Agilent is 2M for one or two channels, 1M for three or four or channels in single acquisition. Memory is not selectable. If in run/stop mode vs single, memory is further halved to 1M for two channels and 500K for four channels. If digital is added i believe it is halved yet again. so you can be down to 250k - 500k vs 10M on tek. You can get 4M option to double these .
The Agilent/Keysight scope under consideration does not reduce memory beyond 2x for analog channels. One acquisition engine is shared between channels 1&2, and another between 3&4. So, it's either 1Mpts or 2Mpts per analog channel depending which channels you have on. Each digital channel is always 1Mpts.
After looking again at the Agilent datasheet, there was some squirrelly language surrounding the digital channel numbers. I did a quick test to find out exactly what it's doing.
This is with SINGLE capture, the MEMUP option, and examining the number of points recorded for each channel via the SCPI commands. You are correct that RUN mode vs. SINGLE will halve these numbers again, but that's already a well known behavior.
Chan1 on, chan2 off, digital off: 4Mpts on chan1.
Chan1 on, chan2 on, digital off: 2Mpts each on chan1 and chan2.
Chan1 off, chan2 off, digital on: 2Mpts on Bus1 (as the digital channels are called on SCPI).
Chan1 on, chan2 off, digital on: 2Mpts on chan1, 250k on Bus1.
Chan1 on, chan2 on, digital on: 1Mpts each on chan1 and chan2, 250k on Bus1.
What surprised me was the drop to 250k digital points if *any* analog channels channels are on. And it doesn't seem to matter which or how many digital channels (D0 through D15) are on.
My apologies for the previous mis-statement, and thank you Agilent Marketing for the crisp specifications.