it should be pretty obvious that a later sensor design from same manufacturer with substantially larger pixels and dual-ISO will certainly perform better
Thanks! I think a number of factors are being conflated or overlooked here. A bit of disassembly might be useful. To back up somewhat, this is probably a comparison between what seems most likely a reuse of some of the
IMX294 masks (GH5S and BHG1) and the
IMX272 (GH5, GH9). So far as I know the relationship between 2x2 binning, long-time and short-time pixels, and dual ISO is murky. Analysis is complicated by Sony's lack of SNR information on the IMX272 and I don't see any clear mapping between the two chip's PGA capabilities and the ISOs available on the camera bodies. But the basic thing is it's likely a comparison between 10.2 effective MP and 20.3MP with the effective MP plausibly having distinct behavior from actual MP.
If one were to make a baseline assumption the 10.2MP sensor had 1.4x the pixel pitch of the 20.3MP the associated null hypothesis would be that the 10.2MP sensor would have one more stop of dynamic range because its pixels have twice the area. We might be sceptical this null hypothesis is a good approximation because
- Pixel scaling rarely exact and neither sensor is back side illuminated. Additionally, effective MP are unlikely to fully reproduce the behavior of actual MP.
- Bill Claff's measurements, linked upthread, show 0.1 stop more dynamic range from GH5 below ISO 800 and half a stop in favor of the GH5S above. If the measurements are correct, the null hypothesis is false for the GH5S. Since the GH5S and BGH1 very likely share a sensor it's unlikely off die changes would introduce a 0.5-1 stop of improvement, implying the null hypothesis is most likely also false for the BGH1.
- The null hypothesis has been shown to be false for similar changes in pixel area in MP to MP comparisons involving the IMX272 which bypass possible effective MP confounds.
- CineD changed methodology between testing the GH5 and BGH1, suggesting the two measurements are not directly comparable. While other causes of differences are certainly not precluded, it appears plausible higher DNR is reported for the BGH1 due to the change from VLog to VLog-L, use of waveform charting, and perhaps the use of different test charts. Though CineD indicates SNR = 1 for both measurements the text of the two articles also suggests different handling of noise that may result in reporting of higher BGH1 DNR.
It does not seem CineD has measured the GH5S or retested the GH5 with VLog-L firmware and BGH1 methods, precluding triangulation of their results. One might hope someone eventually sends BGH1 data to Bill since it is not at all obvious a later sensor design from same manufacturer with substantially larger pixels and dual ISO actually does perform better.
In particular, it appears a plausible interpretation of currently available evidence is the lower IMX294ish ISO underperforms the IMX272 while the upper ISO realizes only half of the potential of the larger pixels. From an information content perspective, this implies the upper ISO is less suited to downsizing workflows due to its lower area efficiency even though it favours more attractive headline numbers.
Unfortunately if BGH1 DxOMarks were available I'm not entirely they would be helpful to our purposes here. DxO hasn't measured the GH5S and it's my understanding their scores aren't always directly comparable. While I would hope DNR measurements could be compared and I suspect DxOMark's methods are probably more stable than CineD's
DxOMark protocol 4 does include some dynamic range changes.
