John Atwood was recently able to buy a tube of TI/Burr-Brown PCM-1704U-K's for $75 each. Oddly enough, the super-grade K series are more available than the lower grades. It's a good question whether or not the PCM1704's you can buy are actually *current* production - maybe TI only wakes up the line when demand builds up enough to justify a product run. (That's the way audiophile caps are made, by the way - they're simply a special, low-volume run made with different films.)
Renee's description of the 20400A lines up with everything I've heard: they're custom-trimmed PCM63's, which offer the unusual option of external trim of the LSB's. Renee, it's the lowest two bits of the 20-bit converter, isn't it?
One difference for the passive I/V conversion crowd is the PCM63 runs at twice the current of the PCM1702 and PCM1704, so a PCM63 can be used with a 100 ohm resistor without turning on the internal diode, while the PCM1702 and 1704 require 50 ohms (preferably a bit less). I do remember measuring the PCM63 and finding that yes, 100 ohms is fine, while 105 ohms you start to see the very first signs of little distortion spurs emerging from the noise floor. With the spectrum analyzers that Matt Kamna had at the time, we never saw any difference in distortion between 10, 20, 50, or 100 ohms.
Regarding the Karna amplifier ... yes, I've been occasionally tempted by direct-heated input tubes, since the already high-performance 5687/6900/7119 series is the only 5th-and-above source of harmonics in the entire amplifier. The PP 45 section is almost frightening to measure: Gary Pimm and I only saw 2nd and 3rd, with the instrumentation noise level at something like -130 or -140 dB. Gary's test setup is pretty exotic, and uses custom-built test fixtures to achieve the astonishing dynamic range. The PP 300B's are a little dirtier, with the 5th way down at something like -80 dB, and the 6th and 7th something like -90 to -110 dB. When I say the Class A PP 45 and 300B are low distortion, I'm not joking.
Old-stock 2A3's were not so great, with 5th around in the -60 to -80 dB range, but this is very vendor-specific, with a remarkable 20 dB spread between brands. To our surprise, varying the operating current only made a 6 dB change in the high-order distortion spectrum, and sample variation was in the same range as well. It was the *vendor* that made the difference. Obviously the difference was in internal filament and grid structure, and most likely the result of the jig used in manufacturing. (I don't have the nerve to try cryo-treating a vacuum-tube, since I'm concerned about damage to the glass-metal seal over time.)
All of the indirect-heated triodes were another step down, with the best we could find somewhere around -60 to -70 dB for the 5th-and-up harmonics. We tried a *lot* of these things, and once again, it was the vendor that really mattered.
Note that the levels of 2nd and 3rd were all pretty similar, falling in a 6 dB range, with high-order harmonics what set them apart. The test circuit was the same as a Karna stage: transformer in, transformer out, all Class A PP, with cathodes bypassed. This was the lowest distortion configuration of any. We tried a current-source cathode (forcing the tubes into exact match), and high-order harmonics skyrocketed. John Atwood, using his Audio Precision setup, has confirmed the same result: a high-impedance shared-cathode circuit lowers 2nd harmonic (which is nearly inaudible) at the expense of high-order harmonics; peak current delivery also goes down by a factor of 2 or more.
If you're really clever, John found an interesting result: there is an optimum shared-cathode dynamic impedance (which is not the same as the cathode resistor) for the lowest possible upper-harmonic distortion. In practice, for a circuit with a shared cathode resistor and a cap bypass, you insert a pot in series with the bypass cap, and adjust the pot while looking at the spectrum analyzer, and look at high-order distortion at the -3 dB and -30 dB modulation levels.
At the 2004 European Triode Festival, the audience got to hear what addition of 2nd harmonic, 3rd harmonic, and 2+3rd harmonic sounded like on a (very) high-quality system with musical selections (the distortion was generated by computer algorithm, as I recall). I was surprised just how hard it is to hear 2nd harmonic: for me, the threshold level was around 1% or so, and sounded kind of like an old jukebox, a warm syrupy sound, like you'd expect. 3rd by itself was audible around 0.3%, and sounded kind of rough, like a trumpet blat, but not too awful. The combination of 2nd and 3rd in a 3:1 ratio sounded surprisingly like the undistorted original, but a bit more rich, more dense, a little thicker, if you will. So these two harmonics, by themselves, are pretty innocuous, and the higher-order harmonics are much more unnatural and "electronic" sounding.
When designers go to extreme lengths to eliminate 2nd harmonics, they are barking up the wrong tree, since it is so hard to hear. Reducing harmonics from 5th through 9th and higher, though, is really worthwhile, since it has a big impact on IM distortion.
