When it comes to audiophile performance, I consider objective tests to be absolutely superior to listening tests. Why? First of all, objective results are final and indisputable. Also, you can't do any listening tests if you don't have the sound cards in front of you! So, objective test results are the best way to compare the performance of audiophile sound cards without first buying them. Okay, let's get this ball rolling...
Note that this section might contain a lot of electronic engineering jargon, which I can't really avoid. I'll try to explain it the best I can though.
Signal To Noise Ratio (SNR)
This is the ratio of how much clear signal there is compared to noise, with the value in decibels. Noise is generally defined as unwanted random addition to the actual audio signals. The amount of noise varies according to factors like board design, parasitic capacitance and inductance, non-linearity, power supply noise, RFI, EMI and others.
A high SNR is good, because it means that ratio between the actual signal and the unwanted noise is high. For example, the noise level is 1% of the signal level. That would mean the audio signal level is far higher than that of the noise. Thus, the higher this number is (in decibels or dB), the better. Just remember that the number is negative, so -10 dB is higher than -1 dB.
Most sound cards' SNR is directly tied into the quality of their Digital-to-Analog Converter (DAC) and their implementations. Tests for this can be done in Right Mark Audio Analyzer (RMAA). Most sound card reviews should have numbers and charts for SNR as well. When you are looking at a noise chart, take a look at the amount of peaks. The fewer peaks there are in a chart, the better.
This is the range of frequencies that a sound card can produce a meaningful output within the definable range of human hearing, which is from 20 to 20,000 Hertz. The better the frequency range adheres to the range of human hearing, the better the sound card.
It should also be followed by a very sharp attenuation or cut-off point past the limit of human hearing. What this means is after 20 kHz, there should be no signal. Of course, this is usually not the case. Thus, the best sound cards have the sharpest cut-off points.
The shape of an ideal frequency response chart should be as straight as possible. There should not be any peaks or sags in the level of audio output at any frequency. Some people will try to hoodwink you by telling you that it might be Head Related Transfer Function (HRTF) which is the added factor of the human head and its effect on sound. But that's plain bullshit because HRTF compensation is not the job of the sound card, rather the job of the speaker or headphone.
While testing for frequency response using the Right Mark Audio Analyzer (RMAA) is all fine and dandy, I do recommend a spectrum analyzer for such duties, failing which an oscilloscope with an Fourier Fast Transform (FFT) analysis function is a good second choice.
Total Harmonic Distortion (THD)
If noise can be defined as random junk in a signal, then Total Harmonic Distortion (THD) can be called random junk that changes the signal's waveform shape, irrespective of whether it is within the frequency, amplitude or phase domain.
The total harmonic distortion should be as low as possible. Because distortion can be quantified as how much the signal has changed, rather than how much distortion there is (because the circuitry itself somewhat distorts the signal), THD is listed in percentage (%). A distortion of 1 % would mean that 1 percent of the total sum of the signal is distorted. Real world figures are much lower, of course. Best of all, you can get these values off the sound card manufacturer's website.