Possible to play @ 2ms latency?

The Pops worked out well for Arthur Fiedler?

It's possible, but you need something with uber-efficient drivers and architecture like those from Lynx and RME. I haven't tested RME in a while, but my Lynx will do the tightest latencies I've ever seen, even slightly better in some apps than the RME PCI stuff. But there are other factors that affect the performance so you have to configure your machine as well as possible. With your motherboard/CPU/chipset/OS combination I can't say for sure if Sonar will let you get down to a reliable 2ms even with a Lynx. I think it's totally possible, as I can go lower than that when needed, but the CPU load goes up considerably.

Anyway, the best you can do is get a Lynx or RME in terms of bus and driver performance, so if you can't do it with those, then you probably can't do it reliably on your system.

To give you an idea, I can easily use 128 sample buffer (2.9ms latency at 44.1KHz) in Sonar, but there begins a slight performance hit. I usually run at 128 or 256. And most cards should be able to handle 128 sample buffer okay. I can also bump it down to 64 samples (1.45ms) but the CPU load begins to ramp up considerably, even on a Core2Duo or Quad CPU. But still basically usable. SOME cards can pull off 64 samples, but not very reliably under load. The Lynx driver even lets me hit 32 samples (about 0.7ms!!!!!!!) but unfortunately it's totally impractical in Sonar. Sonar (and a number of other DAW apps) don't seem to like that level of latency, and as of the last time I tested it, it's a no-go for Sonar 7 at 0.7ms. However, there are a few other apps that I've used that WILL actually allow me to go that low.... so I have successfully run at 0.7ms latency in other apps. Imagine that... that even the ROUND TRIP latency is less than 2ms! Amazing. But no matter what, the CPU load is pretty hefty at that level, and it really almost defeats all the power of a quad-core, for example.

Also keep in mind that only PCI/PCIe can go this low... even the RME proprietary firewire interface has a safety buffer that adds a small delay, which people don't seem to realize. And all other firewire devices are worse, even though their drivers may not report this extra safety buffer, it's measurable. So you CANNOT trust reported latency from any firewire or USB audio devices. Delve into the RME manual and you'll see what I mean -- they're the only ones that actually tell you how big their firewire safety buffer is. :) So if you want the most screaming latency performance, you need to stick with PCI/PCIe. And only PCI/PCIe versions of the Lynx and RME products can reliably and consistently handle 64 or 32 samples (without safety buffers), in my experience (true 1.45 or 0.7ms latency). Various tests and DAW benchmarks for various DAWs out there support that as well, although there are exceptions.

Anyway, good luck! Most cards should be able to handle around 128 samples (2.9ms) reasonably well, and like I mentioned, some can handle around 64 so-so. Your mileage will vary. Good luck!

ORIGINAL: eratu

It's possible, but you need something with uber-efficient drivers and architecture like those from Lynx and RME. I haven't tested RME in a while, but my Lynx will do the tightest latencies I've ever seen, even slightly better in some apps than the RME PCI stuff. But there are other factors that affect the performance so you have to configure your machine as well as possible. With your motherboard/CPU/chipset/OS combination I can't say for sure if Sonar will let you get down to a reliable 2ms even with a Lynx. I think it's totally possible, as I can go lower than that when needed, but the CPU load goes up considerably.

Anyway, the best you can do is get a Lynx or RME in terms of bus and driver performance, so if you can't do it with those, then you probably can't do it reliably on your system.

To give you an idea, I can easily use 128 sample buffer (2.9ms latency at 44.1KHz) in Sonar, but there begins a slight performance hit. I usually run at 128 or 256. And most cards should be able to handle 128 sample buffer okay. I can also bump it down to 64 samples (1.45ms) but the CPU load begins to ramp up considerably, even on a Core2Duo or Quad CPU. But still basically usable. SOME cards can pull off 64 samples, but not very reliably under load. The Lynx driver even lets me hit 32 samples (about 0.7ms!!!!!!!) but unfortunately it's totally impractical in Sonar. Sonar (and a number of other DAW apps) don't seem to like that level of latency, and as of the last time I tested it, it's a no-go for Sonar 7 at 0.7ms. However, there are a few other apps that I've used that WILL actually allow me to go that low.... so I have successfully run at 0.7ms latency in other apps. Imagine that... that even the ROUND TRIP latency is less than 2ms! Amazing. But no matter what, the CPU load is pretty hefty at that level, and it really almost defeats all the power of a quad-core, for example.

Also keep in mind that only PCI/PCIe can go this low... even the RME proprietary firewire interface has a safety buffer that adds a small delay, which people don't seem to realize. And all other firewire devices are worse, even though their drivers may not report this extra safety buffer, it's measurable. So you CANNOT trust reported latency from any firewire or USB audio devices. Delve into the RME manual and you'll see what I mean -- they're the only ones that actually tell you how big their firewire safety buffer is. :) So if you want the most screaming latency performance, you need to stick with PCI/PCIe. And only PCI/PCIe versions of the Lynx and RME products can reliably and consistently handle 64 or 32 samples (without safety buffers), in my experience (true 1.45 or 0.7ms latency). Various tests and DAW benchmarks for various DAWs out there support that as well, although there are exceptions.

Anyway, good luck! Most cards should be able to handle around 128 samples (2.9ms) reasonably well, and like I mentioned, some can handle around 64 so-so. Your mileage will vary. Good luck!

my old maudio delta used to go down to 64 samples, not sure how it performed under load though. my EMU1820m which i use now goes down to 88 samples (lowest setting possible), which equates to 2ms. depending on the type of plug ins, I may or may not get artifacts under load.

There's nothing wrong with reaching for perfection or the ultimate, but isn't 2ms latency like standing two feet from the loudspeaker, and 4 ms latency like standing four feet from the loudspeaker. Does it really affect your musical performance? Or is it so that the above mentioned latencies caused by buffers in FW hardware could add the latency so much, that you actually need to set the sequencer to 2 ms to stay, say, below 10 ms in total?

ORIGINAL: Jim Roseberry

The 0404 is USB...

assuming he has one of the new ones, yes. the original 0404 was pci.

ORIGINAL: mgh

use a mixer and get near zero-latency hardware monitoring...

Agreed (again)...

ORIGINAL: Jim Roseberry

use a mixer and get near zero-latency hardware monitoring...

Absolutely... unless the OP wants to monitor in realtime thru software EFX.
ie: Playing in realtime thru an AmpSim plugin
Low round-trip latency is critical for this application.

You either have to buy a receptor or use another PC for realtime monitoring of softsynths and FX.

One thing that would be nice, I dont know if it's possible as I'm not a programmer at all, is to be able to have two latency settings, one for real time monitoring and another for playback and choose wich processor(s) will be real time. Most of the time one 2Mhz processor will play one or two soft synths without problems so why not use one processor for synths and three for SONAR.

I know it sound's cranky but with my old PC (Single 2Mhz AMD) and an audiotrack Maya (PCI) I was able to use WDM drivers at 130ms in SONAR (I think it was 2 or 3) and at the same time ASIO drivers at 2ms using a standalone soft synth or Guitar Rig. I tested it on my friend's PC with his MOTU 828 and it WORKED. Now with SONAR 7 it seems broken but I have no need for it because I bought a laptop and use it as my realtime station.

i play in real-time all the time using NI GR2 with my echo audiofire 2, using a mixer, with input monitoring turned on, at latency of around 4ms, which i guess will be around 10ms in real-time, and it's fine...is this cos 10ms don't matter in the real world or something else/??

Total perceived latency is much higher than reported latency when input monitoring. This thread gives more info, especially the post at the top of the page: http://forum.cakewalk.com/tm.asp?m=927564&mpage=4

When input monitoring the reported latency doubles as sound comes in and then goes out. And there are other small "hidden" buffers that don't get reported in the latency setting. Any plugin used in the chain can potentially increase latency. Any plugin with look-ahead settings (ex. gating) or its own latency settings (ex PerfectSpace) can add huge amounts of latency.

Also, using headphones for monitoring can make a very noticeable difference in perceived latency assuming you usually sit more than a foot or two from your speakers.

ORIGINAL: Ognis

#0 0x900138ac in close ()
#1 0x0036120c in my_close (fd=7, MyFlags=0) at my_open.c:92
#2 0x000d72f8 in MYSQL_LOG::close (this=0x575a10, exiting=1) at log.cc:1925
#3 0x000d9004 in MYSQL_LOG::cleanup (this=0x575a10) at log.cc:158
#4 0x000d9064 in MYSQL_LOG::~MYSQL_LOG (this=0x0) at log.cc:147
#5 0x000d90e4 in __static_initialization_and_destruction_0 (__initialize_p=0,
__priority=0) at log.cc:37
#6 0x000d9144 in global destructors keyed to mysql_log () at log.cc:2379
#7 0x8fe109dc in
__dyld__ZN16ImageLoaderMachO13doTerminationERKN11ImageLoader11LinkContextE ()
#8 0x8fe038ec in __dyld__ZN4dyld14runTerminatorsEv ()
#9 0x90013a5c in __cxa_finalize ()
#10 0x90013904 in exit ()
#11 0x901209a4 in abort ()
#12 0x00397e78 in __eprintf () at TCP_Transporter.cpp:289
#13 0x00046600 in subselect_single_select_engine::cols (this=0x0) at
item_subselect.cc:1579
#14 0x000456f0 in Item_singlerow_subselect::check_cols (this=0x0, c=1) at
item_subselect.cc:423
#15 0x0009d2f4 in yyparse (yythd=0x384f818) at sql_yacc.yy:4718
#16 0x00089f6c in mysql_parse (thd=0x384f818, inBuf=0x384f854 "", length=79) at
sql_parse.cc:4318
#17 0x0008a6a8 in dispatch_command (command=COM_QUERY, thd=0x384f818, packet=0x3089019
"purge master logs before (select adddate(current_timestamp(), interval -4\nday))",
packet_length=59058295) at sql_parse.cc:1499
#18 0x0008b684 in do_command (thd=0x384f818) at sql_parse.cc:1312
#19 0x0008c1d8 in handle_one_connection (arg=0x384f818) at sql_parse.cc:1044
#20 0x9002c3b4 in _pthread_body ()

Kids say the darndest things!!

Sven

ORIGINAL: Mick

How many hard drives do you have? Do you have your audio on a different drive than the system files?

You know that that is extremely irrelevant to the topic problem, right?

Isn't 4ms latency comparable to standing 4 feet from your amp?

You're right. You can't hear a 4ms delay. Exactly what the threshold of perception is depends on what you're listening to, but it's typically somewhere around 20ms, give or take.

Seems like such a short delay, but anyone who's played on a large stage knows the feeling when the band just isn't as tight as it should be. If the stage is 30 feet wide and you're the keyboard player stage left trying to sync to the guitar player stage right, and you don't have your respective instruments going through the monitors, you're hearing each other's notes delayed almost 30ms. That's enough to make everything sound loose and sloppy. I have experienced that, and learned long ago to set up tight onstage even if there's plenty of room.

Drummers are often more sensitive to the effect, certainly more so than bass players. I don't know if it's due to the percussive nature of the instruments, or just that drummers are more tuned in to timing. But I know that trying to "play drums" on a keyboard with a soft synth is an exercise in frustration, even with short latencies.

But here's the real gotcha. The latency you see in your ASIO control panel is just a calculated time based on the size of the buffer and the sample rate. It is only part of the whole latency picture, since there are other components that add to the total amount of time it takes to process the audio and send it back out.

Because I primarily use hardware synths, I enjoy the luxury of "zero-latency" monitoring. Consequently, I don't fret over latency like I would if I were using only soft synths or monitoring with software effects. And a good thing I don't, because even though my ASIO buffers are set for a conservative 8.7ms latency, the actual round-trip latency is nearly 20ms! And that's a long enough lag that you can definitely hear.


EDIT:
Sorry, I got a little OT there. The answer to the OP's initial question "is 2ms latency possible?" is probably "no", at least not with any hardware that we regular people can afford.

Would the latency be doubled when using softsynths, or only audio effects? In softsynths I would think since the input is MIDI only there would be minimal latency on that part of the round trip.

I used Centrance software last night for the first time. ( thanks Jim Roseberry for the recommendation thru this forum ) I have a Edirol FA-66. It wasn't successful for me at first either likely because the Edirol requires that all apps be shut down and the unit power cycled to reset it's buffer settings. ( I got the same msgs ) It took me maybe 6 trys to get the sequence right where it would measure correctly. I don't know what your interface requires to reset in ASIO mode but I did finally get a round trip of 4.23 and 6.35 ms through mine, depending on the buffer setting. It does work.

Jim,

Got centrace to work.

Chose:
ASIO FW-1884
64 Samples (1.45 ms)
44100 Hz

Results: 65 Samples / 1.47 ms

If I hit the Measure button multiple times I can see as low as 1.45 and as high as 2.18 (after about 50 tries).

I expected 5-11 as you previously suggested. Did I do something wrong or do I have the greatest round-trip-latency ASIO machine available (lol)?

P.S. Using ASIO @ 64 sample buffers I have not had an audio engine halt after about an hour of idle time periodically minimizing and maximizing Sonar.

So far so good. Can't wait to install Q6600/Zalman/Antec 650w PSU.