Thanks for the update. Yes, that is super slow (and I notice the storage requirements are also near prohibitive [roughly quadrupling space taken by each track?]). Is this the only approach to Stems on devices where processing power is limited? How does Pioneer implement this (if at all)?
Rekordbox does not have precomputed stems. In rekordbox, the stems are stored in memory until the track is ejected from a deck.
Just how fast are you expecting pre-prepared stem extraction to work? Are you comparing it to on-the-fly analysis in VDJ for example (that uses the GPU)?
Traktor’s pre-prep extraction (if you care to Google) apparently takes 30-40 seconds per minute of track.
Also when splitting a track into FOUR separate parts, surely quadrupling the size requirements is expected?
Depends on the source. Only when using an mp3 it gets about 4 times the size.
I use flac and wav, then the stem is about half the size of the source. 
But also (I’m guessing) downgraded in quality from the original file?
Yes by the numbers. However, only when you play 1 to 3 stems. When playing 0 or 4 stems, it will use the source file.
1 Like
IIRC correctly with Serato it’s virtually “on the fly”, meaning I’ve never noticed a particularly lengthy delay for stems to activate. Is this observer bias, or?
If rb does not precompute stems, how does Pioneer equipment tackle the performance problem?
No pioneer hardware does stems, only rekordbox in performance mode.
Serato’s stem analysis is pretty quick (4 seconds per track compared to the 11 seconds per track in Engine).
Not surprised to hear Pioneer is lagging behind Denon as usual, while casually charging 4x the price.
“11 seconds per track in Engine” is a typo surely; as documented above, even on my super fast system, performance is 30 seconds per track.
So if Serato really is pulling off 4 seconds per track, this 8x acceleration over Engine then justifies my gripe on this thread about Denon’s current implementation performance shortfall issues. How is Serato able to do it so much faster?
From my testings so far:
- Macbook Pro 16" M1 Max (24C/32GB)
- ≈18sec to render a 5min track
- ≈24sec to render a 7min track
- Imperceptible difference between 320 CBR and FLAC
Totally fine for me, especially since it’s running the job in the background. I wouldn’t render my entire collection anyway. Instead, use and practice it for your favorite tracks.
The M1 Max on a notebook has a maximum of 10 cores (8 performance 2 efficiency). There’s currently no notebook Mac with more than 16 cores (12 performance maximum).
Just tried Serato here. Dropping tracks into the stems crate (which is the pre-preparing crate) a 3 minute track takes 53 seconds - approx, as there’s no actual timer, you just have to watch for the icon to change colour. So… Denon’s 30 seconds is not “super slow”.
How long does Denon actually take on your system, PK? What are your specs?
Slowest I’ve processed so far is 1288 tracks in 5 hours or 14 seconds per track. That was on my laptop which has a 13900hx, 4060x and 64GB ram. My desktop has been running 8-11 seconds per track. Half of ky collection is 320kbps mp3, and the rest is 44.1kHz 24bit aiff.
I’ve got the same CPU but the mobile H variant.
After checking with Serato, there is a small quality difference between live stems and prepared stems. So I imagine you get a rougher algo for the live stems while the track continues preparing the higher quality stems in the background. Precompured stems are still processing at about 6-8 seconds per track in Serato on my laptop tho.
Would love it if someone could benchmark this on an AMD 7945HX3D CPU.
1 Like
This topic was automatically closed 24 hours after the last reply. New replies are no longer allowed.