PDA

View Full Version : SAC load and connected devices.



905shmick
03-23-2009, 09:07 AM
I'm still setting up my SAC rig and came across an interesting observation.

I have 2 focusrite octopres connected to each of my 2 HDSP 9652 units.

With SAC configured with 48 inputs using 2x64 buffers and 12 monitor mixes, the load was about 48%. (Asus p5q se/r, intel e8400, 4gb ram, xp pro w/sp3)

I turned off one of the octopres, load went down to 43%, then I turned off the second octopre and the load went down to 38%. Turned them back on and the load went up 5% with each one.

Is this 5% load increase per device common or is there something odd going on with my setup?

I'm only using the octopre units to test with right now, the actual rig will be using ADA8000's.

Naturally Digital
03-23-2009, 09:45 AM
Bob would need to confirm this but I have a feeling the 'input-driven' technology of SAC refers to its ability to detect 'zero' data coming from the device driver/SAW link. When you turn off one of your octopres, SAC knows not to process any data for 8 of the inputs and the load drops accordingly.

I could be completely out to lunch here of course. :)

DominicPerry
03-23-2009, 10:02 AM
If you switch the Octopres off you get no sync, so the HDSP knows to drop the inputs. I guess SAC sees that.

Dominic

905shmick
03-23-2009, 10:20 AM
So I guess I should expect to see a 5% load gain with every ADAT lightpipe input. This is really going to add up when I eventually have 9 ADA8000 units connected.

DominicPerry
03-23-2009, 10:26 AM
Time to overclock that P5Q board and E8400:D.

Both components are well up to a bit of serious o/c'ing but you'll need to be sure the CPU is properly cooled. I guess you should wait and see what the load is.

9 ADAs at 5% each is 45%, added to the 38% CPU gives you 83% - still a bit of headroom. You could always relax the machine out to 3x64 or 4x64 buffers, that will usually make a big difference.

Dominic

905shmick
03-23-2009, 10:48 AM
That's what I dont understand, I can get sac up to 80% utilization but windows reports 33% CPU util on one core while the other is sitting idle.

905shmick
03-23-2009, 11:22 AM
I don't see any load difference between 1x64, 2x64 and 3x64

If I change to 1x128 the SAC load drops from 48% to 35% and turning off the 2 octopre units drops the SAC load to 27% which takes the adat cpu overhead from 5% to 4% so 9 units would be 36% + 27% base load, for a total of 63%

Though I don't know how safe I'd feel running at 1x128.

Bob L
03-23-2009, 12:29 PM
Larger buffer sizes eases the loop processing load... because the setup time for each loop (tens of thousands of variables and stack instructions) remains the same but you process twice as many samples per loop cycle.

Smaller buffer sizes create many more times that the loop setup of those variables has to occur.

Every connected input device is adding cpu load since all inputs are being live processed... even if the buffers are zero and you do not assign them into the console routing... they are still in the loop, to maintain sync, in case you punch an assignment into the mix.

So... to lessen loads... only setup the amount of devices you will use in the Audio setup options... if you are doing a 24 channel gig... you may want to relax the machine by creating a preference set that only assigns 12 devices in the Audio Hardware Setup dialog.

Bob L

905shmick
04-14-2009, 08:24 AM
Time to overclock that P5Q board and E8400:D.

Both components are well up to a bit of serious o/c'ing but you'll need to be sure the CPU is properly cooled. I guess you should wait and see what the load is.

Dominic

I ended up overclocking the e8400 last night. I took it from stock 3.0ghz to 3.82ghz (approx 27% increase in speed). It dropped my SAC load from 47 to 38 (approx 24% reduction). I ran prime95 overnight and hit a peak of 50C on the cores with an average of 47C at 100% load. So it looks like this is a nice stable overclocked setup with plenty of cooling.

DominicPerry
04-14-2009, 12:36 PM
I ended up overclocking the e8400 last night. I took it from stock 3.0ghz to 3.82ghz (approx 27% increase in speed). It dropped my SAC load from 47 to 38 (approx 24% reduction). I ran prime95 overnight and hit a peak of 50C on the cores with an average of 47C at 100% load. So it looks like this is a nice stable overclocked setup with plenty of cooling.

That's good. Mostly good that you managed to do it so quickly and it is still stable and cool. I spent about 6 hours over 2 days trying to o/c my new Phenom II and after about 50 crashes and reboots, left it at stock speed. Need to fix the problems I have with SAW first and then o/c later.
Glad the E8400/P5Q is working well.

Dominic