Page 18 of the owner's manual https://usa.yamaha.com/files/download/other_assets/7/1180527/cp88_cp73_en_om_d0.pdf

NOTE
• Audio data sending capability for the instrument is a maximum
two channels (one stereo channel) at a sampling rate of
44.1 kHz.

Page 27

USB AUDIO
USB Audio receiving/sending capability for the
instrument is a maximum 2 channels (1 stereo channel)
at a sampling rate of 44.1 kHz. The input signal from the
USB [TO HOST] terminal is output via the OUTPUT
[L]/[R] jacks (XLR type connector), the OUTPUT [L/
MONO]/[R] jacks and the [PHONES] jack.
The input level can be adjusted from the [MENU] button
→ “General → “USB Audio Volume.”
The audio signal output via the USB [TO HOST]
terminal is identical to the audio signal output via the
OUTPUT [L]/[R] jacks (XLR type connector), the
OUTPUT [L/MONO]/[R] jacks, and the [PHONES]
jack.

The supplemental manual has no additional information regarding frequency - so this is the spec.

The Roland RD-2000 is 24-bit/192kHz. I don't see other similar specs from Casio or Korg - so assuming perhaps sampling rate via USB is not a marketing bullet (perhaps standard CD quality).

I'm not sure higher frequency translates to any significant impact on latency because before you have audio data - you have the physical and protocol layers of USB (2.0) which present the lions-share of latency before you get to how many samples your DAC has to translate per second. Audio quality going from 44.1kHz to your ears isn't any worse than 192kHz to your ears - 44.1kHz is way above the noise threshold. Where you would gain benefit is in having more resolution if you're going to post-process - which may be your situation. I don't know exactly what you're doing - even if you're not post-processing, you may just "feel" better about 192kHz - I won't argue with your rationale.

Some reading material: http://lavryengineering.com/pdfs/lavry-sampling-theory.pdf

Thanks Jason.

I don"t think it is up to us to debate whether high quality digital to analog conversion (DAC) is meaningful or not. I believe it is and I am probably not the only one..

For my scenario, it is essential. I dont do post-processing or record, I simply play VSTs and listen.

Higher frequency conversion, such as 192k vs 44k, indeed has a BIG impact on latency. When the computer and interface negotiate at 44k, processing the sample blocks of the VST takes 4 times as long compared with 192k. The cost of faster processing is CPU power. The gain is lower latency. For VST-piano playing (note: NOT post-processing, recording, etc) the rule of thumb is that you need to stay below 5ms latency. Thats pretty much impossible with a 44k audio interface.

Christoph,
It looks like you answered your own question.

Lower sample rates leading to higher CPU utilization is a new one on me. Seems backwards. In general the rule is opposite - higher sample rate is more data to shuffle and thereby more taxing on the system. That's not to say you can't drive a car in reverse on the freeway. Some VSTs may break this rule if they simply cannot natively downshift requiring something external to convert the sample rate. Would be interesting to learn more about this.

A 96 sample buffer on my 44k setup over USB 2.0, Cubase reports 5.011 mS latency. I can get down to 2.562 mS at the lowest buffer size. That doesn't include the VST - somewhat of a limitation in information reporting in Cubase 9.5. I'll need to upgrade to Cubase 10 for the VST latency monitoring now available.

Not doubting your setup may be unique and buck the trend.