I meant I get a clean sine wave up to 841MHz.
Ah, I see. Well, that's just one point in time, and the display persistence hides the deviation of the peaks from the real value. What we're interested in is observing dynamically how the waveform becomes wobbly as the signal frequency increases and gets closer to that point.
It's clearly better than the Rigol, though, which is much worse at 1/2.5x the sampling rate. I wonder how the SDS800X HD scopes perform in this scenario.
Here's, BTW, a 500 MHz sine wave captured at 1.25 GSa/s sampling rate: