SAWStudio / SAC User Group

i found a 2.7g celeron w/ 760meg of ram

with the older of my 2 9652's it's running
+/-70% CPU load w/26 channels and 5 mixes lit up

[so much for the idea of 50 channels in this box]

seemingly rock solid at 2x64 so far

as i'l still need to add a couple reverbs and a DDL
i'm a bit concerned with the load

the cmos reports a 400mhz buss
and 266mhz ram

will getting more ram help to reduce it
or is that just asking more than that era of technology
can provide?

has anybody run a show in the 75-90% load range?

what happens if you hit 100%?

thnx
mh

For an old Celeron processor... that's not bad.

I have run stress tests here with my rig running at 98% load and it has run for hours with no dropped buffers...

Would I run a show that way... probably not.

I would look for a way to reduce the load... are your 5 monitor mixes using the Pst EQ tap and are you keeping those mixes free of extra eq? If not... try that... and conserve the cpu by not repeating many of the eq settings on the monitors.

Bob L

Let me preface by saying I agree wholeheartedly that one shouldn't depend on a box that is running on that high of a load to behave as it should, though SAC is amazingly stable under a stupid load on most boxes. But...in the real world, as it grows in popularity, people will try to use SAC on less computer than they should for too much show.

I ran a show on my test box that was hovering in the upper 80% range at home, and yes it did hit 100% quite a few times during the show, and was running far too high as a habit.
When it maxes out, SAC does it's absolute best to keep your audio uninterrupted (Hail Bob!), however your display and command refreshes get quite laggy. Of course there comes a point when it's too much & it will freeze up, but if you keep your processing trimmed to a minimum it can be done. Well before the system crashes you will become frustrated at the inability to make minor adjustments quickly due to the lag...think of it as an exercise in learning how to not overmix.

I am not pretending that this was kind to nice Mr. Computer, but I had to know, so:
Before I took it out, I spent an evening intentionally overloading different ways, crashing, and restarting from scratch in a hurry at home. This got me comfortable in symptoms of impending doom, and in emergency restarts. It also told me in advance that my soundcards spit out loud/ugly stuff if things are frozen up & I don't have some stage after them I can mute before I hit the power button to reboot, lol

You'll find with a little practice it is FAST to load from scratch, and if you are going to tread a thin line with resources, at least you can be prepared to fix things quickly.
If you have to use an underhorsepowered box, some basics that may conserve resources and save your show are:

Set up your system preferences & save it as default, that way you won't have to reset them all if you need to reboot or restart SAC in a hurry. Once you get the basic stage set-up, save that as a mix for the same reason.

Resave your mix often, be aware that if your CPU load is extremely high, even saving your mix will be laggy.

make Mon1 your master, patch other monitor sources post that mixer, only process monitor channels at Mon1

Watch your load meter & your buffers, look for things that can be done without and notice how any changes you make affect your load

Turn off all features (eq & dynamics, etc) for channels that aren't actually using them. Unassign unused inputs.

Make a separate scene with all processing you can kill in an emergency gone, (like...no effects at all) save THAT as a mix as well, then if things are going south you can load that scene for near instant resource recovery, and if you need to restart that is a decent 'safety' to start from.

Add new effects or auxes ONLY between songs or on break, you want time to see what they do to your load when you have time to UNDO them if they are too much additional strain.

Do not be afraid to increase the size and/or number of buffers to get more out of your system in a pinch. I know we all need as little latency as possible, but under fire if you have to choose between having the show run and having it crash, 3x256 and shopping for a better box tomorrow is a genuine option.

I'd highly recommend swapping the Celeron for its upscale equivalent - P4 or whatever.
Celerons are just too slow IMO.

cheers

mako

mako wrote:I'd highly recommend swapping the Celeron for its upscale equivalent - P4 or whatever.
Celerons are just too slow IMO.

cheers

mako

True but even a P4, Xeon, dual or quad core can be slow if not properly matched with the right MB, RAM and even HD's
do your research prior to just throwing a rig together, you won't regret it later.

I think there is a real risk of bringing SAC and Bob into disrepute by using underpowered systems. How much does (just) a 48 channel desk with 6 aux sends, monitor matrix and high quality reverb cost? Add to that as many comps and gates as you would need. How much now? $10,000? $20,000? More?

Now justify spending $1000 on a PC to do the same task, instead of a proper $3000 - $4000 rig. You can't.

In my opinion, if you want to run SAC, get a proper box to run it on.
A modern X38 or X48 chipset box with E8500 processor and 2GB of memory will run 72 channels at 1x32 buffers on 2x PCIe RME card. Where's the point in compromising?

Dominic

all great comments
especially the note about not overmixing

dominic is right, of coarse, but
in the real world it will be how much can i get it to do
for how little

as major production companies come onboard
quad core rack mount boxes will be the norm

but garage bands of the world
will still be stress testing that old gaming machine...

i got both rme cards in my celeron
[more to make sure they both worked than anything]
and got out to 30 something channels befor it started slipping

i'l also sgree you should have practice
at emergency recovery befor you go to the gig

fwiw

DominicPerry wrote: A modern X38 or X48 chipset box with E8500 processor and 2GB of memory will run 72 channels at 1x32 buffers on 2x PCIe RME card. Where's the point in compromising?

Dominic

I'm doing research to prepare for a corporate band that is also cutting over to In Ear to take me in. So....

The MB I'm looking into is the ASUS P5E which meets those conditions.
Have you used this MB yet?

Hap

Maybe we should start a new thread group like "SYSTEMS that SMOKE and those that DON'T"

Bob as I understand it never address the question "what is the best system for SAW or SAC ?". Bob just left it up to us to make that decision. He has only gave us the minimal requirements then suggested we experiment.

It would be nice to see all the setups (components etc.) and tests performed so we can establish some sort of baseline for system setups. We should try to come up with a standardized SAW/SAC system test to establish a performance baseline as well.

I do know that I have seen systems with much less power then mine have a lot better performance as well as higher track counts.

In no way do I feel that SAC or SAW is the problem with my setup, I could not be happier with them, it’s in not seeing the performance results I feel I should with a system that is by all other standards a smoking setup. It is also tweaked to the hilt and does have high track count with low CPU cycles usage. But it is unstable and hangs at times, so I have to have higher buffer settings than I would like.

Sorry I hope I didn’t ransack your thread to much.

Warren

Warren wrote:Maybe we should start a new thread group like "SYSTEMS that SMOKE and those that DON'T"

This is why I asked about 'who gets 1x32' some time back.
My conclusions, as yet supported by other people's systems but not by anything I have built is

1. Use the single fastest modern processor you can, ignoring things like dual core, quad core. That is, you should buy the fastest individual core speed you can. E.g, a dual core at 3.0GHz is better than a 2.7GHz quad core. I suspect that running SAC on a single core is the best approach for lowest latency, although leaving 2 cores running may provide more power for plug-ins.
2. Use a fast FSB speed. This needs a good board with stable overclocking if needed. Asus are good for this. The Asus P5K (in one of its variations) would be my personal choice.
3. Use fast memory. The X38 and X48 chipsets support DDR3. Not all X38 boards require DDR3, and it is expensive, but faster.
4. Use an RME PCIe card. They have lower latency settings than the PCI cards.

Dominic

DominicPerry wrote:This is why I asked about 'who gets 1x32' some time back.
My conclusions, as yet supported by other people's systems but not by anything I have built is

1. Use the single fastest modern processor you can, ignoring things like dual core, quad core. That is, you should buy the fastest individual core speed you can. E.g, a dual core at 3.0GHz is better than a 2.7GHz quad core. I suspect that running SAC on a single core is the best approach for lowest latency, although leaving 2 cores running may provide more power for plug-ins.
2. Use a fast FSB speed. This needs a good board with stable overclocking if needed. Asus are good for this. The Asus P5K (in one of its variations) would be my personal choice.
3. Use fast memory. The X38 and X48 chipsets support DDR3. Not all X38 boards require DDR3, and it is expensive, but faster.
4. Use an RME PCIe card. They have lower latency settings than the PCI cards.

Dominic

Thanks Dominic:

All great stuff we should concatenate to come up with the more perfect SAW/SAC system, then publish the data so all can easily determine what setup is best for there particular needs.

Any others want to add their setup and testing data?

Warren

DominicPerry wrote: 4. Use an RME PCIe card. They have lower latency settings than the PCI cards.

Dominic

The PCI versions (at least mine)don't have settings under X@64. Curious, would the overall track count increase on PCI-e over PCI with the same buffer settings like 1 X 64?

Bud Johnson wrote:The PCI versions (at least mine)don't have settings under X@64. Curious, would the overall track count increase on PCI-e over PCI with the same buffer settings like 1 X 64?

You must have the latest RME driver and BIOS update to do 32.

My P5K uses a PCI interface and easily runs at 3X32.

cheers

mako

mako wrote:You must have the latest RME driver and BIOS update to do 32.

My P5K uses a PCI interface and easily runs at 3X32.

cheers

mako

I thought I'd done that and it turned out it won't update. Nuendo branded hammerfalls. I could be wrong though. When I'm a little more on top of things I'll try again.
Still curious, would the overall track count increase on PCI-e over PCI with the same buffer settings like 1 X 64?
BTW, 1 X64 is good here up to 20 or so tracks. It would be nice to be able to go 3 X32 rather than having to jump to 2 X 64 though.

Guys - some great info in this thread. Warren mentioned sharing info - it got me to thinking.

Maybe we could set up a mix file and run it on our various machines. Then we could post system specs and CPU usage at various buffers.

I think it would be helpful with planning upgrades and purchases. Like right now I am curious how much laptop I'd need to buy to run 24 channels SAC and leave the rack PC home.

Just an idea.

txshonk wrote:Guys - some great info in this thread. Warren mentioned sharing info - it got me to thinking.

Maybe we could set up a mix file and run it on our various machines. Then we could post system specs and CPU usage at various buffers.

I think it would be helpful with planning upgrades and purchases. Like right now I am curious how much laptop I'd need to buy to run 24 channels SAC and leave the rack PC home.

Just an idea.

That would tell us something about Saw, not sure it directly translates to SAC, in a broad way I guess. I think maybe just 24 tracks, comps-gates-eq enabled on all 24 and 4 mon mixes copied from FOH would yield a more accurate measurement.

I guess the thing to ask Bob is, What channel settings will stress out the individual channels and groups the most? All with only using Native plugs only.
As far as Saw goes a .wav file used for each channel that everyone will use, so there are no variants.

As it stands I don't know if I am having trouble with ECC, Spread Spectrum on somewhere, RAM or CPU voltages, timings etc.

But I think with all the data out there, if it were easy to access it might be much easier to buy a high performance system as well as fine tune the hardware componets and the OS.

If I am off base here, as to if this might help others or not PLEASE let me know.

Thanks

Warren

Wow, I'm glad to see the turn this thread has taken, I completely agree that a standardized test run across a variety of user machines runs a risk of helping our whole community to more quickly be able to finalize what works best, and also allow other potential adopters to predict a bit more accurately their own experience with a variety of entry level and custom designed boxes...sounds like a great thing for all!

I should have two more dramatically more current boxes ready to test by this weekend, both are 1U custom whiteboxes. One e8500, 4GB DDR2-800, ASUS board, and the second has 2 xeon5060, 8GB DDR2-667 EEC&registered, and a supermicro board. The second of the two will be my first venture into the 64bit world...anything I should know there anyone?

I will be honored to run & post their results along with the data from the old P4 the more effective one of the two will replace Can we get a word from our fearless leader or a consensus amongst ourselves as to what might be a preferred standardized test procedure?

the only problem i see with a quote-unquote standard test
is the vast array of audio options that are applicable

i would say most systems would be pointed at +/-24 inputs
but real pro audio will be wanting to stress test 72 input rigs

and i have gigs that a hand carried 12 input rig would do fine on...

i'v proven it's not the smartest thing but firewire will do a gig

more than any one thing input count = CPU load

...jus a thought

After some more of the comments added heres my run at a "standardized" process that probably suits most rigs.

We run 2 mix sessions in SAC:
1) 12 audio tracks active
2) 24 Audio tracks active

(we can go higher track count, but iI figure we will be limited on participation - I am personally limited to 24 tracks)

On all tracks eq, comp, gate, and the SAW reverb in the pre-fader slot.

We we run each at 2 3 buffer settings - 2-128 and 2-64.

We then post following rig info:

• Motherboard (for laptop or off the shelf rigs you wont know) - brand and model
• Processor - model at speed running at (for those who overclock)
• Memory - brand, type, amount and speed
• OS Hard Drive - model, RPM
• Operating System
• Audio Device

SAC detail

• Audio driver type used.

Performance

CPU usage for each file at each buffer setting - so if we go with the above recommendations we post 4 results

1. 12 tracks 2-128 buffers
2. 12 tracks 2-64 buffers
3. 24 tracks 2-128 buffers
4. 24 tracks 2-64 buffers

FYI - personally I have an 8 space rack with my PC in a 4 space rack, 2 space slide out 15" server rack monitor, and a server keyboard tray. And the thing is heavier then all get up. So I am really looking for scaling back to a laptop or lightened 1 space Rack PC. I'd love to run 2-64 on 24 tracks - but am unsure how much laptop I'd have to buy.