Linux: Testing VST CPU performance with Renoise...

I´ve installed OBXD 64bit vst in my Xubuntu Kxstudio Linux yesterday. I know that it is very cpu-hungry VSTi (I used it on Windows) but on the other hand, absolutelly great and free.
Somebody made a linux version cause it is build with Juce and code was published. So maybe it wanna more resource than on windows machine because of it. I dont know.

I have 64bit linux, 64bit Renoise and 64bit version of plugin.
With windows i was using 32bit Renoise and 32bit VSTi but system was 64bit (and one obxd took about 35% cpu).
I dont wanna use 32bit apps with my Linux because i dont wanna install these 32bit dependencies.
Is 64bit Renoise more hungry?

Because when i opened OBXD, recorded some notes and pasted convolver it took about 60% cpu in Renoise and 96% cpu was showed by jack meter (i know that metering in Renoise works somehow different). And cracing and xruns were here.

Because i though (because of renoise meter) that it isnt cpu problem i´ve tried change every setting. When i changed 2cpu to 1cpu crackling stopped! Why? I have AMD Turion II Dual-Core P560 so using 2 cores should be ok.

But selecting 1cpu didnt solve this problem forever, crackles appears when i something little more cpu demanding. Change from 2cpu to 1 just lowers resources about 5% maybe.

So my questions are basically three:

  1. Is renoise 64bit more resource hungry and should i use 32bit? I have 4gb ram and i wont reinstall my system.
  2. Should i just delete OBXD cause it has maybe too much cpu demanding port or try 32bit version with renoise (but sandboxed plugins are more hungry, not?)
  3. Is there any another test i should take? Is Renoise under linux more cpu hungry?

When i use Bitwig Ive never got any Xruns and there can be about ten reverbs and another cpu hungry dsps. So i dont think it relate to realtime or other setting in Jack server.
I am on b8 still, didnt downloaded RC1 yet.

Thank you people. I thought that with Linux there will be better resource managing…

(If you wanna try that OBXD linux port and try it - )

You should’ve tested the OBXD plugin in bitwig (or some other DAW) using jack as output to get a fair comparison.

for the 1 vs 2cpu, IIRC there has been some changes in linux kernel during last year that should improve load balancing using multiple cpus. I’m not sure if they were merged already in 3.11, which is the version in *ubuntu 13.10, but newer version could still improve the situation. I had to use only 1cpu myself before, but I remember that at one point the situation improved significantly, most likely due to kernel upgrade.

Most Linux dist’s CPU energy saving settings are pretty harsh, and not really useful for real-time audio application. Changing the cpu govenor from “ondemand” to “performance” can make a really huge difference.

This can be changed with cpufrequtils. Unfortunately it’s not very easy to change that.
Try googling for “cpu governor performance YOUR_LINUX_DISTRIBUTION” to see how to set cpu frequency to “performance”.
xfce4-cpufreq-plugin may do the job on Xubuntu.

You are right. I dont use cpu freq but pm utils. I thought that it will relate. Problem is that with performance mod my laptop is can hear fan very well (it was not problem with windows). I will test obxd with other host also but!
I can have ten synths and ten reverbs in bitwig (internal) and it takes about half of cpu. With renoise my cpu is full with one sample and four convolvers.

Thank you.
Will try to do a lot of test and report them before final version :)

What going on here? Performance on linux is bad, much worse than with w7 where i had also feeling that R3 is less fluent than R2.8
Regardless on cpu setting (and system cpu governor)…

bitwig: 8 instances of OBXD 64 bit _ every pattern has 4 notes _ cracking is here_ 7 instances are absolutely ok
renoise: 2 instances of OBXD 64 bit with 4 notes causes a lot of crackles _ once instance shows about 40%

On linux too and performance is great. I use mate 1.6 mint 15 64bit. Cpufreq on performance. I5 3ghz 8gb ram geforce 8600gt 1gb sb audigy

Can you compare renoise with some vsti instances
other linux host (maybe the bitwig)?

I only use whysynth sometimes. Mostly samples and internal dsps

So i recommend that obxd

I’ve just been playing with this synth and the highest I could make it go was about 34% for one instance (renoise meter). Added 4 instances, all playing 4 notes and got about 60% but no crackles at all.
(Linux 64 bit rc1)

My concerns are mainly about Renoise itself. One native instrument i create, about three samples, some filters and reverb takes about 20% if i use chords.
When i insert 7x drawbar organ and everyone plays 4 notes it takes about 50% (one take about 20%, cpu use isnt going up lineary…) and when i am on 50% cracking always come in and xruns are here. When i am on 50 or more procent in bitwig, jack behaves normally.
I would assume that cracking should be here with 90% of cpu for example.

I’m not sure if bitwig is calculating the audio only process here or full cpu utilisation so this might be comparing apples to oranges, the other thing is the audiocard latency bitrate and and interpolation being used. Renoise is processing 32-bit audio. What bitrate does Bitwig performs its processing on?

Vv: With Bitwig i can use 7 instances of OBXD. With Renoise i cant use 2 instances. It is more important than cpu meter.
I am going to make these tests tonight:

  1. OBXD with R3.0rc2 (64bit vs 32bit) vs R2.8 (64bit vs 32bit) vs Bitwig
  2. some different system heavy vsti and same test (maybe Aspect vsti will be enough, will see…)
  3. Make R3.0 “hardcore multiband song” (so hardcore that crackles will be there) and open it with R2.8 (64bit vs 32bit again) then and see if there is some difference with cpu uses and xruns.

Bitwig has also 32bit audio engine if we talk about this.
Audiocard latency is same with every aplication so when ardour, tracktion and bitwig works better than Renoise with VSTi plugin (i´ve didnt do tests with all these apps yet) it is obvious where problem is.

Okay i’m curoius to the result Renoise 3.0 vs 2.8
A side note:Renoise does do 64-bit processing in the DSP areas.It might perhaps also does that in the DSP chains in the instrument structure.

Will do.
But for now. Just inserting Movie Dodgy Synth instrument and playing random chords will show 50% of cpu and there is a lot of xruns and crackling (2 realtime cpus).
When i change setting to one in realtime cpu it is much harder to have xruns (i would say twice :-D) and if i have it, dsp meter in renoise shows about 90% not 50%. So it is better setting for me, dunno why.

Selecting ALSA as a driver is maybe even a worse than a jack.

Maybe Renoise isnt friend with some “cpu governors” but other DAWS work fine so i am not so sure if it relates.

Changed name of topic.
I cant run renoise 2.8 32bit cause i have 64bit xubuntu
so i tested R2.8 64bit.
Performance is almost same (xruns are here with 3 obxd vs 2 obxd in R3 - but it is still not 6 like in bitwig.
(2cpu shows 50% where 1cpu shows 100% - 1 cpu is much worse with r2.8, cant say this with R3 where it feels sometimes better)

BUT with Ardour performance is even worse than Renoise! So maybe every software compensate “cpu governors” differently and Bitwig is just the king in this…
Is it possible?

Cpu governors only affect how cpu frequency scaling is handled. There’s no way one application could ‘compensate’ it better (unless something like ‘cpu overload prevention’ was horribly buggy maybe)

ondemand governor which is default for a lot of processors saves power by ramping cpu clock speed up when needed and keeping it down when cpu is idle. Unfortunately it adds nasty latency peak when clock speed is ramped up/down, which is why it’s not good for audio work.

performance governor keeps your cpu at full speed so you should have best performance with it, and no cpu ramping latency issues.

At least newer intel processor handle frequency scaling themselves in cpu, and should use powersave governor by default so the latency issue don’t exist there.

I should test obxd some time if I get it to build and see how it performs on my system.