Author |
Message |
DrJustice
Joined: Sep 13, 2004 Posts: 2114 Location: Morokulien
Audio files: 4
|
Posted: Wed Nov 19, 2008 12:36 pm Post subject:
Audio latency - real results? |
|
|
I'm considering a bold move to save myself a few thousand hours of hard work, namely setting up a PC for use as a synthesizer.
One of the things that has kept me from doing anything of the sort previously is the latency issue. Thus I wonder what the minimum achievable latency is these days. I assume that the audio hardware and its drivers are the main culprits, but I'm also interested in the performance of stand alone VST host software. I also assume that with CPU's of a GHz or more with suitably fast memory, the CPU and memory speed is not much of an issue for latency (although it obviously will be for throughput, but that is a different matter) - after all, it only has to respond to an interrupt and copy/write a fairly small (hopefully!) buffer. It is very hard to get hold of any real numbers, so I'll rely on my EM friends for a reality check.
The synthesis software I'm interested in running is SynthMaker and eventually VSTs generated by it. I will also likely run some homemade software that will not introduce more latency than the minimum buffer size that the audio driver API can work with. The setup I'd run would be based on Windows XP and Core Duo CPU with fast RAM, say around 2GHz and 800MHz, and the OS would be configured to run as lean as possible.
Now to the questions.
What is the minimum stable latency from input to output that you have achieved, and what hardware and software did you use?
What is the minimum latency from a MIDI note on command to the corresponding sound is output, and what hardware and software did you use?
If anybody is actually running a SynthMaker setup, I'd naturally be interested in hearing how well it fares.
Recommendations for Audio and MIDI interfaces that are known to be fast is also appreciated.
I know that these things may be difficult to measure, or to verify in case of software reported latency, but I hope that some ballpark figures for what one can realistically expect can be reported.
TIA!
DJ
-- |
|
Back to top
|
|
|
kkissinger
Joined: Mar 28, 2006 Posts: 1354 Location: Kansas City, Mo USA
Audio files: 42
|
Posted: Thu Nov 20, 2008 8:10 am Post subject:
Latency and more latency |
|
|
ASIO offers good latency response. I use an RME800 and it is possible to run at latencies < 10ms, though for my uses I run from 13 to 25 ms. Such latencies make live performance manageable.
A firewire port offers better performance than USB, in general.
One of the overlooked areas is the PCI bus. Basically, you want to minimize the load on the PCI bus -- you want to disable any unneeded peripherals. You want to choose a video card that uses minimal PCI resources.
The VST synths that I've played with use considerable CPU and memory. Thus, the more RAM and CPU power the better. Latency is related to the buffer sizes for your ASIO driver and your external audio interface and the available bandwidth of your computer's PCI bus. Larger buffers reduce glitches however they increase latency. The more hardware channels you run, the larger the buffers you need. If you are running stereo, you will be fine (I don't run into problems until I try to exceed 18 hardware channels: 8-in, 10-out). I have minimal problems when I run 16-in/18-out. To run 24-in/26-out requires me to run with large buffers and too-much latency for live performance.
(Note: my latency problem is due to the particular video card on my laptop that is not 100% PCI compliant.)
I haven't used synthmaker, however have messed around with VST synthesizers that came with CUBASE. I haven't encountered any noticeable latency when triggering the VST synths from external keyboards and controllers.
Good luck! _________________ -- Kevin
http://kevinkissinger.com |
|
Back to top
|
|
|
DrJustice
Joined: Sep 13, 2004 Posts: 2114 Location: Morokulien
Audio files: 4
|
Posted: Thu Nov 20, 2008 9:37 am Post subject:
Re: Latency and more latency |
|
|
Thanks for the reply!
kkissinger wrote: | ASIO offers good latency response. I use an RME800 and it is possible to run at latencies < 10ms, though for my uses I run from 13 to 25 ms. Such latencies make live performance manageable.
|
I've heard good things about the RME hardware and drivers. Not much more than 10ms would be good. I take it that's the time from the VST instrument starts until sound is heard?
If it can be assumed that the audio input-to-output latency is no more than twice that, then we're getting into the land of feasible realtime fx processing too. The Last time I tried that, the delay made it impossible - it sounded like an echo...
To put things into perspective, if I delay the signal from my VG-99 by more than 15ms, playing starts to get uncomfortable. At 30ms the fun is over. That is in addition to the natural 10-20 ms contributed by the distance to the speakers - I'm not used to being very far from them.
Quote: |
One of the overlooked areas is the PCI bus. Basically, you want to minimize the load on the PCI bus -- you want to disable any unneeded peripherals. You want to choose a video card that uses minimal PCI resources.
|
Good point. I would think there's only any significant gfx-card PCI traffic during actual screen updates, but you have to cover for that anyway. 640x480 at 8 bit colour and no animations on screen is the way to go
Quote: | The more hardware channels you run, the larger the buffers you need. If you are running stereo, you will be fine
|
Another good point. However, I will not need more than 8 channels. That's less than 1MB/sec of audio data bandwidth, and seems like nothing compared to memory and PCI-bus bandwidth, especially as a new system would probably be using PCI-E.
One of the worrying things, and what I suspect is one of the culprits, is that interrupt response times in XP (and probably most other desktop OS'es) are all over the place. I haven't read up enough on that, but one report I read said it ranged from a few µs to several thousand µs with a constant workload. I.e. several ms just to be sure to avoid missed ISR calls. The audio bandwidth requirements themselves should be peanuts in todays super fast PCs, even with the buffers being copied a good number of times.
Quote: |
I haven't encountered any noticeable latency when triggering the VST synths from external keyboards and controllers.
|
Good to hear. I hope a MIDI message can normally trigger a VST within less than 5ms of reception.
Thank you. I will investigate this a bit more before I make any decisions, but if the latency from an external MIDI note-on to sound is heard can be in the range of 10-20 ms, then there's hope
DJ
-- |
|
Back to top
|
|
|
kkissinger
Joined: Mar 28, 2006 Posts: 1354 Location: Kansas City, Mo USA
Audio files: 42
|
Posted: Thu Nov 20, 2008 11:03 am Post subject:
Re: Latency and more latency |
|
|
DrJustice wrote: | If it can be assumed that the audio input-to-output latency is no more than twice that, then we're getting into the land of feasible realtime fx processing too. The Last time I tried that, the delay made it impossible - it sounded like an echo... |
I'm running with a 3.4GHZ Pentium IV with 4 gig of RAM. Even after chaining five or six VST effects, I haven't encountered any noticeable latency buildup. Yes, I know that sounds pretty amazing -- however processing time can be overcome by plenty of RAM and a fast CPU.
In other words, once you have overcome the CPU and memory issue, the the PCI bus, Firewire port, and hardware is the remaining bottleneck.
The latency figures I mentioned are round-trip figures. I have found no noticeable latency for MIDI messages -- of course, there is some latency but nothing I could notice. I agree, that when the latency exceeds 15ms, the rhythmic precision suffers in a real-time application. _________________ -- Kevin
http://kevinkissinger.com |
|
Back to top
|
|
|
BobTheDog
Joined: Feb 28, 2005 Posts: 4044 Location: England
Audio files: 32
G2 patch files: 15
|
Posted: Thu Nov 20, 2008 11:28 am Post subject:
|
|
|
RME drivers are top notch for OSX that is for sure, I had nothing but problems with their VISTA drivers though.
With Logic on my MacPro I can easily run large projects (32 tracks, >50 plugins) at 48K with a 32 sample buffer, so a couple of ms output latency.
I find buffer settings can be set much lower with CoreAudio rather than ASIO. With ASIO and XP a 128 buffer size is required for me (1024 in Vista!!).
I have absolutely no latency issues with the MacPro and a fireface 800.
Andy |
|
Back to top
|
|
|
BobTheDog
Joined: Feb 28, 2005 Posts: 4044 Location: England
Audio files: 32
G2 patch files: 15
|
Posted: Thu Nov 20, 2008 11:30 am Post subject:
|
|
|
Also make sure you use a usb midi interface, midi over firewire seems to have latency problems, have a search on google. |
|
Back to top
|
|
|
DrJustice
Joined: Sep 13, 2004 Posts: 2114 Location: Morokulien
Audio files: 4
|
Posted: Thu Nov 20, 2008 11:32 am Post subject:
Re: Latency and more latency |
|
|
kkissinger wrote: | The latency figures I mentioned are round-trip figures. |
Just to be sure, you mean that you can get <10ms audio input-to-output?
DJ
-- |
|
Back to top
|
|
|
BobTheDog
Joined: Feb 28, 2005 Posts: 4044 Location: England
Audio files: 32
G2 patch files: 15
|
Posted: Thu Nov 20, 2008 11:39 am Post subject:
|
|
|
Yep easy less than 5 ms.
One thing you need to keep an eye on really is Deferred Procedure Call Latency, shite drivers in XP can cause real problems with this.
Get this app http://www.thesycon.de/deu/latency_check.shtml to check out what is going on.
Andy |
|
Back to top
|
|
|
BobTheDog
Joined: Feb 28, 2005 Posts: 4044 Location: England
Audio files: 32
G2 patch files: 15
|
Posted: Thu Nov 20, 2008 11:48 am Post subject:
|
|
|
To be quite honest if you want low latency I would go for OSX.
XP offers lower cpu usage for plugins but higher latency.
A trade off but for me OSX is very solid at low latencies.
Also worth noting is that midi messages will be quantised to the audio buffer as any midi events for plugins will be rendered into that audio buffer.
Lots of people complain about midi interface latency where the real problem is the audio buffer size.
Andy |
|
Back to top
|
|
|
BobTheDog
Joined: Feb 28, 2005 Posts: 4044 Location: England
Audio files: 32
G2 patch files: 15
|
Posted: Thu Nov 20, 2008 11:53 am Post subject:
|
|
|
Whoops answered someone else's post, sorry |
|
Back to top
|
|
|
DrJustice
Joined: Sep 13, 2004 Posts: 2114 Location: Morokulien
Audio files: 4
|
Posted: Thu Nov 20, 2008 11:57 am Post subject:
|
|
|
BobTheDog wrote: | Also make sure you use a usb midi interface, midi over firewire seems to have latency problems, have a search on google. |
Did a search, couldn't really find any hard info... is that an OSX thing or a interface/protocol thing? Like firewire MIDI messages having to wait for a ride in a lengthy data frame or something?
Quote: | Yep easy less than 5 ms.
|
Now we're talking! But that's presumably with your Mac using CoreAudio, which isn't an option on XP, of course. But if ASIO on XP can steady with 128 samples buffer, that's less than 3ms - not bad.
Quote: | Whoops answered someone else's post, sorry
|
No sweat - the info is appreciated anyway
DJ
-- |
|
Back to top
|
|
|
DrJustice
Joined: Sep 13, 2004 Posts: 2114 Location: Morokulien
Audio files: 4
|
Posted: Thu Nov 20, 2008 12:42 pm Post subject:
|
|
|
BobTheDog wrote: | To be quite honest if you want low latency I would go for OSX.
|
OSX is not an option I'm afraid. It won't run SynthMaker, SynthEdit and a few other pieces of software that I'd like to run. Hopefully the difference isn't enormous. 5ms VSTi-to-output on XP should be bearable...
On the VST front, I've found a couple of stand alone VST hosts:
MiniHost and VSTHost. I'd be interested in any experiences with those, or other stand alone hosts.
DJ
-- |
|
Back to top
|
|
|
BobTheDog
Joined: Feb 28, 2005 Posts: 4044 Location: England
Audio files: 32
G2 patch files: 15
|
Posted: Thu Nov 20, 2008 1:04 pm Post subject:
|
|
|
Well I am not sure where I got the midi latency over firewire thing from, I have just done a quick search on google and found nothing but somehow I am sure this is the case!
Have been wrong in the past though |
|
Back to top
|
|
|
destroyifyer
Joined: Mar 22, 2006 Posts: 425 Location: Babylon
Audio files: 4
|
Posted: Sun Nov 23, 2008 4:37 am Post subject:
|
|
|
I'll let it be know that there's a nice freeware modular program floating around...
Actually makes me not so upset that I lost my cd case with my Moog Modular V in it. And my other massive hoard of VST's and other audio programs. Not to mention my entire cd collection. God dammit. Okay I'm still upset about that. But other than that it's great. Top secret shit.
for the topic I've got it down to 11ms latency and my computer's kind of a shit box. 1.5ghz cpu. 1.5 gb ram. Something to think about...
That's a latency I could never notice. |
|
Back to top
|
|
|
DrJustice
Joined: Sep 13, 2004 Posts: 2114 Location: Morokulien
Audio files: 4
|
Posted: Sun Nov 23, 2008 10:55 am Post subject:
|
|
|
destroyifyer wrote: | I'll let it be know that there's a nice freeware modular program floating around... ...Top secret shit.
|
You mean the nice freeware modualr is?
Quote: |
for the topic I've got it down to 11ms latency and my computer's kind of a shit box. 1.5ghz cpu. 1.5 gb ram. Something to think about...
|
Not too bad then. I wouldn't say that's a slow computer; 1.5GHz should be capable of quite a bit of synthesis...
DJ
-- |
|
Back to top
|
|
|
destroyifyer
Joined: Mar 22, 2006 Posts: 425 Location: Babylon
Audio files: 4
|
Posted: Wed Nov 26, 2008 4:25 am Post subject:
|
|
|
Thanks, my computer is subject to flattery. Hey this latency stuff is really lame. I noticed it today with my audio-in. Our audio signals should be traveling at radio-wave speed, like fiber optics or whatever. This shouldn't be an issue. I'm saying that audio latency is just a tiny kink in computer world that needs to be eliminated. My audio experience lies mainly in the computer world, and I'll find a way to get an un-notice-able amount of latency. I'll get back to you in a few years on this one. |
|
Back to top
|
|
|
|