Anyway, what about noise floor of this new scope ?

It's in line with other Siglent X-series DSOs.

Here are some measurements with 1Mpts, 2GSa/s (resulting measurement bandwidth is 2kHz ~ 1GHz) at 1mV/div:

Full bandwidth (570MHz): 80µV RMS

Bandwidth limit 200MHz: 54µV RMS

Bandwidth limit 20MHz: 30µV RMS

As expected, 10 Bit Mode does not change much at high input sensitivities, because the noise from the analog frontend is dominating:

Full bandwidth (100MHz): 36µV RMS

Bandwidth limit 20MHz: 27µV RMS

See attached screenshots.

What aquisition modes are available too ?

Only Normal (aka Sample) and Peak Detect.

Average and ERES are now implemented as math functions. I'll post some demonstrations of ERES later.

Does 10 bits mode improve FFT dynamic range ?

No, It cannot. 10bit mode is accomplished by oversampling, taking advantage on the highly linear ADC.

FFT implicitely does something similar and vastly increases the dynamic range by itself, so 10bit mode is no benefit here.

Here's an example of a 30MHz carrier, 5% amplitude modulated at 100kHz. Visible (and spurious free!) dynamic range is 80dB and the peak table confirms that it is fully operational. That's better than many old swept SA boat anchors out there...

SDS2354X+_30MHz_3dBm_AM100kHz_5%