A way increase latency more than card settings
Hello there,
Here's another of my not so frequent contributions to this forum, seems to be 95% reading and not so much writing for me.
I read a thread where tweaking the .ini file was discussed and began to wonder if I could set the latency for my Luna card higher than the options in the ordinary control dialogue by tweaking the buffer settings in the .ini file.
I'm an 100% audio guy (= use no MIDI) and don't care much for low latency. When mixing I would like to set latency a lot higher than possible with the the ordinary Scope settings.
Even when running in the highest ordinary SCOPE latency setting (25 ms) I feel that I waste a lot of CPU that could be used for better purposes.
Other cards have options for much lower latency. With the very nice real-time monitoring in the mixer it seems to be such a waste not beeing able to set higher latency.
Is it possible to do such a tweak? Am I totally on my own with theese thoughts?
Mikael
<font size=-1>[ This Message was edited by: BB54 on 2004-09-15 04:56 ]</font>
Here's another of my not so frequent contributions to this forum, seems to be 95% reading and not so much writing for me.
I read a thread where tweaking the .ini file was discussed and began to wonder if I could set the latency for my Luna card higher than the options in the ordinary control dialogue by tweaking the buffer settings in the .ini file.
I'm an 100% audio guy (= use no MIDI) and don't care much for low latency. When mixing I would like to set latency a lot higher than possible with the the ordinary Scope settings.
Even when running in the highest ordinary SCOPE latency setting (25 ms) I feel that I waste a lot of CPU that could be used for better purposes.
Other cards have options for much lower latency. With the very nice real-time monitoring in the mixer it seems to be such a waste not beeing able to set higher latency.
Is it possible to do such a tweak? Am I totally on my own with theese thoughts?
Mikael
<font size=-1>[ This Message was edited by: BB54 on 2004-09-15 04:56 ]</font>
Its actually a slight misconception that lower latencies increase cpu usage. What actually happens is that irq calls to the pci bus are increased (dramatically) due to smaller buffer sizes (more buffer transfers to/from pci bus) and hence more wasted cycles by the cpu waiting for things to move one way or the other.
There's usually a 'sweet spot' for performance, and you can go above this a bit to decrease the number of irq calls (and hence increase cpu performance). However if you go TOO far you'll actually create buffers that are so large the the cpu has a harder time dealing with the current dataset and this will reduce performance dramatically, even compared to low latencies.
Of course we're talking REALLY large buffer sizes so the rule of thumb that everyone is used to (higher latencies = better cpu usage) works out in the end, its just not 100% accurate (much like understanding what 'poles' in a filter really are).
There's usually a 'sweet spot' for performance, and you can go above this a bit to decrease the number of irq calls (and hence increase cpu performance). However if you go TOO far you'll actually create buffers that are so large the the cpu has a harder time dealing with the current dataset and this will reduce performance dramatically, even compared to low latencies.
Of course we're talking REALLY large buffer sizes so the rule of thumb that everyone is used to (higher latencies = better cpu usage) works out in the end, its just not 100% accurate (much like understanding what 'poles' in a filter really are).
Yes, perhaps it's the PCI bus that is the first to run out of resources.
Anyway, I think the possibility to run at higher latencies than 25 ms is an essential tool for maintaining stable PC performance at every stage of the recording process (and even may extend the computer hardware life cycle).
From experience with other cards I'd say that the useful latency range would be up to at least 250 ms.
Seems like an easy thing to implement too, I can't figure out why it's not there already.
Anyway, I think the possibility to run at higher latencies than 25 ms is an essential tool for maintaining stable PC performance at every stage of the recording process (and even may extend the computer hardware life cycle).
From experience with other cards I'd say that the useful latency range would be up to at least 250 ms.
Seems like an easy thing to implement too, I can't figure out why it's not there already.
A choking computer is far more annoyíng than a quarter of a second's delay. Still can't see why Creamware don't offer its users the opportunity to decide what's best for themselves in this matter - like a majority of competing vendors do.
Like I said before, using non-ASIO drivers isn't an option for me since I often record multiple tracks simultaneously.
<font size=-1>[ This Message was edited by: BB54 on 2004-09-17 05:18 ]</font>
Like I said before, using non-ASIO drivers isn't an option for me since I often record multiple tracks simultaneously.
<font size=-1>[ This Message was edited by: BB54 on 2004-09-17 05:18 ]</font>
My RME card has the ability to go up to 183ms latency, but I'll be honest with you--above 46ms I do start to exhibit a lot of cpu issues (read statement above)
The card has 18 ASIO i/o so for a latency of 93ms (the next step above 46ms) that's a buffer size of 4096. 18*4096 is 73728 24-bit samples, and each sample is a 24-bit value moving through a 32-bit bus (8 unused bits). Now consider that this is only IN or OUT--so the actual amount of 'bits' is 2x as much (ins+outs) for a total buffersize of 147456 24-bit 'samples', in 32-bit form (again with the wasted 8bits of headroom) that's 4,718,592 'bits' of data being shuffled about.
At 6ms the buffersize is 256(samples)*18*2 = 9216 'samples' or 294,912 'bits' of data to crunch on. Quite a massive difference.
Use the wav drivers as others have suggested. I personally use 4-7ms latency when 'composing' and if necessary I bump it up to 13ms for a bit of extra cpu headroom while mixing. This holds true both for the RME and Creamware cards.
If you really do need better performance out of your system I might suggest checking this thread:
http://www.planetz.com/forums/viewtopic ... forum=19&1
<font size=-1>[ This Message was edited by: valis on 2004-09-17 05:45 ]</font>
The card has 18 ASIO i/o so for a latency of 93ms (the next step above 46ms) that's a buffer size of 4096. 18*4096 is 73728 24-bit samples, and each sample is a 24-bit value moving through a 32-bit bus (8 unused bits). Now consider that this is only IN or OUT--so the actual amount of 'bits' is 2x as much (ins+outs) for a total buffersize of 147456 24-bit 'samples', in 32-bit form (again with the wasted 8bits of headroom) that's 4,718,592 'bits' of data being shuffled about.
At 6ms the buffersize is 256(samples)*18*2 = 9216 'samples' or 294,912 'bits' of data to crunch on. Quite a massive difference.
Use the wav drivers as others have suggested. I personally use 4-7ms latency when 'composing' and if necessary I bump it up to 13ms for a bit of extra cpu headroom while mixing. This holds true both for the RME and Creamware cards.
If you really do need better performance out of your system I might suggest checking this thread:
http://www.planetz.com/forums/viewtopic ... forum=19&1
<font size=-1>[ This Message was edited by: valis on 2004-09-17 05:45 ]</font>
Would that be the ASIO Multimedia Driver? Then I'd have to set up and switch to a special SFP setup for that, and if I don't remember wrong there is other drawbacks apart from that. Not really what I'm looking for. I am serious with my audio work and don't see why I should settle with something less than ideal with my Creamware setup.On 2004-09-17 13:32, garyb wrote:
you can record as many tracks at a time as you want with wav drivers, just add more modules. i suspect that punchins will suck with that kind of delay, but if you monitor in scope i guess it doesn't matter.....
I still don't see why the Creamware latency window is so extremely narrow when compared to other vendors. Really, there must be a lot of users out there that could make good use of higher latency than 25 ms during mixdown.
No, just add the desired number of 24-bit Wave Source/Destinations from the SFP Software IO menu.On 2004-09-20 16:39, BB54 wrote:
Would that be the ASIO Multimedia Driver? ...
Set the maximum number in Window's device manager tab (dunno the limit right now)
you already do by running that OS......and don't see why I should settle with something less than ideal with my Creamware setup. ...

cheers, Tom
You can try it. The file you are looking for is in SFPAppBin and it's called cset.ini. The ASIO block size (and hence latency value) is under the [hw] header, and it called intBlkSize. You can try changing the value in the ULLI setting screen in SFP and looking at the file afterwards to see this is the case.
I've never tried changing it by editing that file, but now that you mention it, I'll probably try it. There might be something in SFP to override ASIO settings from the .ini file if it's too high. I suggest using a power of 2 (like 4096 or 8192 etc) as a buffer value, otherwise weird (the not-nice kind of weird) things might happen. In fact, weird (still of the not-nice kind) things might happen anyway if you input a non-standard value, so try at your own risk! I'm not responsible for any puff of smoke or alien invaders that might crawl out of your card afterwards.
If anyone else tries it, please report. I'm gonna see if I can get a lower CPU usage when using Logic/SFP. Mostly out of curiosity because I've never had a problem, but I guess more CPU can often drive a mixdown session down into deeper ends.
<font size=-1>[ This Message was edited by: symbiote on 2004-09-25 01:22 ]</font>
I've never tried changing it by editing that file, but now that you mention it, I'll probably try it. There might be something in SFP to override ASIO settings from the .ini file if it's too high. I suggest using a power of 2 (like 4096 or 8192 etc) as a buffer value, otherwise weird (the not-nice kind of weird) things might happen. In fact, weird (still of the not-nice kind) things might happen anyway if you input a non-standard value, so try at your own risk! I'm not responsible for any puff of smoke or alien invaders that might crawl out of your card afterwards.
If anyone else tries it, please report. I'm gonna see if I can get a lower CPU usage when using Logic/SFP. Mostly out of curiosity because I've never had a problem, but I guess more CPU can often drive a mixdown session down into deeper ends.
<font size=-1>[ This Message was edited by: symbiote on 2004-09-25 01:22 ]</font>