USB 'vs' Firewire 'vs' PCI ...... Best Round Trip Latency .... ?

Hi Timboe,
 
 
The MOTU 8-Pre delivers 5.35ms total round-trip latency at a 64-sample ASIO buffer size/44.1k
The RME Fireface 400/800/UC deliver 6ms total round-trip latency at a 64-sample ASIO buffer size/44.1k
The M-Audio FastTrack Ultra 8R delivers 5.5ms total round-trip latency at those same settings.
 
All the above are great choices...
I wouldn't have believed the FastTrack Ultra 8R's performance had I not seen/measured/experienced it first hand.    It's a great little unit at $499 (has rock-solid Win7_x64 drivers)

Timboe, great advice from Jim Roseberry, as always. However, I think it is important to point out that in a mobile rig (a laptop, I assume?), the interface itself will not be the only limiting factor.

Your current rig specifies a "4.2 GHz" i7 processor. Unless that's a typo, you're running an overclocked CPU which requires advanced cooling for stable/reliable operation. The 4.2GHz is also why you are able to achieve such outstanding roundtrip latency results -- and most likely not available in a laptop (maybe Jim can help clarify).

Therefore, my best advice to achieve the absolute lowest levels of roundtrip latency in a mobile platform would be to opt for a rackmountable PC over a laptop. You may even consider converting your current rig into a rackmount case if you haven't already.

Houndawg


Timboe, great advice from Jim Roseberry, as always. However, I think it is important to point out that in a mobile rig (a laptop, I assume?), the interface itself will not be the only limiting factor.

Your current rig specifies a "4.2 GHz" i7 processor. Unless that's a typo, you're running an overclocked CPU which requires advanced cooling for stable/reliable operation. The 4.2GHz is also why you are able to achieve such outstanding roundtrip latency results -- and most likely not available in a laptop (maybe Jim can help clarify).

Therefore, my best advice to achieve the absolute lowest levels of roundtrip latency in a mobile platform would be to opt for a rackmountable PC over a laptop. You may even consider converting your current rig into a rackmount case if you haven't already.

I am almost shure "muscle" has nothing to do with roundtrip latency or simply" latency". Muscle WILL determine how many tracks/ Softsynths/ Effects you can run at your lowest latency settings though.

Muscle= CPU, Memory performance and quantity, HD speed, 2nd HD or not etc....

Windows drivers behaving badly and bad device drivers/ poor DPC latency performance will be THE limiting factor and determi!ne if you can run such low buffer sizes at all.

I remember running an AMD Athlon XP 2800 on a Nvidea chipset/ Gforce video card/ 1 gig ram, Windows XP and having very good DPC latency results (7usec, no pikes) and using a terratec MT88 interface at its lowest buffer settings (32) with NO PROBLEMS at all . I couldn't run many tracks though :-), not at that latency settings.

I am SHURE it had very good roundtrip latency if I would have measured it. It felt incredibly responsive at these settings, but as I said not for that many tracks/ synths etc...

So, there is no reason why a good "audio laptop" with good DPC latency results would have worse roundtrip latency with these same interfaces. You WILL bring the laptop sooner to his knees then an equally priced desktop (audio- ready) when you start adding tracks / softsynths/ effects...

I am almost shure "muscle" has nothing to do with roundtrip latency or simply" latency". Muscle WILL determine how many tracks/ Softsynths/ Effects you can run at your lowest latency settings though.

 
I think we're all saying the same basic thing...   (just different ways)
 
I'll reword to clarify:
The audio interface will determine the lowest round-trip latency you can achieve...
Provided... that your machine can keep up with the associated CPU load.
ie:  Using a FW1814, you'll never achieve 5.5ms round-trip latency at a 64-sample ASIO buffer size/44.1k.
Doesn't matter what machine it's connected to... (as the driver uses a large hidden safety buffer)
The 8-Pre and FastTrack Ultra 8R will deliver total round-trip latency of 5.5ms at those same settings
(if your CPU can handle the load).  With the memory controller on the CPU itself, the i7 is a particularly good choice for working at ultra low latency settings.
 
FWIW, My favorite live rig build (currently) is a Cube.
I connect a touch-screen monitor (so I don't have to tote keyboard/mouse)
This gives me loads of speed/power... it's quiet... and it uses standard parts should something fail (quick/easy/inexpensive fix).  I put a pair of 1TB HDs in RAID-0 for streaming samples (sustains 260MB/Sec).  This allows me to run complex splits/layers of the most taxing VST instruments... with loads of polyphony.

Out of personal curiosity...... I've only ever had a PCI interface (Aardvark Q-10), I'm thinking of getting an Echo Audiofire 12, anyone have any idea what kind of latency it has?
Also, I've only ever used WDM drivers (It's what's always worked best for me with Sonar). Are there significant differences between WDM and ASIO?

Will the new USB 3 make any difference?

No. USB2 is already many, many times faster than necessary for digital audio. Making it faster won't reduce latency, since audio data comes along at a constant rate regardless of the speed of the transport mechanism. USB disk drives will probably benefit, though.

timboe


Hi all 
  
I am a long-time  PCI  Audio Interface user - Layla 3G.   I am  *thinking*  of  going more-mobile  and hence therefore I am looking at  USB  or  Firewire interface.
  
Low latency is  very  important to me and my  " real-world-round-trip "  CEntrance  figures are as follows:- 
 
- Intel 4.2gig i7 
- Windows 7 X64 
- Sonar 8.5.3 X64 
- Echoaudio  8.5  Windows 7 X64 Driver 
- ASIO Mode 
- Echo Layla 3G PCI 
  
CEntrance Figures are:- 
  
44.1k / 24 Bit / 32 Samples 
  
-24db     154 Samples / 3.49ms 
-18db     155 Samples / 3.51ms 
-12db     155 Samples / 3.51ms 
-6db       185 Samples / 4.20ms 
  
  
44.1k / 24 Bit / 64 Samples 
  
-24db     218 Samples / 4.94ms 
-18db     219 Samples / 4.97ms 
-12db     219 Samples / 4.97ms 
-6db       281 Samples / 6.37ms 
  
In short then ....... is there any currently available  USB  or  Firewire  Interface  that  comes  close  to the above  " real-world-round-trip "  CEntrance figures  ? 
  
All suggested hardware /   CEntrance  Test Result figures would be hugely appreciated. 
  
Thanks, 
timboe

I compared my latency in a 32bit and 64bit environment ... they were identicle. I was kind of expecting better latency with 64bit. I'm hitting 5.8ms @ 128samples round trip and can't complain about that.
 
A little off topic ... Don't forget about recording latency also. I calibrated and had to adjust by 23 samples ... not bad at all and really unoticable but I did it anyway. There's a tutorial on how to do the calibration along with a test .wav by Danni Danzi I think somewhere on the forums or net somewhere. Can't quite remember where I saw it.

papa2005


Hugo,

Just for the record (and because I think you're a nice guy) "SHURE" is a brand of microphones..."Sure" is a word that means "certain"...

Ah,

Thanks, I will try to remember :-). English is not my native language (Dutch is).

Hugo

This is a long shot - but my Layla 3G is rock solid at 32 samples. Do any of the interfaces you mentioned run solidly / reliably at 32 samples ? Also Jim - when I ran CEntrance, the results at -24db / -18db / -12db were essentially identical but once I went to the -6db setting, the round trip latency spiked up condideraby buy a good ~1.0ms / ~1.5ms ..... ? Any idea why this is the case ? Are your results above "an average" or are they at a specific -db setting ?

 
Hi Timboe,
 
Few audio interfaces provide a 32-sample ASIO buffer size. (as you know)
I'd like to see that change... as it's gotten to the point where a fast i7 based DAW can work very effectively at a 32-sample ASIO buffer size.
 
MOTU had a 32-sample buffer size in their 32Bit driver (back prior to the mkIII units).
RME FireFace 400/800/UC have a 48-sample buffer size - absolutely rock-solid
Steinberg MR816 units have a 32-sample ASIO buffer size (but the unit uses a large/hidden safety buffer).  It's absolutely rock-solid, but RTL is similar to the other units running at a 64-sample buffer.
The M-Audio units don't provide a 32-sample ASIO buffer size.  Shame as their drivers are rock-solid...
The Lynx units provide a 32-sample ASIO buffer size.
The MOTU PCI/e units provide a 32-sample ASIO buffer size.
 
Regarding the dB level affecting latency:
I've seen a very small amount of fluxuation... but I've not encountered 1.5ms difference.