When I first entered this hobby, I started investigating the best high-end, widely available headphone at the time. My research seemed to keep bringing up the Sennheiser HD580 and then the HD600. Trouble is, those headphones had an impedance (300ohms) that far exceeded other widely-available "hi-fi" headphones available at the time - Koss, Sony, etc.
The higher impedance meant that they weren't being fed with optimal amplification when used with portable CD and tape players, due to the voltage limitations. (No one was interested in sound cards at the time.) That led Chu Moy to build the first CMoy. Soon after, Apheared built a CMoy and put it in a mint tin, and the rest is history.
Unfortunately, this same voltage limitation is part of what started this thread and causes most people in this thread to doubt the capability of sound cards directly feeding high-end headphones, such as Sennheisers.
Let's look at a few numbers ...
Here's a typical set of sound pressure values for a baseline:
Normal speaking voice: 65-70 dB
Orchestral climax: 105 dB
Live Rock music: 120 dB+
Pain Threshold: 130 dB
Jet aircraft: 140-180 dB
Now, if we look at the HD580/HD600, it has an efficiency of 97dB at 1mW, along with a maximum sound pressure capability of 116dB (
HeadWize - Technical Paper: Understanding Headphone Power Requirements by Dennis Bohn). If we're talking about peaks - not just steady-state listening levels, and the difference between listening levels and peaks are part of what define "high-fidelity", then we'd want an amplifier with the capability to utilize that peak in the Sennheiser headphone of 116dB. The reference above seems to indicate that falls just between an orchestral climax and a live rock concert, so it's not really a stretch to want to have that 116dB available - for peaks. Just figuring for those values, using Ohm's Law and the Power relationship, we have P = (V^2)/R. Or, to find the voltage needed, V = SQRT(P x R).
We start with 97dB and 1mW. Every 3dB increase is going to double the needed power. So, 100dB needs 2mW, 103dB needs 4mW, 106dB needs 8mW and 109dB needs 16mW. Let's just stop there - barely perceived above the volume level of an orchestral climax. (Note that it may still not be as great a level as a single cymbal crash transient.) The voltage needed at 109dB is SQRT(0.016 x 300), or 2.19V. Since music is reproduced in alternating voltages (sine wave), the complete voltage swing needed is +or- 2.19, or 4.38V. Note that the typical power supply for a PC is only 5V, assuming no losses. Every opamp needs a base-loaded voltage, so the sound card will probably need at least 1 or 2 volts more than the 4.38V, so the sound card's ability is already impacted.
However, if we want to utilize near full capability of the headphone, we need 64mW at 115dB! Using the equations again, that results in +or- 4.38V, or a total swing of 8.76V!
This is physically impossible for a sound card to provide! (A CMoy can easily manage it if using 2 x 9V batteries, though.) An amplifier must be able to supply this voltage, even for split-second, transient peaks or clipping/distortion and accompanying degradation in sound quality will occur.
Of course, the same situation existed in the old days with portable CD and tape players, where voltage supplies were often limited to 2 x 1.5V AA or AAA batteries, or at best - a 4.5V AC-DC adapter. Hence, Chu Moy's CMoy.
One might say why would such a high-listening level be needed? Again, one of the primary ways we define "high-fidelity" is the ability to produce a wide difference between sound levels - iow, "dynamic range." Even with the prevalent sound studio producer-compressed music that many of us listen to, a 30dB swing or more is fairly common , at least with transients. Listening at very low levels, you may be able to produce this dynamic range with a Sennheiser connected to a sound card, but you're likely to miss most of the music, too - since the Fletcher-Munson curves for the human ear have the most effect at low levels.
Current ability in amplifiers follows a similar logic, but it's a bit harder to define as easily as voltage was done above.
EDIT: This is just a simple rough estimate. Typically, RMS voltages are used to calculate power, meaning the voltage peaks must be higher than the values used above. Strictly speaking, the average power is corrected by a factor using the cosine of the angular frequency, or something similar - I'm sure you can all poke holes in the logic above, but maybe it illustrates the physical limitations of sound cards and fits in with the general theme of this thread.