PDA

View Full Version : Excessive CPU usage with AutoTune (FIXED)


teej813
11-25-2006, 09:11 AM
Greetings! Recent Reaper convert here.

I've been monitoring the progress of Reaper for about a month now and have been very impressed with its progress. Purchased a license a few days ago and have started some serious testing.

Reaper has done most everything i've asked it to, but i ran into a problem this morning with a plug i've used successfully in the past with my previous DAW software (n-Track), Antares AutoTune.

AutoTune works fine with low CPU usage (1.5 - 2%) as long as it's being fed a signal (even silence, such as between verses). But when there's no signal being sent (empty space on track between items) CPU usage jumps to over 70% and playback stutters.

Could this be a problem with Reaper or isolated to my system?


tj

Justin
11-25-2006, 12:04 PM
Hmm I have tested autotune extensively without such a problem.. you can try turning on the prefs/audio option "always run FX"... that might help you..

-Justin

teej813
11-25-2006, 02:28 PM
Thanks for your quick reply, Justin. I'll give it a shot!


tj

teej813
11-26-2006, 11:46 AM
This doesn't seem to be isolated to AutoTune, Justin.

I took a screenshot, just in case that gives you an idea of what the problem might be. http://ssmrocks.com/tj/error.jpg

As long as there is a signal being sent to the effect, CPU usage is low (hovers around 2.5%). When there's no signal, CPU usage jumps way up. Sometimes, it jumps to around 25% or so. Other times, it jumps to well over 50% and causes stutters.

I've tried a few things:
- Deleted the Verb track the Vocal track was sent to
- Deleted the parent folder and tested
- Recreated the parent folder and tested

Same result.

If you look at the image, you'll see that the play cursor is over a silent section of the vocal track and causing high CPU usage. This is different from what i saw and posted originally.

Any chance this is related to I/O? The track was originally sent to both a parent folder and a Verb track that was also part of the folder. I reason i ask is that i see a similar thing on the Snare track using SStillwellrbj4eq (http://ssmrocks.com/tj/error2.jpg). In both situations, the track causing trouble is part of a folder that has a Verb aux track included. Am i setting these up wrong, perhaps?

Ideas? Anything i can do to give you more info?


tj

Justin
11-26-2006, 01:01 PM
This doesn't seem to be isolated to AutoTune, Justin.

I took a screenshot, just in case that gives you an idea of what the problem might be. http://ssmrocks.com/tj/error.jpg

As long as there is a signal being sent to the effect, CPU usage is low (hovers around 2.5%). When there's no signal, CPU usage jumps way up. Sometimes, it jumps to around 25% or so. Other times, it jumps to well over 50% and causes stutters.



Hmm what CPU? This sounds like a denormal issue..

-Justin

teej813
11-26-2006, 01:27 PM
i appreciate you taking time to try to help me, Justin. I believe you when you say you've done full testing, and I've not heard from anyone else having problems like this. It's probably a problem with my system.


Hardware:
Msi Neo2
P4 2.4g
1 gig Crucial RAM (matched sticks)
Matrox G550 video
Emu 1820m
Samsung SATA data disk
Western Digital PATA OS disk
Windows XP SP2


So i assume, then, that this isn't an I/O, routing problem? It's OK to add a Verb folder inside a parent folder?


tj

LOSER
11-26-2006, 02:51 PM
Hi there,

it really seems like a denormal prob.
You could try this plug-in: http://www.digitalfishphones.com/main.php?item=2&subItem=6
Put it right in front of the problematic plug-in (there're also instructions provided on the page).

BTW Hasn't Stillwell made a similar JS plug-in?

Edit: Is "Reduce denormalization from plug-ins" enabled? (in prefs -> plug-ins)

teej813
11-26-2006, 05:35 PM
thanks, Loser. I wasn't familiar with this issue, but it certainly seems to describe what i'm seeing. I'll DL the plug and give it a shot.

Yes, the Reduce option is still enabled.


I'm still working my way thru this long list of rather nondescript plugs and can't say i know the diff between SStillwell plugs and JS plugs. The rbj4eq seems pretty nice. Frankly, can't make much sense out of many of the others.


tj

LOSER
11-27-2006, 01:27 AM
I'm still working my way thru this long list of rather nondescript plugs and can't say i know the diff between SStillwell plugs and JS plugs. The rbj4eq seems pretty nice. Frankly, can't make much sense out of many of the others.


JS (short for jesusonic) is just the FX API where users can program there own plug-ins/FX and Stillwell is the guy who coded some e.g. the rbj4eq and others that are really nice too.

Justin
11-27-2006, 01:31 AM
Try extracting this .zip over 1.44, and see if it helps your autotune issues:

http://reaper.fm/files/reaper144-denorm2.zip

Thanks,

Justin

teej813
11-27-2006, 05:05 PM
Thanks for explaining, Loser. That helps.



oops... sorry, Justin. This is the first chance i've had to get back online.

I see that 1.45 is out. I'm DLing it now and will do a test. If i'm still having the problem, i'll try your .zip fix.


tj

teej813
11-27-2006, 05:45 PM
Ok, i upgraded to 1.45 and did some tests. The vocal track with AutoTune works great... no CPU spikes between items or after the vocals end.

The kick track, however, still spikes at the end of the song. Last hit, kick track ends, guitars ring past the kick item, and CPU spikes.

http://www.ssmrocks.com/tj/CPU.jpg

I loaded several EQs and tested them all. Here's what i found:
- nTrack Graphic EQ played fine
- JS-parametric_eq played fine
- ReaFir played fine

CPU usage on all of the above 3 was reduced after the kick item ended.

- nTrack Parametric EQ spiked (85%)
- Rbj4eq spiked (177%)
- rbj7eq spiked (240%)
- rbj1073 spiked (282%)

Again, they all played fine with reasonable CPU usage until the kick item ended.

I see now that your zip fix is designed for 1.44. I'm hesitant to install it until i hear from you.


tj

Justin
11-27-2006, 05:53 PM
Ok, i upgraded to 1.45 and did some tests. The vocal track with AutoTune works great... no CPU spikes between items or after the vocals end.

The kick track, however, still spikes at the end of the song. Last hit, kick track ends, guitars ring past the kick item, and CPU spikes.

http://www.ssmrocks.com/tj/CPU.jpg

I loaded several EQs and tested them all. Here's what i found:
- nTrack Graphic EQ played fine
- JS-parametric_eq played fine
- ReaFir played fine

CPU usage on all of the above 3 was reduced after the kick item ended.

- nTrack Parametric EQ spiked (85%)
- Rbj4eq spiked (177%)
- rbj7eq spiked (240%)
- rbj1073 spiked (282%)

Again, they all played fine with reasonable CPU usage until the kick item ended.

I see now that your zip fix is designed for 1.44. I'm hesitant to install it until i hear from you.


tj

yeah that fix was included in 1.45 so no need to try that .zip.

teej813
11-27-2006, 06:16 PM
Understood. Again, sorry i didn't respond sooner.

If nothing else, all this testing forced me to play with a bunch of different EQs, and i was able to determine which i can use safely inside Reaper. While i like the graphical interface and ease of use of the nTrack graphical, ReaFir, without a doubt, produced the best overall sound.


tj

Justin
11-27-2006, 07:21 PM
hmm, what source material is it? .wav? what bit depth if so?

teej813
11-28-2006, 06:50 PM
yup, WAV. 44.1k at 24 bits.

I saw the CPU spike on several channels at one time or another, depending on what fx i had loaded at the time. I've been using this test song for all my Reaper testing. There are 24 tracks, but 1 is a Click, and several are folders or verb (aux) sends. Most are mono, most are full length renderings, but a couple (lead track, vocal track ) are several short parts spaced throughout the song.


tj

Justin
11-28-2006, 06:55 PM
yup, WAV. 44.1k at 24 bits.

I saw the CPU spike on several channels at one time or another, depending on what fx i had loaded at the time. I've been using this test song for all my Reaper testing. There are 24 tracks, but 1 is a Click, and several are folders or verb (aux) sends. Most are mono, most are full length renderings, but a couple (lead track, vocal track ) are several short parts spaced throughout the song.


tj


does the CPU spike show up in the windows task manager as well? or just in the FX/performance window?

teej813
11-30-2006, 07:18 AM
Haven't had a chance to check this, Jusin. Sorry! I definately get stutters and dropouts... big time, actually. Should have a chance tonight.


tj

teej813
12-02-2006, 05:59 AM
Sorry for the delay, Justin. This is the first i've been in my studio in a few days.

Yes, the CPU spike is reflected in Task Manager as well. The following screenshots were taken as the song is fading out. The CPU Usage History graphs reflect approx 3 secs of no activity followed by approx the last 10 seconds of the song. I do not have a dual-core processor. HyperThreading is enabled.


This image was taken with ReaFir enabled. No spike, no stuttering.
http://www.ssmrocks.com/tj/CPUreafir.PNG


This image was taken with rbj4eq enabled.
http://www.ssmrocks.com/tj/CPUrbj4.PNG

This image was taken with rbj7eq enabled.
http://www.ssmrocks.com/tj/CPUrbj7.PNG

This image was taken with rbj1073eq enabled.
http://www.ssmrocks.com/tj/CPU1073.PNG


Hope this helps some.


tj

LOSER
12-02-2006, 06:20 AM
Try those: [link removed, bad try]

These are slightly modified Stillwell rbj EQs. I made them add a little DC-offset (just unnoticable -600dBFS), thus they don't denormalize.

teej813
12-02-2006, 07:04 AM
Thanks, Loser.

I reran these tests. The DC versions of 4eq & 1073 now do not spike. 7eq still does. Is it possible that the 7eq you zipped is still the old version?


tj

LOSER
12-02-2006, 07:58 AM
Ahh sorry I forgot that the 7 EQ has a highpass so the DC needs to be injected after the highpass or it gets filtered :)

[link removed, bad try #2]

P.S. There might still be some spiking due to the highpass...

teej813
12-02-2006, 10:24 AM
Ahhhhhh... yeah, that makes sense.

This version is better. Didn't totally eliminate the spike, but much better and certainly useable. Thanks!


tj

RedStone
12-08-2006, 09:54 AM
woah LOSER ... sweet!

LOSER
12-08-2006, 02:19 PM
Ahhhhhh... yeah, that makes sense.

This version is better. Didn't totally eliminate the spike, but much better and certainly useable. Thanks!


tj

So I had/took a minute and (hopefully) fixed this issue, too.

http://stash.reaper.fm/oldsb/6728/SStillwell_rbj_dcADD.zip

The avarage CPU usage might be a tiny bit higher but should at least spike no more...