I just obtained a 490EX. I was a bit puzzled about the claimed that no darks were needed. So I sat down and made some measurements on various duration darks as well as collecting a bunch of bias frames and making masters.
What I found is that the claims are correct - you should not do dark frame subtractions -- because they are apparently already doing that on the sensor. Every one of my dark frames were within 1 e- of the master bias frame.
I fed the frames into my script in PixInsight that can characterize the sensor. What it shows is that there are very close to 6 ADU per e-, that the effective dark current is vanishingly small (see reason from above), the full-well capacity is on the order of 11K e-, and that you should get peak SNR of around 33 dB.
But I did find that the variance in the dark frames grows with time, just like you'd expect. I see the standard deviation in the darks ranging from around 35 ADU at 1 sec, and rising to around 113 ADU at 100 sec. But the mean and median counts remain astonishingly constant at around 360 ADU = 60 e-. Typical readout noise in the master bias is as they claim at 5 e-.
(It seems impossible to find any solid technical details on these sensors.)
What this all translates to is that you should subtract a master bias frame from each new image, then flat field correct the image.
Don't bother subtracting dark fields because the sensor is already doing that on the fly for you. Apparently there are some masked cells on the sensor that accumulate dark current charge, and some measure of that charge (mean? median?) is being subtracted from each pixel value before being sent out the A/D to your computer. Doing a dark subtraction yourself only compounds the image noise.
:-) DM