Login

More Processing Power : Server or Cuda?

Tips & tricks, working results, technical support

Re: More Processing Power : Server or Cuda?

Postby Definity » Mon Sep 19, 2011 6:19 pm

CoolColJ wrote:Ok tried AlexB's TFT console demo which is also super CPU hungry, and at first it didn't work with a DSP buffer setting of 512 or under, but once set to 1024 it works

CPU load also 0-2%

The reverb worked fine with a buffer of 512 and under, but not the console one

Can't tell you how many instances I can run yet, I'll test that later today.
But I did try insert 2 instances in Wavelabs master FX chain and then it stopped working...


AH Sweet! could you see how many you could run before the Cuda and how many after. it would be good to finally get some acutal numbers going rather than just gustimates
Definity
Member
Member
 
Posts: 308
Joined: Tue Apr 06, 2010 2:54 am

Re: More Processing Power : Server or Cuda?

Postby Classik » Mon Sep 19, 2011 10:17 pm

Definity wrote:
CoolColJ wrote:Ok tried AlexB's TFT console demo which is also super CPU hungry, and at first it didn't work with a DSP buffer setting of 512 or under, but once set to 1024 it works

CPU load also 0-2%

The reverb worked fine with a buffer of 512 and under, but not the console one

Can't tell you how many instances I can run yet, I'll test that later today.
But I did try insert 2 instances in Wavelabs master FX chain and then it stopped working...


AH Sweet! could you see how many you could run before the Cuda and how many after. it would be good to finally get some acutal numbers going rather than just gustimates


+1

Coolcolj, thanks for sharing this with us.

Besides, if I go with Server, it is not clear to me whether it is possible to run simultaneously a local AND a remote Server? And if yes, upon creation of a new instance of Nebula, how do you choose which PC (local or remote) will handle the load?

Best,
Thomas
Studio Sequenza
Acoustic Recordings
http://www.sequenza.fr
Classik
User Level 0
User Level 0
 
Posts: 8
Joined: Sun Sep 12, 2010 7:10 pm

Re: More Processing Power : Server or Cuda?

Postby Arne » Mon Sep 19, 2011 10:44 pm

You can run both server remote and locally at the same time, and you can run several remote servers on dedicated machines, meaning you can set one IP per server, and with a dedicated client setting (XML file) which point to the IP of each spesific server.

In all practial terms, just copy the client VST Nebula Reverb Pro.dll file and related XML file, rename them to whatever you want, (important to maintain the same name on both files) and set the Server IP for each of the client XML file in use. For local server set just a . or 127.0.0.1 as IP.
Arne
User Level XI
User Level XI
 
Posts: 170
Joined: Sat Nov 20, 2010 10:31 pm
Location: Bergen, Norway

Re: More Processing Power : Server or Cuda?

Postby CoolColJ » Tue Sep 20, 2011 1:58 am

I gave it a good shot this morning, and it seems like there is a "hard cap" as to how many instances you can run regardless of wether it's Native or Cuda doing the work. And from what I can see, Native CPU power is way more useful right now, except maybe for reverb :D

----

Firstly I found Nebula3 Pro way more effecient than the Free version! Like 2-3 times more

I used Reaper for this test. Started with an all CUDA Nebula test. I experimented with the various CUDA options in the master page, and found only the highest no.15 setting to be the only one that could do more than one instance glitch free.

System specs - Windows 7 64 bit
i7 930 at 4ghz. Gigabyte UD3R X58 motherboard, 12gb of triple channel ram. Gigabyte 570GTX OC video card


CUDA only

44.1khz, 512 sample buffer size, DSP buffer = 1024 in Nebula. AlexB's TFT Pro console demo library needs this DSP buffer size as a minimum to run glitch free.
Maximum quality in Nebula, program rate set at 1ms.

I used the one master bus program on the Master, and the stereo line in on the tracks.
I capped out at 9 tracks total + Master with CUDA only...

Screencap of Reaper and windows performance meter
http://i.imgur.com/JEXTx.jpg


Native only

For native I tried to use the Nebula 64 bit, but Reaper kept crashing on loading the plugin... funny because the CUDA test used a copied 64 bit plugin with different master page parameters...
So I had to use the x86 version, but Reaper allows you to run plugins as separate instances, which does away with the 32 bit ram limits

I used the same settings and repeated the test with Native CPU power only. The CPU load per instance was a lot lower than I expected in Nebula, around 12% from memory. It was in the 30s with the Free version. Reaper wasn't reporting the CPU load properly, but the Windows performance meter is accurate.

I did hit master + 37 tracks, but Master + 35 tracks is about where is was glitch free. with 36 tracks, you would get some clicks once in a while.

screencap
http://i.imgur.com/JR1xk.jpg


Native + CUDA

I kept the above Master + 35 native tracks, and then started adding CUDA tracks. 2 extra was the max before glitches happened....

Not a lot more, but I think maybe if I drop the native track count down to 30-32, I might get those 9 extra CUDA tracks in.. :geek:

In any case those 2 extra CUDA tracks could be used for reverb etc


--------------


So I would say a $400 video card is no match for a $2000-3000 computer for Nebula grunt :mrgreen:
CoolColJ
User Level IX
User Level IX
 
Posts: 99
Joined: Wed Sep 07, 2011 4:05 pm

Re: More Processing Power : Server or Cuda?

Postby Definity » Tue Sep 20, 2011 2:46 pm

!!! Top review mate! thanks for putting the time in, finally good to see how it matches up with a CPU!
Definity
Member
Member
 
Posts: 308
Joined: Tue Apr 06, 2010 2:54 am

Re: More Processing Power : Server or Cuda?

Postby CoolColJ » Tue Sep 20, 2011 8:10 pm

I used the Nebula 3 Pro reverb BTW.

Don't know if the normal version would be lighter on CPU load.

Also maybe the experimental version might be even better


I know from playing around with CUDA on graphics and rendering that it's much faster than my CPU for those tasks, but for audio, it's not quite as suited. Yet
Like a lot faster, from 30 mins down to semi realtime
CoolColJ
User Level IX
User Level IX
 
Posts: 99
Joined: Wed Sep 07, 2011 4:05 pm

Previous

Return to Working with Nebula

Who is online

Users browsing this forum: No registered users and 6 guests

cron