Processing Power

cleanbluesky · Post by **cleanbluesky** » Sun Sep 05, 2004 3:11 pm

Hello people, find myself with this sort of setup...

P4 2.8ghz, 1.5gig RAM, fast and large IDE HDs, Creamware Pulsar 2 (6DSPs) - I also have this computer networked with another P4 1.6ghz and use them both to process sound at the same time in Cubase.
My question is...

Why, when I have a fast computer, a slave computer and a DSP card... CAN I ONLY GET LESS THAN 10 CHANNELS OF F*****G REVERB...
Am I doing something wrong here? Do I really need more computers?
I wish to have SIR convolution on perhaps 20 tracks. Does this program really take up THAT MUCH PROCESSING TIME?
So I like to record in 96/24...

Suggestions anyone...

bassdude · Post by **bassdude** » Sun Sep 05, 2004 5:15 pm

working at 24/96 is going to chew up a lot of resources.
Are you trying to run individual reverbs on each track (the wrong way to do it) or are you putting 1 or 2 on an aux?

astroman · Post by **astroman** » Sun Sep 05, 2004 5:56 pm

On 2004-09-05 16:11, cleanbluesky wrote:
... I wish to have SIR convolution on perhaps 20 tracks. Does this program really take up THAT MUCH PROCESSING TIME? ...

YES - no chance to change this in 'regular' programming style.

imho 2 options exist:
transfer the critical part of the processing to machine language
or
apply a technology like the one AndreD mentions in the OffTopic forum under 'Bionic FX' - in other words give that graphic CPU a really useful job

considering the age of the author option one is not very realistic, but the technology of #2 is certainly known to him.

Obviously he's a smart guy (living in the neighbour town btw

), so you can count him in, if that kind of processing gets more exploited.
Currently it's still a bit exotic, but technically convolution is one of the most appealing tasks for such engines.

cheers, Tom

symbiote · Post by **symbiote** » Sun Sep 05, 2004 6:21 pm

I can't even run one instance of realtime convolution reverbs on this P3-1GHz ^_^ Convolution reverbs (and convolution in general) are terribly demanding computing-wise, honestly if you can get 10 of them running at the same time, it's pretty good. I'm not sure why you would want to run 20 of them, tho, I never use more than 2 or 3 reverbs per track. Worse case, you can try bouncing stuff to audio to free up some processing space.

Maybe there's a cheap Cray on eBay that could do it for you, heh.

Post by **hubird** » Sun Sep 05, 2004 7:49 pm

On 2004-09-05 16:11, cleanbluesky wrote:
CAN I ONLY GET LESS THAN 10 CHANNELS OF F*****G REVERB...

you really mean 10 different reverbs??
as Symbiote mentions this is rather weird...
I guess you are familiar with the idea of 'effect send' as distinct from 'insert effect', although you give reason to doubt it...

edit: ah, Bassdude already came up with this...

[ This Message was edited by: hubird on 2004-09-05 20:50 ]

symbiote · Post by **symbiote** » Sun Sep 05, 2004 10:33 pm

Yep, I think he meant 10 reverbs total, not all SIR ^_^

valis · Post by **valis** » Mon Sep 06, 2004 3:53 am

Why do you need SIR on up to 20 tracks in realtime?

bosone · Post by **bosone** » Mon Sep 06, 2004 4:42 am

what are you doing with 10 reverbs???
normally two or maximum three are sufficient.
even with only one reverb you can live well!

cleanbluesky · Post by **cleanbluesky** » Mon Sep 06, 2004 4:57 am

I can get about 3-4 SIR1008 at 96/24 using BOTH computers with 1 or 2 stereo tracks. If I wanted, with both computers and the DSP card I can get perhaps 10-15 non-convolution reverbs. The reason I would want 20 is
A) Drum tracks

I would like to be able to listen to a track with full effects in real time.

I must admit I like the BionicFX idea, and I have a NVIDIA card. It would also give me an excuse to buy a super-fast graphics card for Half Life 2!

How would I go about routing audio channels together in Cubase and processing them as one? Can I do this without merging the audio as well, just temporarily group and process them as one?

cleanbluesky · Post by **cleanbluesky** » Mon Sep 06, 2004 5:05 am

How do I use the effects send... this sounds promising and I think I may be about to get blasted with a faceful of the Pie of Stupidity. Is there something I have missed?
I use all my 'verbs as inserts. Also, amplifier convolution (can be very nice if recorded from the right pre-amp) takes up a lot of processing...

Post by **hubird** » Mon Sep 06, 2004 7:45 am

oh my god...

inserts are usually for 'in line' effects, like distortion, eq, etc.
'global effects like delays and specially reverbs are typical send effects, equally used by more than one (or all) channels

Life can be easy

btw forget about 24/96, unless you're recording a super classic orchestra with immense dinamics...first try to get to the basics, may be by reading the manual thoroughly, good luck

[ This Message was edited by: hubird on 2004-09-06 08:48 ]

[ This Message was edited by: hubird on 2004-09-06 08:54 ]

cleanbluesky · Post by **cleanbluesky** » Mon Sep 06, 2004 8:12 am

Well that clears that up then. Thank you for your advice, although I still can think of situations where I would want at leave 7 instances of convolution running at the same time, but the whole FX send thing can help a lot...

How I would like to use lots of convolution

2-3 guitars (each using amplitube, which doesn't hog quite as much as SIR1008)
8 drum channels (most using a send reverb, but two channels from drums using different reverb for certain effects)
2 (or even 3) vocals, which could easily use two reverbs between them.
Which all might be possible using my setup. Possible, but not easy.

In the mean time, if anyone has any advice on getting extra juice...

Thanks

Post by **hubird** » Mon Sep 06, 2004 8:32 am

On 2004-09-06 09:12, cleanbluesky wrote:
In the mean time, if anyone has any advice on getting extra juice...
Thanks

yes...less is more

Post by **garyb** » Mon Sep 06, 2004 12:19 pm

yeah, with 7 verbs going you're probably making mess more than gold....
also, it has been discussed ad nauseum, 24/96 is of little advantage when you are ending up on a 16/44.1 cd. use 24bit, but 44.1 samplerate. that will save a lot of resources. resampling will likely kill all of the gains of a high samplerate.

cleanbluesky · Post by **cleanbluesky** » Mon Sep 06, 2004 12:37 pm

Thanks, but the whole downsampling and dither thing has been discussed before and we have no conclusions on...

Sampling rate 96/24 versus 44.1/16
Does sampling from 96 have any advantage over recording simply in 44.1?
Does using 88.2 and sampling down to 44.1 have better results than 96 to 44.1?

In the meantime, does anyone have ideas how to boost my processing power...

Also, as far as using 5-7 'verbs are concerned, they would not be obvious enough to screw up the sound

wayne · Post by **wayne** » Mon Sep 06, 2004 1:20 pm

On 2004-09-06 13:37, cleanbluesky wrote:
Thanks, but the whole downsampling and dither thing has been discussed before and we have no conclusions on...

there are conclusions to be read....

In the meantime, does anyone have ideas how to boost my processing power...

processing power is not what you need, methinks

Also, as far as using 5-7 'verbs are concerned, they would not be obvious enough to screw up the sound

good luck!

if you can't achieve it with 2 or 3 (at the most verbs), 5 or 7 "not obvious" verbs ain't gonna do it.

as has been mentioned, check out your mixer's architecture, and make good use of the power at hand. You don't need to overkill with gear to have a good mix.

astroman · Post by **astroman** » Mon Sep 06, 2004 4:49 pm

On 2004-09-06 13:37, cleanbluesky wrote:
... Does using 88.2 and sampling down to 44.1 have better results than 96 to 44.1? ...

definetely not - if it's faster then it's just a trick which leaves out every 2nd sample, possibly interpolating the remaining 2 neighbours.

That's not downsampling but just spoiling the sound - it's not about 'even' division but about placing a value at the proper position on a curve.

The 'distance' between the virtual points isn't constant, so every interpolated sample has a 50% chance to be at the right position

cheers, Tom

symbiote · Post by **symbiote** » Mon Sep 06, 2004 6:56 pm

On 2004-09-06 13:37, cleanbluesky wrote:
Thanks, but the whole downsampling and dither thing has been discussed before and we have no conclusions on...

I just want to point out the 1337 timestamp of the posts. Props to that ^_^

The thing with using 2 or 3 reverbs tho, is that you can send different amounts of each signal to each reverb. So by setting 2 or 3 reverb, you can accomplish a pretty wide palette of effects/sound/stuff. By using 5 or 7, you most likely end up setting up fairly identical reverbs on a few tracks, or smearing things up by creating too many different reverb tails that all kind of interact weirdly together.

One situation where I could see using 8-10 reverbs is setting up 10 of them, one for each octave, heavily band-limited. I'm sure that would be quite fun. But I can't try it now

Then again, using a reverb on an insert still has its place for custom jobs, like a 100% wet reverb output for weird atmospheric stuff. Of course you can do the same with aux sends by muting the channel playback, so I guess I'm just being rhetoric (and I also like to process reverb outputs.)

As for sampling and dithering, I'm sure you can look up quite a bit of research papers (or you can pickup a signal processing book and work it out for yourself) pointing out to some level of distortion in resampling. Sampling at 96khz makes sense if it's going to stay at 96khz for playback, for like dvd audio, or to mixdown on a nice phat analog neve/amek console (so you'd use 96/24 dacs to feed the console (sorry I'm just daydreaming here.))

Post by **hubird** » Mon Sep 06, 2004 7:12 pm

On 2004-09-06 13:37, cleanbluesky wrote:
In the meantime, does anyone have ideas how to boost my processing power...

yes, reduce tracks and reverbs, your music will win from it and your processor will shine again

nprime · Post by **nprime** » Mon Sep 06, 2004 9:02 pm

To answer your question:
How about dual Xeon 3.06 GHz CPU's?

That should be enough power to do just about anything.

R