This are the rise and fall time when generating a square wave (HighZ) I haven't loaded any custom arb waves to see if they are faster.
Rise and fall time are 6.1ns - 6.4ns The specs define it as <10ns for the DG4102 and <8ns for the DG4162 so that is very good.
I didn't see a difference for small or big amplitudes.
Edit:
If you take about 0% to 100% rise time then you get about 10ns.
If we assume that the following is an acceptable square wave:10ns rise, 10ns high, 10ns fall, 10ns low... as in the picture below, then the max square wave freq is 25Mhz. (the manual is very optimistic about this max. frequency given the rise and fall time)
I don't have one so all my "knowledge" is based on what people have published here on this forum (this thread in particular) and the spec sheet and various tear-down photos and videos.
What appears to be the case is that there is a 150MHz (or perhaps 160MHz) image reject filter and for arbitrary waveforms this is all you see. For official waveforms I think there is an additional output filter which removes ringing on square waves and reduces the rise time a bit.
Amplitude wise the full 10Vpp (into 50ohms) is only available to 20MHz and the roll-off of the amplifier is such as to give a 3dB point at around 28MHz. (The gain*bandwidth product is around 240MHz so Gain is 10 to 20MHz, 4 to 60MHz, 2 to 120MHz and then the image reject filter is the limit) Rigol extends the frequency range by decreasing the gain and this switch happens at just over 4Vpp (see the curves earlier in this thread). This change is very visible with arbitrary waveforms but doesn't appear on the standard waves presumably because of the extra filtering (or perhaps instead of filtering they use sloped square waves in the arb).
Rigol have been quite clever in trading off gain with frequency. Everyone else (Agilent, TTi, Tabor etc) design for a given output (10Vpp) and then for a bandwidth at that output. The fact that with sine waves you can calibrate for the roll off and still get 1Vpp even when the roll off is 10dB allows a much higher frequency to be used.
Edit: The square waves presumably, as in the Agilent 33250 and other DDS generators, uses comparators and a generated sin wave otherwise at high frequencies there would be unacceptable jitter at in-between frequencies. (e.g with 2ns a sample point, 50MHz is 100 points but 101 points drops to approx 45MHz)