Ursa 12k Data Rates at 4k vs BMPCC4K Data Rates

Hi All,

Perhaps a dumb question, but I have noticed that the memory usage on the Ursa 12k whilst shooting at 4k is allot more than the memory usage when shooting at 4k with the BMPCC 4K.

If "4k" is an expression of the number of pixels on screen (resolution), then why would there be such a big difference between cameras when using the same compression / continuous quality ratios both at the same resolution?

Also, when shooting in 4k using the Ursa 12k, does the Ursa 12k initially capture everything using the entire sensor and then down sample to 4k?

Any insights / education on this would be interesting to hear.

Lee

Ursa 12k downsamples.

Perhaps a dumb question, but I have noticed that the memory usage on the Ursa 12k whilst shooting at 4k is allot more than the memory usage when shooting at 4k with the BMPCC 4K.

Which URSA12K? Cine or Mini? At the same BRAW compression settings, the 12KMini should match the amount of data you'd get from a BMPCC4K exactly. The 12KCine has a taller sensor than either the BMPCC4K or 12KMini though, so in 16:9 or 3:2 format its 4K mode will need more data per-frame.

Maximum Data Rates 24p BRAW
4K DCI 4096x2160 CFast2

UMPro12K 5:1 161 MB/s
UMPro12K Q0 161 MB/s to 268 MB/s

BMPCC4K 5:1 82 MB/s 30 fps
or about 66 MB/s at 24 fps (almost 60% less!)
BMPCC4K Q0 102 MB/s to 203 MB/s
or about 82 MB/s to 162 MB/s (almost 50% less)
NB: BMCC6K LF and all pocket cameras use the identical rates

Maximum Data Rates not listed for UMPro4.6K G2 4K DCI so I calculated based on UHD rates 30 fps:

UHD 3840x2160 30 fps -> 24 fps
5:1 76 MB/s calculated 24 fps 4K DCI 65 MB/s, lower than the pocket cameras!
Q0 76 MB/s to 191 MB/s calculated 24 fps 4K DCI 65 MB/s (less than pockets) to 163 MB/less (same as the pocket cameras).

The UMPro12K is an outlier with much higher data rates for 4K DCI. I can only assume the difference in data rates is attributed to the way BRAW is designed when working with the RGBW sensor that uses many more photosites than the traditional Bayer RGB layout to determine pixel values.

The 12k doesn’t downsample in the traditional sense. Can’t really downsample in raw, rather it skips rows and columns of pixels. Shouldn’t really matter though? I’d think they’d be the same but it looks like there’s a 2.5x difference in file size at 5:1.

The rgbw sensor constructs a luminance image from filtered and unfiltered pixels, whereas the 4k sensor uses only filtered pixels. At the end of the day they’re both 4096x2160 12bit 420 @ 5:1. I’d definitely be interested in the explanation.

Good Luck

Even if BRAW is not RAW in a conventional sense, the better dynamic range must be stored somewhere before final decode.

I was thinking something similar but the disparity still seems too great. Maybe in linear, but not log.

Good Luck

I doubt we'll ever get more info than that, after all, it's BM's "Coke Recipe".

The 12k doesn’t downsample in the traditional sense. Can’t really downsample in raw, rather it skips rows and columns of pixels.

No, that is not correct. The 12k does not skip any rows or columns of pixels. What it does is binning - grouping pixels together to form a bigger pixel.

Better to use the term photosites, so a pixel is the result of a grouping of photosites.

Absolutely! Thanks for demanding precise definitions.
One that recently got some acceptance is "sensel" (sensor cell).

The 12k doesn’t downsample in the traditional sense. Can’t really downsample in raw, rather it skips rows and columns of pixels.

No, that is not correct. The 12k does not skip any rows or columns of pixels. What it does is binning - grouping pixels together to form a bigger pixel.

It’s my (admittedly limited) understanding that 8k and 4k use smaller unit cells to acheive faster readout at the cost of reduced spatial resolution.

Good Luck

Just conjecture: If 12K uses a 6x6 matrix for 36 photosites, perhaps 8K uses 5x5 for 25 photosites and 4K uses 4x4 for 16 photosites given that is part of the technique to maintain the same frame but reduce the resolution. That wouldn’t be downsampling and theoretically might reduce the dynamic range. Would be interesting to see each resolution’s DR and latitude then.

Just conjecture: If 12K uses a 6x6 matrix for 36 photosites, perhaps 8K uses 5x5 for 25 photosites and 4K uses 4x4 for 16 photosites given that is part of the technique to maintain the same frame but reduce the resolution. That wouldn’t be downsampling and theoretically might reduce the dynamic range. Would be interesting to see each resolution’s DR and latitude then.

If it's able to use a matrix approach, how is it able to increase the scan speeds and increase frame rate, since a matrix approach would still need to reach the data on all the photosites.

I must be skipping some photosites in order to increase the readout speed when going to the 8k and 4k modes. This has me scratching my head.

If my conjecture is correct regarding use of a smaller matrix, if recording 4K then perhaps the gain in frame rate comes from processing less than half the photosites.

However if that is true, one would think the dynamic range would be reduced. I doubt BMD would reveal everything they’re doing if this is part of their secret sauce. The consumer can judge the quality they see with each recording resolution and frame rate without needing to know everything about the camera image processing.

If I had the camera, I’d be testing the available combinations in camera and the post production alternatives in DaVinci Resolve to get to a 4K or 8K delivery.