ampfixer
I should also add that the phantom power is supplied to the 4 mic pre amps in pairs. You can have it on none or any 2 or all 4. 
 

 
That is actually a nice feature for those who use ribbon mics. I have never had or needed 4, but I always assumed "all or nothing" for phantom power until you got higher up the food chain.
 

I am sorry, I stand corrected! The UR44 does allow printing EFX aside from reverb. Here it is from the owners manual:
 
http://screencast.com/t/35HAiDHHAx
 
Thank you mettelus for pointing that out!

My original intention was to get the ur44 but my dealer said i will have to wait 2 weeks. I end up with the 18i8. I only tested it for a few hours. I can say that 18i8 output is clean, clear and just NICE. It's headphone out is loud enough. A lot louder the ua25ex.

jbraner
Larry - I don't think you're going to like going from 6 ms to 12 ms. I record guitar through amp sits like you, and get 5 - 6 ms rt latency at 64 samples with a PCI card (like you). At 128 samples, it's 7-8 ms but you can hear the lag, if you hear the actual electric guitar strings being hit (vs the amp sound coming through headphones).

I'm going to bite the bullet (I hate spending this much on a audio card) and get a MOTU Ultralite AVB or RME fireface UC. Jim Roseberry says they both go down to 4.8 ms at 64 samples.

jbraner -  The MOTU looks fabulous, but it's out of my budget. I would have to stretch to get the Scarlett 18i8 at $350. Needless to say, the RME is out of the question.
 
I don't know what the cutoff would be for me in terms of latency for playing guitar and monitoring back through a host and an amp sim. I'm fine with my current card (M-Audio Audiophile 2496) with a round trip of 6.1 msec (according to Sonar), but I don't know just how much delay I can tolerate, as all my DAW experience has been with this PCI card. I think M-Audio is done writing drivers for it, so sooner rather than later I will need something new. I hope I can find something that works.

FWIW I am in a similar boat. My m-audio Profire has low latency yet I can still feel it when I play. My next interface needs to have low latency. While the MOTU Ultralight is more money than I want to spend and has more I/O than I need, it is the interface I am most likely going to upgrade to if my Profire quits working in Windows 10. The fact that Jim Roseberry is recommending it means a lot to me.
 
 

Larry, I would say it makes more sense to just get the latency as low as possible and play to the actual amp you're using.

Maximum - I feel exactly the same as you and will probably go with the Ultralite AVB in the next month or two. I'll be sure to report back - maybe in a new thread. I made sure if I buy it via Amazon that I can return it within 30 days if I can't get the low latencies that I'm hoping for.

I can get my Scarlette 6i6 down to 8.5 ms RTL @ the setting of 5ms , not sure but I think that translates into what most people call 128 samples? ) and my system is stable if the project is stripped down and I just run something like Guitar rig. And I'm using 44.1. It guess it would perform better at higher clock rates? Below 5ms I have issues with dropouts. 

Cactus Music
I can get my Scarlette 6i6 down to 8.5 ms RTL @ the setting of 5ms , not sure but I think that translates into what most people call 128 samples? ) and my system is stable if the project is stripped down and I just run something like Guitar rig. And I'm using 44.1. It guess it would perform better at higher clock rates? Below 5ms I have issues with dropouts. 

Cactus -  What the picture shows is my driver settings from the Sonar Preferences dialog. Sampling rate: 96000. (I have that same rate set in the ASIO panel, with a buffer size of 256 samples.) In case the picture doesn't appear,  the ASIO reported latencies displayed in the Sonar Driver Settings are:

• Effective latency at 96k stereo: 2.7msec
• Input: 3.1msec 299 samples
• Output: 3.0msec 284 samples
• Total roundtrip: 6.1 msec 583 samples

This gives me no dropouts or instability, and undetectable lag when monitoring my guitar through Sonar with an amp sim. My projects typically are around 30 tracks, including sampled drums, synth bass and a couple of keyboard synths plus vocals. I don't know how these settings translate to what you see in your Focusrite box. I'm just looking for a USB box that is as capable as my old inexpensive sound card.
 

mettelus
For an interface with that mix capability, I would assume you can send different mixes to different outs easily. Two headphone outs is needed for such. In Focusrite's mix control, analog 1 is different from FX1, and I would think the UR44 is similar (if not almost identical).

mettelus - You'd think so, but the Steinberg site is quite vague about exactly how this works in their interface.

mettelus
LOL, 96K is "cheating" I never record at that, but finally have the system back online so can give you quick numbers, 96K/256 buffer is 733 samples total (8.3ms), 44.1/128 is 367 (also 8.3ms), and 44.1/64 is 239 (5.4ms). It will go down to 32 "supposedly" but never tried it. As Jim Roseberry mentioned, I assume there is a "wall" at around 5ms anyway (and the CPU load is unnecessary).
 
From Johnny's thread on offsets, it seems the Focusrite products are more optimized to the 48K spec. Another thing to bear in mind is my offset is 76 samples - it seems this is "built-in" to accommodate loopback functionality inside the box.
 
I may have missed it in the thread, but which amp sims are you running?
 
Your one comment made me chuckle. Personally if it isn't broke, I don't try to fix it (but planning is good). You are "banking" on impending doom, but worry like that is best dealt with when the time comes.

mettellus - Yeah, I've been cheating since the beginning. I got into Sonar by myself, and my whole learning process was setting it up (by myself) and recording & mixing (by myself), so I had no one to tell me that 24/96 was "too much." After all, the sound card is actually called "Audiophile 2496." How was I to know? But it has worked well for me: Until I started shopping for a replacement, I didn't even know latency was an issue.
 
As for worrying too far in advance, Part B of my plan is to start using my laptop for recording on location, so that's what got me started looking for a new interface. Worrying about drivers for Windows 10 was just a fun bonus.
 
Thank you so much for all your help!
PS: I use Guitar Rig 3 (that came with an earlier version of Sonar), TH2 Producer and various guitar FX chains by Craig Anderton. I just don't want to mic an amp and make a whole bunch of noise, and most of my music doesn't really require that sound anyway.

metellus - I'm messing around with my current setup trying to increase the round trip latency and see how much it screws up my playing. It occurred to me yesterday that we are talking about milliseconds -- thousandths of a second! Intuition suggests that I may be freaking out about a time lag that I would not even notice. I mean, Sonar is showing me an RTL of 6.1 msecs and others are saying this or that device yields 11.4 msecs or whatever. So yeah, it's almost double, but we're still talking about just five thousandths of a second.

I'm messing around with my current setup trying to increase the round trip latency and see how much it screws up my playing. It occurred to me yesterday that we are talking about milliseconds

That is a really good idea!
 
I know we're talking about milliseconds - but when I'm playing my guitar quietly, through headphones - I can hear the actual electric guitar strings being plucked *and* the (delayed by a few milliseconds) sound coming out through the amp sim etc. It *is* annoying when it's more than 8-10 ms.
Try it yourself - set big buffers, to get a RT latency of 20ms or so. If that doesn't bother you - than any audio interface will be great. If it does - find out how low you need to go.
 
Please let us know how you make out ;-)

 
 
Since I finally got around to re-installing Platinum (Foxboro), I figured no time like the present to stress test the 32 sample buffer using GR5 on "HI" CPU settings, and the new "Upsampling." If I cannot break it, it is "damn good"... and well... I couldn't break it (yet). 127 sample round trip at 44.1K (is fast enough I wouldn't be able to tell if this number is BS, since I cannot hear it).
 
 
And performance from inside Platinum - GR5 CPU hit 14% max, and GR5 was upsampled (although I heard no difference)... also ran GR5's out through Focusrite's Red EQ (VST3). Went for about an hour, then got bored, but took a capture to show the comp didn't explode (no pops, crackles, or anything). This is one of only 2 audio tracks in the project, so could easily overload the 32 sample thing with more tracks/effects.
 
 
Again, this is firewire (so no potential USB conflicts) although is running through the onboard VIA chipset (folks often recommend TI only). MixControl version 3.4.
 
Edit: The upsample actually means nothing to the CPU as this was not a bounce!

mettelus

I just realized I misled you a bit before. Since I reloaded this machine there is an additional "Firewire Driver Latency" setting which defaulted to "Medium" which I since set to "Short" (the numbers I quoted above were with it set to medium). This maybe a Firewire device only feature (not sure what the USB versions of MixControl have??), and can only bet set when the device is not locked to a program.

 

 
Since I finally got around to re-installing Platinum (Foxboro), I figured no time like the present to stress test the 32 sample buffer using GR5 on "HI" CPU settings, and the new "Upsampling." If I cannot break it, it is "damn good"... and well... I couldn't break it (yet). 127 sample round trip at 44.1K (is fast enough I wouldn't be able to tell if this number is BS, since I cannot hear it).
 

 
And performance from inside Platinum - GR5 CPU hit 14% max, and GR5 was upsampled (although I heard no difference)... also ran GR5's out through Focusrite's Red EQ (VST3). Went for about an hour, then got bored, but took a capture to show the comp didn't explode (no pops, crackles, or anything). This is one of only 2 audio tracks in the project, so could easily overload the 32 sample thing with more tracks/effects.
 

 
Again, this is firewire (so no potential USB conflicts) although is running through the onboard VIA chipset (folks often recommend TI only). MixControl version 3.4.
 
Edit: The upsample actually means nothing to the CPU as this was not a bounce!

That is some low latency you got there! :-)
32 sample buffer is pretty low. I have read, hear an elsewhere, that there are internal safety buffers and some manufactures use more than others. Still, 2.9ms round trip! wow!

Yeah, I would never use that as a default setting, just wanted to try it out to see what happened. TBH my buffer is normally 96 or 128 when tracking.

Although it would be interesting to find where the "belly up" point occurred. Also I am curious if the loopback offset changes.

Again... This is only one part of the entire system from string attack to ear drum (take it with a grain of salt)!

I circled back on this and found something "odd" using CEntrance's ASIO Latency Tester from this thread
(is a neat thread to read for loopback latency/offset).
 
Bottom line, I started of lazy and just tried the latency tester with 32 sample buffer... first offset was 32, then 52, then 206 and seemed to "settle" there. Since this is almost 3 times what I know it is at 64/128 (76), I then reset the interface to 128 samples and the Latency Tester jumped into the 400's.
 
Since I know this is not true, I closed out everything, set my AI back to 32 samples and jumped into SPlat. I still had the loopback project I created, so simply ran the test again, and hit perfect on tracking with the 76 sample offset already enabled (my prefs are set there now). So the offset didn't change with the 32 sample buffer (external loopback is 76 samples, internal loopback is 6 samples).
 
Be advised... if doing a physical loopback test (cables from AI outputs to AI inputs), be sure to zero inputs and outputs before moving any cables and MUTE the track to be recorded before arming it (and leave it muted until the track is unarmed)!!! Do not willy-nilly test this without understanding how to protect your gear.