What are you going to do when the ripple is even smaller? For example, here is a ~150uVrms 10MHz signal being clearly triggered and displayed. This signal was displayable--triggerable and above the noise threshold on a Tek 2465B (shown), a Tek 2221A (digital and analog) and the Siglent 1104X-E (although just barely and not as reliably). I was able to do the same thing with a 1MHz and 100Hz signal of approximately the same amplitude. What would it look like on the Rigol?
You can't average your way out of this when you are looking for noise--possibly non-periodic--in the first place. On the Tek 2221A and the Sig 1104X-E, averaging made the signal look nicer but I'm not convinced that means better.
I have run across that problem a couple times, where only a lower noise oscilloscope would do when looking for subtle anomalies. The usual solution is to use averaging or high resolution mode and trigger off of a different channel from a signal which is synchronous to the signal of interest, but this is not always possible. See below about differential probes.
My 10 bit digitizing but not a DSO Tektronix 400 MHz 7854 can reveal signals that my 40 uV RMS over 100 MHz 2232 or 2230 cannot even see in digital storage mode, and do it with an RF sampling front end if needed, but an equivalent instrument today would be 10s of thousands of dollars and we are not discussing those.
Which initiates another question; what would the MSO5000 sampling drop to if one more channel was activated ?
Lower cost DSOs do that because they either have only one bank of acquisition memory with limited bandwidth shared between 4 digitizers, or because the digitizer is interleaved between the 4 channels. The later is practically universal in lower cost DSOs.
Either can be seen as a cost saving measure or a way to maximize performance when only a single channel is used, and more expensive instruments may use completely separate digitizing and storage for each channel so sample rate, and usually record length, does not depend on the number of channels. Tektronix used to refer to a DSO as being "real time" if its maximum sample rate, and record length, did not change with the number of channels and for many years they maintained a separate line of DSOs which worked this way because some applications demand it.
Also you keep repeating about some magical amplifiers. Amplifiers that have DC-100 Mhz bandwith and less noise as even a little Micsig or Siglent SDS1104X-E cost as much as a good scope from Keysight.
Yes, but amplifiers from DC to 1MHz are incredibly cheap (ie. a $2 OP-amp plus power supply) and would be perfectly adequate for audio work and looking at power supply ripple.
Amplifiers from 10kHz to 2GHz are also incredibly cheap.
What is not cheap or easy are getting good flatness and settling time over the bandwidth of interest. However the high noise is a problem of design and not cost. Up to 350 MHz, there is no excuse for such a noisy implementation and obviously Siglent is doing something right that Rigol is not. The noise on the Siglent is high compared to what it could be, but that makes the noise on the Rigol much worse than high.
In the end neither is suitable for looking at 350MHz signals using 4 channels. The Siglent SDS2k due to low samplerate, the Rigol MSO5000 due to excessive noise. Also note what David Hess wrote: Rigol typically performs math on decimated data which can give the wrong results when doing measurements on noise.
It is not the decimated data which is the problem but performing math on the display record which has already been processed for the display. More expensive instruments maintain a separate full resolution record for processing what is essentially the "raw" data, that is separate from the processed display record.
Probing a 350MHz signal with passive probes is also a minefield, the artifacts from the probe will usually be bigger than the signal.
Passive high impedance probes are more difficult to use at higher frequencies than active probes, but if this is taken into account, the usual problem is noise from the ground loop with a singled ended probe, which even an active probe does not solve. A differential probe solves this but at the expense of greater noise, and this tradeoff is almost always worth it if their higher cost can be accepted. Several times I have probed signals approaching the limit of the input noise of my oscilloscope where differential probing was a solution because it removed common mode noise.
I wouldn't buy either of these for the bandwidth, I'd buy them for the big touch screens, large memory, etc.
I was thinking that about the Rigol MSO5000 series before I saw how much better the Siglent is for noise. They are both noisier than the general state of the art of more than 2 decades ago, but for most applications that is good enough. By preference I do not even use the lowest noise oscilloscope that I have available. Fiorenzo did the right thing by inquiring.