CPU usage in DAW apps.

[AdamCR]

I posted something similar over at KVR so appologies if you read this already!

I'm in the middle of seriously attempting a migration away from Cubase and I've been running an experiemental project (13 tracks, each with one significan VST plugin inserted (4 of which are UAD), first three tracks send-bussed to a reveb buss) on a few DAW apps to see how things are CPU usage-wise.

The reason for my experiment is that I have a rock-solid, if elderly PC which I don't really want to upgrade. It's a P4 2.4 with 1GB RAM.

I'll be using the app mostly for mixing audio, midi is a non-issue (I'll probably stay with Cubase for midi).

Results:


Sonar 6: 85%-90% (!)

Reaper: 75% - 80%

Samplitude 7 (that's the last version I bought): 75% - 80%

Cubase SX3: 50%-55%

I was expecting Reaper to win the contest, (I have the newest version) but it actually didn't do all that well.

Are there any CPU-usage tweaks that I should make?

I'm running my soundcard at a fairly high latency since it's only for mixing (1024 clicks).

Thanks!

Adam

[Bevoss]

What version of Reaper did you use, because it just got seriously faster, lower latency and cpu. I'm talking about today, as in 1.66(Edit) build. Seriously. I can run 20+ track projects at 48 samples now, I can't do this on any other program.

Also, in preferences-audio-buffering, set threads to 2 if you are dual core. 100% in Reaper actually means 50% (200% dual core), check windows task manager readings.

I would count on this getting even better, Justin is a serious genius. I dub him Nerdmaster! :)

And in conclusion, try build 1.66 (Edit) before you draw your conclusions, it's that much improved. Cheers and good luck.

[Bevoss]

I use almost no midi, so I won't comment on that. Regarding audio editing, there is really only one feature I miss from Nuendo, that's Track lanes. You can do many takes here, but you have to pick one to be active, there's not the blending and combining of tracks within the takes...you would have to explode them to new tracks which is a bit messier and not as convenient.

But as regards the rest of the editing, it's almost a step up on every other program, bit's of Pro Tools (Tab to transient), Nuendo (Strip Silence), and some I haven't seen before (the keypad editors). Plus it's routing and keymapping are second to none.
Try it for a week and you will see how very deep it is.

[AdamCR]

What version of Reaper did you use

I used 1.65.

Cubase is far more economical at the moment, at least when using the apps for audio-only mixing.

I was surprised, I'd expected Reaper to be the least CPU-greedy!

[Bevoss]

How did you measure the cpu usage?

EDIT: I meant 1.66 earlier, the one on my machine.

BTW if you do a search I think UAD cards sometimes have problems with Reaper, and it seems to be at UAD's end ie not helpful.

[Billoon]

If you disable all plugins in both appps and just play back the tracks is the cpu usage much different?

[Art Evans]

I can't believe that any of these programs is optimised for all processors. I get the feeling that Reaper is, if anything, optimised for dual core machines although there are loads of tweaks available in its prefs. I guess what I'm saying is that what works best on your hardware might not be what works best on mine.

[Bevoss]

Good point, 2.4 G may not give the same results we're getting from dual cores. I got more than double the tracks once I turned on thread 2.

[malcolmj]

Are there any CPU-usage tweaks that I should make?

Hi Adam,

With regards to your UAD, I'd suggest watching the video in this thread:

http://www.cockos.com/forum/showthread.php?t=4643

Cheers,

Malcolm.

[Bevoss]

Great video! A key point was run your UAD cards in Reaper at the lowest latency for the lowest cpu hit. 1024 samples gives a cpu increase 40-50% higher than 128 it seems, so if you can't lower it Adam then Cubase may be a better bet (with all the bugs I'm hearing about). Or you could upgrade and join the revolution...

[AdamCR]

I can't reproduce your UAD behaviour at all. Reaper behaves exactly as I would expect in terms of latency settings; sweet-spot (CPU-wise) between 512 and 1024 (with or without UAD plugs).

If I lower the latency to 254 then CPU usage rises, if I raise it past 1024 nothing much happens.

I've been messing around all morning and I can't get Reaper's CPU usage down to Cubase's usage.

[Bevoss]

Download 1.66 if you haven't yet, it's lower latency and lower cpu. As I said, I can go down to 48 samples now (I meant 1.66 earlier, sorry)

[billybk1]

I can't reproduce your UAD behaviour at all. Reaper behaves exactly as I would expect in terms of latency settings; sweet-spot (CPU-wise) between 512 and 1024 (with or without UAD plugs).

If I lower the latency to 254 then CPU usage rises, if I raise it past 1024 nothing much happens.

I've been messing around all morning and I can't get Reaper's CPU usage down to Cubase's usage.


Hey Adam, you do not notice any native CPU munching, using 1024, at all? I have used multiple UAD-1's for years in my DAW. I just recently added (2) more cards (I now have 4 in my DAW), so I tend to use a lot of DSP intensive PPI's in my projects and at most, maybe a couple of native plugins, if I run out of DSP or if I need a particular FX that the UAD-1 does not provide.
In multi-card setups, using a lot of DSP, I have found you will get increased native CPU munching, as you increase your project latency (1024 & above being the worst), with 128 samples being the sweet spot. You may or may not get a brief CPU spike when you first start playback with PPI's enabled.
I just tried a little experiment, just to verify what I am seeing. I opened up a (23) track (all recorded audio tracks)/ (4)bus project and added (5) UAD-1 plugins so I had 23% DSP usage (equivalent DSP to (1) card in a (4) card setup). I used no other FX, just the (5)PPI's:

(2) 1176LN
(1) CS-1
(1) Plate 140
(1) Neve 1081

I tested playback @ 1024 & 128. To verify any excess native CPU usage (native CPU munching), you need to do the ol' disable/re-enable trick from the UAD-1 Performance Meter, during playback to see how far you native CPU usage drops. The native CPU between what it first reports, at initial playback and what you see after you do the disable/re-enable sequence is the native CPU munch (extra native CPU being consumed).
You need to have your PM and an actual native CPU meter (like the Windows Task Manager) to accurately and effectively test this.

1024 project sample latency:

During initial playback:
Native CPU usage: 12%-13%

After disabling and then re-enabling the PPI's from the UAD-1 PM, during playback:
Native CPU usage 3%-5% (this is what it should be using)

Clear indication of native CPU munching. The Windows Task Manager CPU meter dropped considerably. You are actually using more than twice as much native CPU than you should be using. This phenomenon gets more severe as you add more PPI's. Try using (20) or more PPI's and multple DSP cards and you can easily get 30%-50% extra native CPU load


128 project sample latency:

During initial playback, after an initial brief CPU spike to 17%:
Native CPU usage: 5%-8%

After disabling and then re-enabling the PPI's from the UAD-1 PM, during playback:
Native CPU usage 4%-8%

Not much of a difference in native CPU usage. No native CPU munching and overall native CPU usage is actually lower @ 128 samples than running the project @ 1024 with the native
CPU munching


In most cases you will see native CPU munching @ higher latencies using UAD-1 plugins, if you know how to properly identify it, during playback. You can also use REAPER's Performance Meter view window to see this as well. If you see any tracks with PPI's showing more than 1% CPU usage (should be lower than this per PPI) that is a clear indication of excess munching. Typically, you will see one PPI with 18%-22% CPU usage. It is usually one PPI that will set the native CPU munching off. The only way to get rid of it is to do the ol' disable/re-enable trick from the PM, during playback or lower your project latency. The lower the latency the less severe the CPU munch.

I have actually created identical multi-track audio (23 or more tracks) projects, using the exact same PPI's (20 or more using 4 DSP cards @ 80% plus load), in both SONAR 6.01 & REAPER 1.66. I can easily run the projects @ 128 samples in REAPER, whereas SONAR playback is all garbled and distorted with CPU overload. I had to raise it to 256 for SONAR to playback cleanly, albeit @ higher CPU usage than REAPER @ 128 samples. That tells me that REAPER's audio engine is performing much better at lower latencies than SONAR. At least, when using my E6300 Duo Core and (4) UAD-1 cards.

Cheers,

Billy Buck

[AdamCR]

Ok, confirmed.

As you say, at lower latencies (I didn't try mine lower than 256) the 'munching' is less.

Interesting stuff.

1.66 does seem to lessen CPU use; I'll continue my experiments!

[Justin]

It may be noted that since REAPER uses 64 bit values for samples at every stage in the processing pipeline, it uses a bit more memory bandwidth for the same tasks.. so if you're comparing it on a P4, that may hurt it.. Try it on a newer box and the difference should be less...

[billybk1]

It may be noted that since REAPER uses 64 bit values for samples at every stage in the processing pipeline, it uses a bit more memory bandwidth for the same tasks.. so if you're comparing it on a P4, that may hurt it.. Try it on a newer box and the difference should be less...


Yes, REAPER really does shine with Duo/Dual Core setups. It really is amazing to see. UAD-1 plugins were not really designed to be used at super low latencies. I believe the lowest recommended sample buffer is 128. Which really does seem to be it's sweet spot and works quite good in REAPER. Today, with the latest v1.66 I did another little test and the UAD-1 plugins do work @ 64. Of course, you get higher native CPU usage and stopping and starting playback, with a couple of UAD-1 cards worth of PPI's can be somewhat glitchy and the audio can be garbled for a few seconds. But it does play the PPI's @ 64 samples, which can't be said for any other DAW app that I have used before. I was playing a (30) track project (all recorded audio tracks) with (20) PPI's (80% DSP load across 4 UAD-1 cards) and the total native CPU usage was 20%-25% @ 64 samples and 10%-15% @ 128 samples (with no glitching or garbled audio, of course). That is quite remarkable to say the least. :D

From a performance standpoint, with my e6300 Core Duo, (4) UAD-1 cards & REAPER, I have never had it so good! I do all my work using REAPER (tracking, mixing, etc.)@ 128 now. I never have to change to a higher latency for anything.

Great piece of work Justin!

Cheers,

Billy Buck

[RokkD]

With 1.66 I am now able to step down from 512 to 256spls on my p4 machine (1.64 and 1.65 made no difference)

Good work J.. :)

--