Cactus has summed it up well.
The fidelity and performance of most sub $500 audio interfaces is very similar.
If absolute rock-solid lowest latency performance is important, the OP should look at RME, Lynx, or MOTU.
 

A quick on-topic comment so the OP isn't tripped up should this happen. Several USB interfaces do not work well on USB 3 ports. If you get one and have issues, be sure it is plugged into a USB 2 port. If your computer has no USB 2 ports, think ahead (like purchasing from a vender with a liberal 30-day return policy). Should buy from such regardless actually.

Off topic, I happen to like "Uncle Bob."

Occide,
 
My long-winded point is that... I've personally build not a handful of machines... not 20 or 50.
I've built many hundreds... and (by default) have a lot of experience with the hardware side of things.
As a Cakewalk user, I go back to when the forums were on CompuServe.
Yes, I'm probably as old as your uncle Bob.  
 
Look, we go off on the wrong foot.  My apologies for that and taking offense to "Uncle Bob".
The forums here are by-and-large full of good folks.
We all have the same basic interests (music, technology, etc).
Welcome to the forums.
Stick around... and I'm sure we'll have many (better mannered) discussions.  
 

Here is an interesting thread related to RTL on the Reaper forums. 
This link has the Focusrite Scarlett 2i4 2gen.
Make sure to read the entire thread.
Also understand that just because an interface can produce a given measured RTL doesn't mean that it's usable or that it can perform under load. Some interfaces can and others can't. There is another thread over n GearSlutz that also measures performance under load and that's really the number that counts.
Anyway, happy reading.
 
http://forum.cockos.com/showthread.php?t=174445&page=3
 

Occide, my honest advice is wait and save. Get an RME when budget allows, anything from a FireFace UC and up is awesome, though I hear the BabyFaces are quite good too. From everything you're saying, you want solid low latency performance and you don't want to waste time "fixing problems". The only thing I could personally actually recommend with confidence given those parameters is an RME card. I've used a bunch of them and they've all been rock solid. They have best in business driver support, bringing out new drivers for devices ten years old or more. They design their own USB chips rather than relying on off-the-rack components. I have the impression that MOTU is more Mac oriented but I haven't used many of their cards. Lynx I only know by name, I don't think they're very popular over here.

You ask if we would find it reasonable that you as a hobbyist would use a $700 (or more) card. I would say if you're serious about your hobby then of course. I find it pretty much unreasonable what you're demanding of a budget interface. There is no guitar player who would expect the same kind of performance from a $200 Squier as from a $3000 Fender Custom.

I'm not pretending RME is the only good brand. I think you might even get lucky with a budget interface. But that's exactly what you'd be - lucky. I run an i7 6700 in my current system. Just yesterday, an Audient iD22 wouldn't run without clicks and pops even at 256 samples buffer size. This system is tweaked for live audio and has dpc latency values of around 30 microseconds. With my RME FireFace UCX I can run it at 48 samples (at 44.1KHz) and get 3.1ms round trip latency. Anecdotal "evidence" I know but it happened yesterday and I once again was struck with how happy I am with my FireFace.

So I spent this Sunday morning dealing with dropouts again. People been asking about a specific scenario / plugins, so here we go:
 
9 Tracks, 5x Iris 2, 2x Falcon (single instrument per instance), 1x Ample Guitar Light, 1xTAL NoiseMaker
Moderate amount of plugins per track, Neutron on every track + 1-4 Plugins from this list: MiniFilter V, BT Tempo Delay, BT Stereo Image, TAL Reverb II, OrilRiver, Spaceship Delay, BlueVerb DRV
ASIO Buffer Size: 1024
 
So far no real issues except for that spike came back on Core 0. Technically not really the first core, but the first bar in the Performance meter, set to Audio Processing. 
(Quote: Measures SONAR’s audio processing load, which is the time it takes to process a buffer full of audio data as a percentage of the maximum time available to process that data and maintain uninterrupted playback. There is one CPU meter for each core in your computer, up to 32 cores. There is some tolerance built into the meter, so it is possible that it will exceed 100% at times (more so during recording than playback).)
 
Then I started playing with the LP plugins on the Master channel. WUT, plugins on the master channel? You, the reader, have done this before, too. ;)
LP EQ, LP MB, Adaptive Limiter + GStereo + MStereoScope
LP Plugins all set to maximum quality.
 
Now the spike on Core 0 went up higher and started to peak around 70%, causing Sonar to dropout.
The same time the overal CPU load was barely noticable, between 1% and 20% per Core.
 
EDIT: Just for comparison, I'm able to run 24 Tracks of Iris 2 + 24 Tracks of Falcon + Neutron on every channel with the exact same configuration as above (except for the Master Bus), without any dropouts, CPU load about 70% higher than the above scenario.
 
I did some research and found the setting ThreadSchedulingModel should be set to "2" on 4-Core (or more) CPUs, according to various threads around here.
 
Now the spike went away, in fact the whole first bar in the Performance meter went away. Also the "main" dropout went away, therefor more issues came up. I started to get Midi-Dropouts, a lot of them. Notes just didn't play, each track swallowed some notes. My Midi-buffer was already at 1000ms, increasing it further didn't help.
 
After couple of hours of trial and error I found now with the ThreadSchedulingModel change I needed to decrease the Midi-buffer to 250ms.
 
This is exactly the kind of stuff I want to go away - it cost me so much time and kills all the fun. Basically I wanted to create some sampler instruments this morning. Instead I did nothing, but dealing with dropouts.
 
Some people around here, who have more experience with Sonar than me should sit down together and write an Official Sonar Performance Guide with FAQ section, Common Problems and Solutions Section etc. 
 
 

Jim Roseberry
My long-winded point is that... I've personally build not a handful of machines... not 20 or 50.
I've built many hundreds... and (by default) have a lot of experience with the hardware side of things.
As a Cakewalk user, I go back to when the forums were on CompuServe.
Yes, I'm probably as old as your uncle Bob.  
 
Look, we go off on the wrong foot.  My apologies for that and taking offense to "Uncle Bob".
The forums here are by-and-large full of good folks.
We all have the same basic interests (music, technology, etc).
Welcome to the forums.
Stick around... and I'm sure we'll have many (better mannered) discussions.  

Thanks for the welcome, Jim. And I want to give it back again, my apologies as well. Didn't have an exactly shiny day that day.
Looking forward to some interesting, in-depth discussion about tech with you!

Wow, some news!
 
After the day had been a waste anyway I thought "maybe I'm gonna try ASIO4All again before I buy anything.
 
The last version of ASIO4All was broken beyond repair for me. But a new version came out this May, I didn't know.
 
TL;DR with ASIO4All I can set the buffer size down to 64 (doesn't go smaller on the panel) and don't get any dropouts or issues at all. Even when I use the Realtek Audio. So the 5 years old M-Audio driver definitely was the main issue. Let's see how it turns out in the long run.
 
So if you have an old M-Audio like me, your buffersize is already at 1024 samples and you still have a ton of problems, even though your PC is fast: Try ASIO4All.

There you go folks after spending all this time reasoning and typing clear bullet proof advice and  recommendations, the answer was Asio4all all along    Are you all thinking what I'm thinking! 

Cactus Music
There you go folks after spending all this time reasoning and typing clear bullet proof advice and  recommendations, the answer was Asio4all all along    Are you all thinking what I'm thinking! 

Not really. Glad I said "Let's see how it turns out in the long run."
 
After a while some weird stuff started to show. I also made a mistake at first and adjusted the buffer size on the wrong device.

So I'm basically back to those 1024 samples but the Midi dropouts went away, at least in this particular scenario.
The onboard Realtek still performs a notch better than the M-Audio and is able to playback this with "only" 512. Therefor some paning goes bonkers now and then, not exactly sure why, but one track got louder and louder on the right channel each time I hit playback again.

Some might not believe but I do listen to any helpful advice and do my research on it. I came here for solutions, not to waste my time.