R5 Mark II worse sensor readout speed in video mode, why?

[dtgq]

Not only is there a head-scratcher of 4K60 SRAW performance, but there's also a head-scratcher of video sensor readout performance:

ES 6.3ms

Data from DPReview:

17.3ms RAW 24-30

7.2ms line-skipped modes

12.8ms RAW 60 and all other modes

Data from Gerald:

8K(24?) RAW 17.3ms

8K60 RAW lite 12.6ms

4K24 RAW line-skipped 17.3ms

4K "Fine" oversampled 17.3ms

4K60 RAW line-skipped 12.6ms

4K24 "Standard" line-skipped 9.6ms

4K60 "Standard" line-skipped 9.6ms

4K120 "Standard" line-skipped 7.2ms

I don't understand why rolling shutter should be worse than 6.3ms for any video mode at all unless Canon is deliberately sandbagging.

Also if I understand correctly, the R5 sensor is read in 16 line chunks right? So with any other camera, rolling shutter at least looks smooth, but with the R5 it looks like stair steps?

 

[Ephemeris]

Not only is there a head-scratcher of 4K60 SRAW performance, but there's also a head-scratcher of video sensor readout performance:

ES 6.3ms

Data from DPReview:

17.3ms RAW 24-30

7.2ms line-skipped modes

12.8ms RAW 60 and all other modes

Data from Gerald:

8K(24?) RAW 17.3ms

8K60 RAW lite 12.6ms

4K24 RAW line-skipped 17.3ms

4K "Fine" oversampled 17.3ms

4K60 RAW line-skipped 12.6ms

4K24 "Standard" line-skipped 9.6ms

4K60 "Standard" line-skipped 9.6ms

4K120 "Standard" line-skipped 7.2ms

May I ask the source of the data?

I don't understand why rolling shutter should be worse than 6.3ms for any video mode at all unless Canon is deliberately sandbagging.

Also if I understand correctly, the R5 sensor is read in 16 line chunks right? So with any other camera, rolling shutter at least looks smooth,

but with the R5 it looks like stair steps?

Should this be R52?

 

[dtgq]

Not only is there a head-scratcher of 4K60 SRAW performance, but there's also a head-scratcher of video sensor readout performance:

ES 6.3ms

Data from DPReview:

17.3ms RAW 24-30

7.2ms line-skipped modes

12.8ms RAW 60 and all other modes

Data from Gerald:

8K(24?) RAW 17.3ms

8K60 RAW lite 12.6ms

4K24 RAW line-skipped 17.3ms

4K "Fine" oversampled 17.3ms

4K60 RAW line-skipped 12.6ms

4K24 "Standard" line-skipped 9.6ms

4K60 "Standard" line-skipped 9.6ms

4K120 "Standard" line-skipped 7.2ms

May I ask the source of the data?

Canon EOS R5 II in-depth review: Digital Photography Review

Canon R5 Mark II Review: A Truly Impressive Camera

I don't understand why rolling shutter should be worse than 6.3ms for any video mode at all unless Canon is deliberately sandbagging.

Also if I understand correctly, the R5 sensor is read in 16 line chunks right? So with any other camera, rolling shutter at least looks smooth,

but with the R5 it looks like stair steps?

Should this be R52?

Both.

 

[Ephemeris]

Not only is there a head-scratcher of 4K60 SRAW performance, but there's also a head-scratcher of video sensor readout performance:

ES 6.3ms

Data from DPReview:

17.3ms RAW 24-30

7.2ms line-skipped modes

12.8ms RAW 60 and all other modes

Data from Gerald:

8K(24?) RAW 17.3ms

8K60 RAW lite 12.6ms

4K24 RAW line-skipped 17.3ms

4K "Fine" oversampled 17.3ms

4K60 RAW line-skipped 12.6ms

4K24 "Standard" line-skipped 9.6ms

4K60 "Standard" line-skipped 9.6ms

4K120 "Standard" line-skipped 7.2ms

May I ask the source of the data?

Canon EOS R5 II in-depth review: Digital Photography Review

Canon R5 Mark II Review: A Truly Impressive Camera

Thank you - an interesting video, most appreciated. I really aught to measure our r52s and send it over to Horshack

I don't understand why rolling shutter should be worse than 6.3ms for any video mode at all unless Canon is deliberately sandbagging.

Also if I understand correctly, the R5 sensor is read in 16 line chunks right? So with any other camera, rolling shutter at least looks smooth,

but with the R5 it looks like stair steps?

Should this be R52?

Both.

 

[weigertj]

There is no intentional dumbing down here. Reading a sensor for photography and video is totally different. All the cameras have different values as you can see here:

https://horshack-dpreview.github.io/RollingShutter/

There must be a scientific explanation behind it, maybe someone knows more about it.

 

[PicPocket]

I don't understand why rolling shutter should be worse than 6.3ms for any video mode at all unless Canon is deliberately sandbagging.

I hope you have seen this thread

https://www.dpreview.com/forums/post/67470284

As for sandbagging, that just seems like wide speculation. It isn't uncommon human nature to assume when not knowing, but given this phenomenon has been observed for a wide range of gear across manufacturers, so there is more to it for sure before we go conspiracy

Can it be something as simple as we don't run different length and terrain races at same pace

--
PicPocket

 

[dtgq]

I don't understand why rolling shutter should be worse than 6.3ms for any video mode at all unless Canon is deliberately sandbagging.

I hope you have seen this thread

https://www.dpreview.com/forums/post/67470284

As for sandbagging, that just seems like wide speculation. It isn't uncommon human nature to assume when not knowing, but given this phenomenon has been observed for a wide range of gear across manufacturers, so there is more to it for sure before we go conspiracy

Can it be something as simple as we don't run different length and terrain races at same pace

I hadn't, but now have! If I understood correctly, at least with Nikon, in video mode the readout speed per line is the same, but it's reading fewer lines simultaneously (than the 16 line chunks it does in stills, for example).

 

[simon_dpr]

I think it's caused by different pixels.

R52 has 8192x5464 pixels.

But even for 8K, it' only 7680x4320.

When recording video, it will need to do something like oversample, to scale down the picture size.

Otherwise the video quality will be very poor.

 

[PicPocket]

I think it's caused by different pixels.

R52 has 8192x5464 pixels.

But even for 8K, it' only 7680x4320.

When recording video, it will need to do something like oversample, to scale down the picture size.

Otherwise the video quality will be very poor.

R5 / R5 II are 8192x4320 for 8K DCI. For 8K UHD, you lose 256 pixels from each side of 8K DCI. It's basically a slight crop. From what I know, readout speeds are same for DCI and their UHD counterparts, but do vary quite a bit with the resolution, frame rate and sampling method (4k vs 4k fine). So it's not just based on pixels, at least not in a way we can easily correlate

 

[koenkooi]

I think it's caused by different pixels.

R52 has 8192x5464 pixels.

But even for 8K, it' only 7680x4320.

When recording video, it will need to do something like oversample, to scale down the picture size.

Otherwise the video quality will be very poor.

R5 / R5 II are 8192x4320 for 8K DCI. For 8K UHD, you lose 256 pixels from each side of 8K DCI. It's basically a slight crop. From what I know, readout speeds are same for DCI and their UHD counterparts, but do vary quite a bit with the resolution, frame rate and sampling method (4k vs 4k fine). So it's not just based on pixels, at least not in a way we can easily correlate

The sensor has to read out a complete line, regardless of the desired width. This is why rolling shutter only improves 1.6x in stills crop mode and not 2.56x.

 

[PicPocket]

I think it's caused by different pixels.

R52 has 8192x5464 pixels.

But even for 8K, it' only 7680x4320.

When recording video, it will need to do something like oversample, to scale down the picture size.

Otherwise the video quality will be very poor.

R5 / R5 II are 8192x4320 for 8K DCI. For 8K UHD, you lose 256 pixels from each side of 8K DCI. It's basically a slight crop. From what I know, readout speeds are same for DCI and their UHD counterparts, but do vary quite a bit with the resolution, frame rate and sampling method (4k vs 4k fine). So it's not just based on pixels, at least not in a way we can easily correlate

The sensor has to read out a complete line, regardless of the desired width. This is why rolling shutter only improves 1.6x in stills crop mode and not 2.56x.

Hence the same for UHD and DCI

 

[Ephemeris]

I think it's caused by different pixels.

R52 has 8192x5464 pixels.

But even for 8K, it' only 7680x4320.

When recording video, it will need to do something like oversample, to scale down the picture size.

Otherwise the video quality will be very poor.

Looking at some datasheets for Sony sensors Horshack shared links to it relates to how the data is buffered (or not) transporting off the sensor.

This is why if we take full frame stills at 30fps, and video and 30(near to)FPS the effect of rolling shutter / read out/ data transport can be seen to be different in the results

 

[Ephemeris]

I think it's caused by different pixels.

R52 has 8192x5464 pixels.

But even for 8K, it' only 7680x4320.

When recording video, it will need to do something like oversample, to scale down the picture size.

Otherwise the video quality will be very poor.

Looking at some datasheets for Sony sensors Horshack shared links to it relates to how the data is buffered (or not) transporting off the sensor.

This is why if we take full frame stills at 30fps, and video and 30(near to)FPS the effect of rolling shutter / read out/ data transport can be seen to be different in the results

This video from Horshack relating to the Z6III shows how with a different design we can have similar video performances to JPEG stills.