Why is latency...

[Ben B]

...such a hard nut to crack?

With native CPU's these days smoking the processing power of older TDM cores, why can't they compete in the area of latency? I've heard stories of 8-core CPU users not being able to set their buffers lower than 128.

This is why I've kept my old MIX-era TDM system. I'm tracking and overdubbing into that, and then mixing on a secondary LE system at my project studio. I wish I could just skip the TDM system entirely, but I'm not confident that I'd get adequate performance during tracking out of my LE rig. (Plus, I don't care for the limitations associated with Low Latency Monitoring Mode in PTLE.) I also like to monitor post-DAW, so I need a system that can keep up (as opposed to pre-converter monitoring).

So why do native CPU's have such a hard time with latency, compared to older, "less-powerful" MIX cores? My guess is that it has something to do with the TDM bus, but that technology has been around for decades (even before Digidesign implemented it).

-Ben B

[Rqrstudio]

I use the 64 setting always while tracking and never have a problem.
I use at least an eq plug on every track and some compressors here and there also.
I only have a dual core!
(I obviously dont use low latency mode)

[Rev2010]

Not sure what you mean by Native CPU's vs. TDM Cores. So, if I'm misunderstanding your question accept my apology.

I have an Intel Core i7 920 which is quad core with hyperthreading so it shows as 8 cores. I have no problems whatsoever in Cubase SX3 (yes, the old version!). I run my Mackie Onyx 1220 Firewire mixer with the latency at 88 buffers and not have a single problem. I've gone lower too perfectly fine, I just feel 88 buffers (2ms) is a good balance between CPU usage and response.

I've read some tech articles outlining why additional cores over 4 create problems. It has something to do with the synchronization between the cores and the added overhead required. But with the i7 and it's 4 additional virtual cores I couldn't be happier.


Rev.

[Subversounds]

Some latency can come also of how much megabytes the harddisk can write per second. And most common hard disks are 7200rpm, while the files only increase in size (ie: 24bits 96khz against the 24bit/44.1 standard from a few years ago)

[kosmokrator]

Originally Posted by Ben B

...such a hard nut to crack?

With native CPU's these days smoking the processing power of older TDM cores, why can't they compete in the area of latency?

Because they ('TDM chips') are ASICs and don't have to do anything else.

[Ben B]

Originally Posted by kosmokrator

Because they ('TDM chips') are ASICs and don't have to do anything else.

Yes. But don't modern CPU's have more than the equivalent surplus horsepower? It sounds like you might know the answer!

-Ben B

[Jake Dempsey]

Logic 9 on a 8 core, 8 gig ram macpro, buffer at 64. FF800. No latency.

I ran logic 7-9 on a 4 core 2 gig macpro with 2 gigs ram for over 3 years. No latency. Wth both systems, I've used the ADAT outs on the converter to create headphone mixes. Never had any complaints. I think the nut was cracked years ago. It just depends on your computer/interface.

You could always use an analog board for monitoring before signal hits the A>D.

[robertshaw]

If you have a sound card that supports 'direct monitoring' 0 latency shouldn't matter even on a 5400 drive, ebev on a PIII cpu
I could do 0 latency in 1999 2000 on a PIII or a P4 never have any issues. Even on a dual core and now a quad I can get
close to zero wothout direct monitoring at 96/24 on a 5400 drive.

Sometimes even virus software and background processes can inhibit performance. Lots of things to look into, but hardware at this point in time should never be an issue even with really cheap gear. Maybe a soundblaster still can't do zero but anyhting thing like a lynx, presonus, rme9652 etc..... all support direct monitoring and therefore all have virtual zero letency. I mean at really can never be truly zero latency since it does take clock cycles but @ the speed of sound and a few million clock ticks @ speeds measured in nanaoseconds it could never be meaured by human senses. yes Hard disk though would be the only bottleneck but with low track counts you shouldn't notice it with direct monitoring.

[Ben B]

Originally Posted by Jake Dempsey

You could always use an analog board for monitoring before signal hits the A>D.

Yes, I know. But I have reasons to not want to do this.

-Ben B

[loopy]

Originally Posted by Ben B

Yes, I know. But I have reasons to not want to do this.

-Ben B

It's the software that is the weak link, specifically the driver for the card.

[theblue1]

Originally Posted by kosmokrator

Because they ('TDM chips') are ASICs and don't have to do anything else.

Originally Posted by Ben B

Yes. But don't modern CPU's have more than the equivalent surplus horsepower? It sounds like you might know the answer!

-Ben B

Also, a dedicated system can have extremely tight throughput tolerances that would not be possible in a multi-purpose system like today's computers. We use multiple buffers throughout something like a computer based DAW in order to 'provide slack' and provide tolerance to prevent processing bottlenecks for the system in case there are -- as there often are -- sudden, frequent spikes in demand.

[DontLetMeDrown]

Honestly I think core of the problem is with the o/s. There really should be an easier way for us to tell any o/s: "When I'm recording, don't do ANYTHING else!" We need some kind of brute force DAW firewall that stops the computer from doing anything that's not related to recording.

[theblue1]

Originally Posted by DontLetMeDrown

Honestly I think core of the problem is with the o/s. There really should be an easier way for us to tell any o/s: "When I'm recording, don't do ANYTHING else!" We need some kind of brute force DAW firewall that stops the computer from doing anything that's not related to recording.

Sounds like a good idea -- but the logistics are daunting.

Don't forget, it's not a matter of telling the OS to shut everything but Cubase or Sonar or Logic out... you've got to allow for all the drivers, utilities, background services, FX plug ins, VI's, etc, etc that are necessary to a full featured recording platform. It could be, literally, hundreds...

[zosthegoatherd]

The os does add to latency quite a bit. Windows is the worse culprit, osx is a bit better, linux can be customized with the right kernel to get very low latencies (korg used this on the oasys keyboard) In fact I think in another few years as the user space programs get better we will see custom Linux based recording solutions start hitting the market with close to zero latency and very good audio performance.

[loopy]

Originally Posted by zosthegoatherd

The os does add to latency quite a bit. Windows is the worse culprit, osx is a bit better, linux can be customized with the right kernel to get very low latencies (korg used this on the oasys keyboard) In fact I think in another few years as the user space programs get better we will see custom Linux based recording solutions start hitting the market with close to zero latency and very good audio performance.

Linux?With like 5 different sound systems (ALSA, pulse, OSS, ESS, arts, etc) is a mess.

The problem with Linux is there are far too many Indians and few chiefs.

How many Linux distributions are there?
Something like 300+ ?
Why?

Linux is a cluster**** and until the Linux community gets off the "choice" bandwagon and re-groups, Linux will remain the disaster it is now.
Windows 7 is killing Linux.

Linux just sucks for everything a human being has to get involved with.
And the Linux community is a bunch of pompous assed idiots who think their poop doesn't stink.
RTFM comes to mind.

[stagefright13]

As far as I am concerned latency is what I add to my session tracks. If it was easy everyone would do it...

John

[synthoid]

I'm working at 32 samples of latency more and more. Using an RME RayDAT card with Logic 9 on an 8-core Mac Pro. Once you get to 32 samples, you're in the ballbark of TDM systems anyhow.

I've done several live shows, doing a submix of instruments and voice from the stage at 32 samples of latency with a FF400 as well. Sometimes I use 64 samples on stage; that doesn't seem to pose much of a problem as far as noticeable latency in the monitor mix.

Seems to me like native systems are either all the way there or very nearly there at this point, with the right audio interface.

-synthoid

[Ry-Fi]

Summarily, it's because operating systems for general purpose computing are not currently designed to provide any kind of near-real-time scheduling for certain activities.

The real question is, why haven't they implemented such options yet? Why isn't there a Windows/OSX version customized for DAW/audio/video/etc use? It may or may not need to be a radically different kernel, but what it would need is a 'control panel' that allows you control over the scheduling schemas and ability to fine tune the performance of your system depending on your particular applications and needs. For example, it would allow you to mark the drivers and processes that require high priority scheduling at the expense of all other systems.

Surely this is feasible, no? To anyone's knowledge, have the big OS vendors ever considered making such "near-real-time" versions of their OSes for the omnipresent and ever-growing market where personal computers are the cornerstones of all modern technological crafts ?

[synthoid]

Originally Posted by Ry-Fi

Surely this is feasible, no? To anyone's knowledge, have the big OS vendors ever considered making such "near-real-time" versions of their OSes for the omnipresent and ever-growing market where personal computers are the cornerstones of all modern technological crafts ?

I spent many years working on software and computer architecture for real-time stuff (cell phones, communications, etc.) The thing is that it is not only an OS consideration. Take the Motorola 56K for example (the DSP inside the TDM cards). They don't use the same kind of cache or paging structure that an Intel processor does, for the very reason that their applications require knowing exactly how many cycles such-and-such an operation will take (no iffy stuff about cache misses or page faults or context switches or variable pipeline stalls allowed). So it's a whole system design and not exclusively an OS thing.

But something like Core Audio is very much an OS vendor doing near-real-time stuff in the OS. They lock all kinds of critical pages and processes down and tune them for real-time response, to get as close as is feasible on a general-purpose Intel platform. It works well, don't you think? I mean, there's a fixed latency in a hard real-time system like a 56K-based system too, and it's not much better than what is currently delivered by Core Audio with a high-end interface.

-synthoid

[Ry-Fi]

Originally Posted by synthoid

But something like Core Audio is very much an OS vendor doing near-real-time stuff in the OS. They lock all kinds of critical pages and processes down and tune them for real-time response, to get as close as is feasible on a general-purpose Intel platform. It works well, don't you think?

-synthoid

I don't know - I'm on PC. I know Windows sure picks some HORRIBLE times to do random things like launching Windows Update, disk indexing, etc. But for what it's worth, my system is tweaked and I have no complaints or unrealistic expectations about the performance I get anyways.

Thanks for the interesting perspective. It seems like you've worked on some cool stuff! thumbsup

[synthoid]

Originally Posted by Ry-Fi

I don't know - I'm on PC. I know Windows sure picks some HORRIBLE times to do random things like launching Windows Update, disk indexing, etc. But for what it's worth, my system is tweaked and I have no complaints or unrealistic expectations about the performance I get anyways.

Yeah, it's pretty f*cked when your important audio sessions comes to a grinding halt because of some high-level activity like a sw update, haha. That's not a real-time thing, that's more like a common sense thing. Even your cell phone doesn't work the same when you're updating the firmware in it haha.

Thanks for the interesting perspective. It seems like you've worked on some cool stuff! thumbsup

Yes, but these days I'm full-time in music and it's a LOT more interesting than high tech stuff IMO. (It's not nearly as lucrative though. )

-synthoid