7 -9 times, and there will be no problem. However, for real-time STEMS2.0, the GPU will reach 100%, which will increase the DPC delay to a certain extent. However, in actual use, as long as you play music normally and don't scratch it, there will be no problem, because if it is scratched, ASIO may cause the sound quality to tear. I suggest that you pre-process the commonly used music with STEMS 2.0 first, and set it to On-demand in the settings. I have used this computer for 1 year and there have been no problems in the performances. However, if you want to use this kind of MINI PC, you must do deep optimization to reduce the DPC delay to use ASIO devices more perfectly. Another friend of mine uses a MINI PC with HX370, but he uses SERATO. At the beginning, he only did simple optimization, but there were some minor lags and sound problems. So he did the same deep DPC delay optimization as me, and used it for nearly 2 months without any problems. I didn’t feel that his 890M was faster than my 780M. When analyzing STEMS2.0, the GPU was also 100%. But the same thing is that only VDJ was installed in the computer, and other system services or other things not related to VDJ were disabled or deleted. I couldn’t open STEMS with SURFACE before, and if I opened it, it would overheat and shut down. Now AMD’s CPU+GPU has solved my troubles!
This is the photo of the MINI PC when I first got it!

great news, I don't scratch so I shouldn't have latency problems, it seems strange to me that the hx370 (so 890M) doesn't give a bit of an advantage over the 780M. However, it's already good that you can do it with the 780M. I imagine that 100% of the GPU is until it has finished decoding the stems. Do you have any idea if the amount of RAM dedicated to the GPU and the type of speed can affect it?
By the way, I didn't know that you could see the calculation of the stems and so I tried on my current notebook and it says 1.2X
 

DJ Dash El Mas Guapo wrote :
Works flawless with Nvidia


what processor do you use?
 

I own the Beelink with the 780m 8845HS combo. It works pretty well with STEMS 2.0. I'd estimate a standard 4 min song takes about 10 seconds to fully separate which is on par with my laptop i7 RTX 2060.

Aside from that, the Beelink was a decent price, is dead silent and has been stable and trouble free so far. I'm very happy with mine.
 

schmidi_0 wrote :
I own the Beelink with the 780m 8845HS combo. It works pretty well with STEMS 2.0. I'd estimate a standard 4 min song takes about 10 seconds to fully separate which is on par with my laptop i7 RTX 2060.

Aside from that, the Beelink was a decent price, is dead silent and has been stable and trouble free so far. I'm very happy with mine.

thanks, what model do you have? why do you say "it works quite well"? is there something that you're not convinced by?