View Issue Details

Jump to NotesJump to History
IDProjectCategoryView StatusDate SubmittedLast Update
0003650ardourbugspublic2011-01-03 14:442012-01-12 22:53
Reporterpaul Assigned To 
PrioritynormalSeveritymajorReproducibilityhave not tried
Status newResolutionopen 
Product Version3.0-beta1 
Target Version3.0 
Summary0003650: DSP usage is double compared to Ardour2
Descriptionbased on currently unconfirmed reports from Dave Phillips.
TagsNo tags attached.

Relationships

related to 0003833 closedsteph5150 DSP gets high with little number of audio/midi tracks (causing lots of xruns) in Ardour3.0 svn 9084 

Activities

2011-01-03 15:42

 

ardour-comparison.txt (932 bytes)    
I ran identical sessions on A2 and A3. The sessions included two tracks created with imported audio files. One plugin - the CAPS 2x2 Plate Reverb - was used on the Master track in each session. The differences are notable:

A3 produced xruns reliably on opening and during operation, including playback. With the test file up to 25 major xruns are typical. The DSP usage idles at ~20%, going to ~30% on playback, with at least one spike to 58%.

A2 produced 0 xruns. DSP usage idled at ~8%, moving to ~10% on playback. No spikes.

Ardour2, 2.8.11, rev 8197
Ardour3, 3.0pre0 rev 8413

JACK 0.120, home-built
  realtime
  default priority
  frames 128
  sr 44100
  periods 2
  timeout 5000
  mididriver seq

Ubuntu 10.04, 32-bit
kernel: 2.6.31-11-rt #154-Ubuntu SMP PREEMPT RT

AMD64 3800+, single-core 2.4GHz CPU
4G memory

M-Audio Delta 66 audio interface

I will furnish the session files on request.

Dave Phillips
3 January 2011
ardour-comparison.txt (932 bytes)   

paul

2011-01-03 16:41

administrator   ~0009800

I ran identical sessions on A2 and A3. The sessions included two tracks created with imported audio files. One plugin - the CAPS 2x2 Plate Reverb - was used on the Master track in each session. The differences are notable:

A3 produced xruns reliably on opening and during operation, including playback. With the test file up to 25 major xruns are typical. The DSP usage idles at 0000023:0000020%, going to 0000010:0000030% on playback, with at least one spike to 58%.

A2 produced 0 xruns. DSP usage idled at 0000014:0000008%, moving to 0000009:0000010% on playback. No spikes.

Ardour2, 2.8.11, rev 8197
Ardour3, 3.0pre0 rev 8413

JACK 0.120, home-built
  realtime
  default priority
  frames 128
  sr 44100
  periods 2
  timeout 5000
  mididriver seq

Ubuntu 10.04, 32-bit
kernel: 2.6.31-11-rt 0000154-Ubuntu SMP PREEMPT RT

AMD64 3800+, single-core 2.4GHz CPU
4G memory

M-Audio Delta 66 audio interface

I will furnish the session files on request.

Dave Phillips
3 January 2011

cth103

2011-01-04 17:11

administrator   ~0009820

Broadly similar results here; I have got 8 of the reverbs on the master bus rather than 1 to get the DSP loads out of the noise a bit, and I'm seeing

A2: DSP 10% idle, 12% play
A3: DSP 15% idle, 20% play

No xruns in either case. This is with jack2, 256 sample buffer, 2 periods on a Core 2 Duo E4600 (2.4GHz)

joegiampaoli

2011-01-09 00:03

reporter   ~0009887

Same problems here with xruns.

With the same jack settings I use for Ardour2, Ardour3 fires up quite some xruns (between 20 -30) when launching it.

Sometimes mouse clicking or changing things will fire an xrun or two once in a while, not so bad but compared to A2 which never happens, but kind of makes me feel like something not working as it should.

When using Ardour2 I get absolutely no xruns unless I open up something quite heavy like a calf plugin with moving gui (like multichorus or phaser) but that's just about it.

MIDI is what mostly gives me a lot of them in A3. Mostly using the step editing with the piano window or sending messages from midi keyboard to record them to midi track, almost every keystroke will give me an xrun. In A2 I use seq24 without issues.

Also sending MMC command to control the transport gives me xruns once in a while, especially when engaging the Main REC button and hitting play to start tracking gives me about 5 of them.

My basic qjackctl settings:

  realtime
  priority 89
  frames 128
  sr 48000
  periods 3
  timeout 5000
  mididriver seq

Hardware:
Fast Track Pro (USB1)
1 GB RAM
AMD Athlon 64 bit 2.0 GHz single core

Software:
Ubuntu Karmic with self compiled rt kernel (32 bit)
Ardour 2 and 3 from SVN
JACK2 SVN 4096

Thanks

joegiampaoli

2011-01-09 01:16

reporter   ~0009888

Just compiled 8483 and performance is much better, not getting so many xruns.

My imagination or something has been improved?

Thanks

realhangman

2011-01-09 22:52

reporter   ~0009892

Unfortunately no improvement here with 8488. A3 still consumes about 2 times as much DSP as A2. But A3 looks much nicer now ;)

cth103

2011-01-10 17:31

administrator   ~0009905

I'd be interested to know if things are any better in rev. 8492.

joegiampaoli

2011-01-10 22:19

reporter   ~0009910

Last edited: 2011-01-10 22:20

 8493 here, things are pretty stable ATM (no xruns).

realhangman

2011-01-10 23:05

reporter   ~0009913

Hey, definitely better here with 8494 than with 8488! I will have to measure the difference to A2, but A3 still consumes more DSP power. But I guess it should be even much less with 4 cores in A3 (configure "All but one cores") ?

realhangman

2011-01-31 15:37

reporter   ~0010005

Hi, I've just tried rev. 8622 and it is getting better and better! A test session with 50 of LinuxDSP's plugins (EQ/Comp/Gate) is using 25% DSP in Ardour 2.8.11 and 30% in Ardour3 rev 8622 on my AMD Phenom 4x2,5GHz CPU (Option: All but one processor in A3).

realhangman

2011-07-15 13:46

reporter   ~0011135

Update for rev 9881 with more specific setup:
AMD Phenom 4x2,5GHz
Jack 1.9.7 running @ 128 samples buffer size
Session with 16 Mono busses, each with LinuxDSP Pro-Gate, Pro-Dyn & Pro-EQ
Master with LinuxDSP GR-EQ2
= 49 Plugins

A2: 24% DSP usage

A3 with 2 processors for DSP: 26% DSP
A3 with 3 processors for DSP: 21% DSP
A3 with 4 processors for DSP: 17% DSP

Not sure whether this is what it should be, but I guess it's pretty close :)

iliverez

2011-08-30 11:07

reporter   ~0011426

Last edited: 2011-08-31 06:58

 DSP usage on A3-alpha 10 is increased when adding plugins, even if the relative tracks are muted, or playback is stopped. This appeared with Calf reverb and an equalizer plugin ticked (both LV2), even if playback or recording is stopped. So, idle DSP usage with no plugins is 0000033:0000015%, with one stereo track with these two plugins 0000025:0000028%, and with a second track with another two plugins 0000010:0000040-45%. When playing back/recording this may go up to 65%. Sometimes, after stopping playback or recording, usage remains up to 65% or even more to 75-80%.

I managed to record a couple of tracks without using plugins with no xruns, but after inserting the plugins, I got about 2-5 xruns for each 4 to 6 minutes recording session. Will try more recording without plugins in the session to check if this works consistently without xruns.

DSP usage was also randomly high (0000071:0000060-70%) when doing some midi recording with calf organ. The usage remained high even after recording or playback was finished.

System specs:
amd athlon x2 5200+, 5gb ram
fedora fc15 with linux-3.0.1-rt10 (custom build)
jack2 1.9.7-2 (fedora) (rtprio 89), 2 periods, 128 frames, 44100
RME HDSP 9632 (rtprio 92)
also tested sata irqs rtprio from 10 to 70 with no difference

UPDATE: Today I tried the same setup with A2, where DSP usage reached a maximum of 15% when recording or idle, with no real difference between recording and idle state

paul

2011-11-15 00:07

administrator   ~0012023

anyone providing any information for this bug MUST include the "DSP CPU Utilization" setting from the Misc tab of Edit->Preferences. they should also specify the output of the following command:

  grep 'model name' /proc/cpuinfo

(for linux; on OS X use the system profile to get information about CPU configuration and include it). you need to also specify whether you are using any denormal protection (Edit->Preferences->Audio).

in addition, i also intend to mostly ignore reports that do not include side-by-side tests of identically sized sessions with NO plugins. that doesn't mean that testing with plugins is a bad idea, but it introduces many, many new variables. we need to see numbers without plugins as well.

realhangman

2012-01-12 22:53

reporter   ~0012549

Update for A3 rev 11219:

$ grep 'model name' /proc/cpuinfo
model name : AMD Phenom(tm) 9850 Quad-Core Processor
model name : AMD Phenom(tm) 9850 Quad-Core Processor
model name : AMD Phenom(tm) 9850 Quad-Core Processor
model name : AMD Phenom(tm) 9850 Quad-Core Processor

Jack 1.9.8 running @ 128 samples buffer size
Session with 16 Mono busses, each with LinuxDSP Pro-Gate, Pro-Dyn & Pro-EQ
Master with LinuxDSP GR-EQ2
= 49 Plugins

Ardour 2.8.12 (10144), no denormal protection:
24% DSP usage with plugins, 2,5% without (plugins deleted from the session)

A3 rev 11219 with "no processor handling" and "Use DC bias" not ticked:
With plugins:

"all available processors" - 17%
"all but one" - 21%
"2 processors" - 27%
"1 processor" - 47%

Without plugins:

"all available processors" - 4,5%
"all but one" - 4,7%
"2 processors" - 4,7%
"1 processor" - 5,5%


Hope this helps!

Issue History

Date Modified Username Field Change
2011-01-03 14:44 paul New Issue
2011-01-03 14:44 paul cost => 0.00
2011-01-03 15:42 Musaeus File Added: ardour-comparison.txt
2011-01-03 16:41 paul Note Added: 0009800
2011-01-04 01:14 cth103 Target Version => 3.0-alpha1
2011-01-04 17:11 cth103 Note Added: 0009820
2011-01-09 00:03 joegiampaoli Note Added: 0009887
2011-01-09 01:16 joegiampaoli Note Added: 0009888
2011-01-09 22:52 realhangman Note Added: 0009892
2011-01-10 17:31 cth103 Note Added: 0009905
2011-01-10 22:19 joegiampaoli Note Added: 0009910
2011-01-10 22:20 joegiampaoli Note Edited: 0009910
2011-01-10 23:05 realhangman Note Added: 0009913
2011-01-31 15:37 realhangman Note Added: 0010005
2011-03-07 00:51 cth103 Relationship added related to 0003833
2011-07-14 14:59 cth103 Target Version 3.0-alpha => 3.0-beta1
2011-07-15 13:46 realhangman Note Added: 0011135
2011-08-30 11:07 iliverez Note Added: 0011426
2011-08-31 06:58 iliverez Note Edited: 0011426
2011-11-15 00:00 cth103 Target Version 3.0-beta1 => 3.0
2011-11-15 00:07 paul Note Added: 0012023
2012-01-12 22:53 realhangman Note Added: 0012549