How does CPU utilization and RAM usage compare with Kontakt?

Official support for: tx16wx.com
RELATED
PRODUCTS

Post

Yeah it's realtime, you get loop start and loop length as modulation destinations for sampler mode. In DFD mode it's not available, though.

Post

Right, but that is what I meant. It is near impossible (and impossible to do efficiently) to have sample accurate loop modulation in DFD.
Btw, I tweaked sinc algorithm a bit. I think I managed to get the compiler to unroll a few more loops, and I might have shaved a few percent off high quality more cpu usage. Not a game changer, but every bit helps etc. I'll do a few more experiments. But the balance between features and performance is tricky.
TX16Wx Software Sampler:
http://www.tx16wx.com/

Post

elcallio wrote: Wed Nov 13, 2019 6:38 amBut the balance between features and performance is tricky.
That is true!


BTW I should have also mentioned - the numbers I got from task manager were with plugin GUIs closed.

Post

Thanks ED and Calle for all the insights! I´ve already sent to Calle a test project with the Salamander Piano, both in Tx16wx and Kontakt (as ED says, the Free Player may be useful to test things out). I hope TX16´s performance can be improved, since I´m very interested in it.

Post

So, warning kids, because the rant is due:
I looked at the salamander, and to be honest, this is the kind of sample instrument I _really_ hate. The samples are loooong, and contain at least half only silence. Not only that, but each note-off triggers _three_ additional voices, one which is a barely audible "churg" sound to simulate the key release noise, and two who are even less audible to simulate "string resonance" (all which could at the very least just have been baked into a single sample).
The envelope falloff is of course set to several seconds, ensuring we play loads and loads of silence...
So it is not super strange that doing a heavy arpeggio you pretty quickly hit the poly limit (4 voices for each key played, each release-playing nearly the length of the song (sic)).

Honestly, this is what I would call a _very_ lazy sample instrument, and just a sign that giving powerful computers to people makes for wasteful usage. Honestly, I don't even think this sounds better than my old AKAI pianos from the nineties.

All that anger being vented, two good things came out of this though: I did find a bug in the age counter which might cause voice stealing to produce less desirable results. With the fix applied, I can play back the example file at ~40% core CPU load (i.e. 40/8 total in my case), and it sounds pretty much as it should, since now only the silence is clipped. In fact, playing this, I would calculate the max number of voices I intend to play (I would guess max 10), multiply by 8, and set a slot poly limit to this to encourage killing off useless trailing silences as much as possible.

Wait with that until the fix for age counter and slot poly stealing is in though, should be max Friday.

I've also, as stated before tweaked the high quality interpolator some. There are some other low hanging fruit that might be able to shave a few percent here and there, but those will come a little later I think.

As for multi-core rendering, I am sceptical, mainly because it violates the standard, and there is additional costs involved that would hurt more sane loads. If one needs a lot of voices spread over more cores, putting a separate TX instance on a different track is a more DAW-friendly way to process in parallel.

Now, let me finish my rant by saying that all annoyance aside, this was a great provocateur to find some very subtle issues. So thanks for that.
TX16Wx Software Sampler:
http://www.tx16wx.com/

Post

Thanks for the detailed reply, Calle!

I´ll test the fixes for sure. Personally, I don´t want this to sound like some kind of competition against Kontakt. In fact, if you read again my first post in this thread, I compared TX´s performance against Sforzando, which is free and loads Sfz as well. Its performance is much, much lower than Tx´s one, and similar to Kontakt, so there must be something else that is making the difference. If you don´t like the Salamander Piano, you can take Piano In 162 for the tests - you won´t tell me that it also sounds as a 90´s Akai sampler, do you? :) - . I´m more intrigued by Sforzando´s performance than Kontakt´s, really. After all, I already have good commercial Kontakt pianos and even have these freebies for it. I joined this thread because it referred to TX´s performance, but I probably should create another topic for the direct comparison with Plogue´s Sfz player. The fact is that I´m interested in TX as an all-round sampler for various reasons, and I´d like to contribute to its improvement when I see something that is not quite right, specially when compared to a freebie like Sforzando. For example, playing live with a notebook is a very common situation for many of us, and probably, right now, TX could not be a completely viable tool for the task (haven´t tested it yet for that use-case, but I´ll do it at any moment).

I also see some problems when loading Sfz files that load flawlessly in Sforzando, but that will certainly require a thread of its own.

Well, whatever you need to further test things out, just tell me.

Greets, and thanks again.

Post

Sforzandois a dedicated SFZ player, so it is really comparing apples and oranges. Because it has neither real editing nor general purpose sampling properties, it can quite easily streamline certain aspects. For eaxmple, it has some very nice handling and visualization of SFZ key switching.
SFZ in TX is an import option, it is not a first-class citizen. I'm happy to try to improve it, but it is never gonna be as good as hand-made programs since some of the core engine properties are different.
Sforzando is great if you want to play existing SFZ:s. TX is for when you also want to edit stuff, or even more, build your own things.
But again, I welcome suggestion on improvements to both engine and importers.
TX16Wx Software Sampler:
http://www.tx16wx.com/

Post

New build is up. CPU usage should be slightly lower now, and more importantly, voice stealing should be a lot more pleasing to the ears.
TX16Wx Software Sampler:
http://www.tx16wx.com/

Post

Thanks so much for the quick improvement, Calle! From tomorrow on I´ll use it and will give you some feedback next week. I´m beginning to use tx16wx on a notebook too, so I have another environment to test things out.

Regarding the importers, I´ll probably create a specific thread for the problems I meet with Sfz.

Have a good weekend, and thanks again for your commitment.

Post

FYI, since it became an issue (I am looking at you ED): The latest beta build of TX Pro includes multi core processing. Take it for a spin.
TX16Wx Software Sampler:
http://www.tx16wx.com/

Post

Multi-core processing in an afternoon? Tipping my hat to you, sir. :)

Post Reply

Return to “CWITEC”