Hacker News Reader

Comment by tetrep | Hacker News Reader

Could it be a bandwidth issue? I remember reading about 120fps 4k video on an Android phone being limited to only a few (real-time) seconds because there wasn't enough I/O bandwidth to stream the video to storage so it had to stream into a specialized hardware buffer that was relatively small and expensive.

View on HN

Modern SD cards go up to 300 MB/s for writes. That should be enough. Even if you dump the pixel data with zero compression.

8-bit pixels: 1920x1080x30 = 62.2 MB/sec

14-bit: 109 MB/s

Even UHD in 8 bits should work.

I think if you're Raw spatialy down-sampled you'll want RGB and not Bayer under-sampled so triple that. No SD is reasonably over 250MB/sec and they are all 64GB. Even at FHD 30Hz you'll only get 30sec of 8bit (not really Raw) video on your insane SD card.

https://havecamerawilltravel.com/photographer/fastest-sd-car...

No, that makes no sense. Why would you store debayered data when you can store the original with the same exact results?

You realize that Bayer data counts every column and row whether R G or B, right? Half the pixels (21Mpix) are green and the other 2 quarters are R and G.

So if you're going down to FHD (2kx1k) from 42Mpix (8kx5k) you have already lost massive ammounts of resolution (and noise). Bayer FHD would be SPR (sup-pixel rendered) and even lower resolution (1/3 RGB e.g. 1080 lines of 1920 RG/BG alternating columns) than most JPEG FHD (natural scene) not suitable for video compression much less Raw. That would be silly.

If you don't down-sample the data somehow you can't even get it off the CIS which are limited to ~2-5Gb/sec. Since most reasonable image sizes are large enough you can see color at full resolution we typically use RGB at FHD, compressed or not. Alternately, you can go to 4:4:4 to 4:2:2 or even 4:1:1 in Yuv (YCbCr), but that's not Raw or anything like it.

Definitely a lot of bandwidth required. 60 fps 4k would be 1.7 GiB/sec, 120 fps twice that.

3840×2160×10×3×60÷8÷1024÷1024÷1024

RAW format doesn't just mean an uncompressed image like a PNG. It's a direct readout from the sensor with minimal processing. It can be substantially larger than you'd expect just based on the resolution * bits per pixel.

You are correct, but don't forget the Bayer encoding that makes things smaller!

Good point! But in the other direction, don't forget that even those there's less "pixels" than expected from the sensor, there will be more than 10 bits per channel, typically 12 or 14. I shouldn't have said it's larger, not necessarily - but it's definitely not 1:1!

You don't need a x3 there since most cameras use a single monochrome sensor with a colour filter array on top – the debayering process can happen offline.

Reply on news.ycombinator.com