With cameras that support variable offset, the offset is in ADC counts, or camera ADU. For Chris, this would be:
Signal*Gain + Offset
The offset is added to the signal during analog to digital conversion. The offset in a shallow exposure can be very easily measured in any FITS editor.
With a variable gain camera, as you increase gain, you lose dynamic range, so it becomes important to find the optimal offset for a given gain setting, to ensure you are not needlessly throwing away dynamic range. As such, it is not really ideal to use a single offset at every gain setting. Exactly how much offset you need can also change depending on what you are filtering with…NB filters might need a bit more offset, LRGB filters under light polluted skies could add a significant skyfog offset of their own in mere seconds, so a large offset may simply be unnecessary, and could be costly in terms of dynamic range at higher gain settings. Variable offset is a means of optimizing performance, and in practice we CMOS camera users don’t use a single offset for all gain settings.
Since this value is in ADU, a single value means different things for different gain settings. At Gain 0 on the ASI1600 or QHY163, the actual conversion gain is ~4.88e-/ADU. With the ASI1600, the default driver offset is 10, which can be easily measured in a bias frame as the mean will be ~10 ADU. In contrast, at unity gain, the conversion gain is ~1e-/ADU, and the default driver offset is 21. Again this can be easily measured in a bias frame. Now, an offset of 10 ADU @ Gain 0 is representative of ~48.8e-, while an offset of 21 ADU @ Gain 0 is representative of ~21e-. The default high gain (setting 300 on the ASI) offset is 50 ADU, which at that level would be about 8-9e-.
Thing is, if we used an offset of 50 ADU at all gain settings, the actual electron count of that offset will increase, dramatically. For example, at Gain 0, an offset of 50 is representative of ~244e-. Hardly necessary, when the read noise is only 3.5e- and dark current @ -20C is ~0.006e-/s. This is particularly true for LRGB exposures…I find myself using only 30 second L exposures in my back yard, which still clips stars, so gaining back every ounce of offset that I can is useful.