ALl-I Or IPB on MKIII?

Kholi

Section Moderator
Anyone reach a conclusion on this? Finally landed in my hands, about to go and rumble with it for a few days.
 
IPB unless there's lots of movement and detail. All-I bitrate can drop as low as IPB. IPB has B-frames so it looks cleaner than the old H.264.

All-I is easier on CPU.
 
5D Mark III uses H264 High Profile Level 4.1 for IPB encoding, and High Profile Level 5.1 for ALL-I encoding. These are really high standard in H264 specification and support for them in NLE is not very perfect yet.
 
Cool! VBR used to crap out at around 60Mbit/s on the MKII. Looking forward to CBR ALL-I and variable IPB. What card were you getting 122Mbit/s with? Was it stable?
 
Cool! VBR used to crap out at around 60Mbit/s on the MKII. Looking forward to CBR ALL-I and variable IPB. What card were you getting 122Mbit/s with? Was it stable?

With IPB we managed to max out at around 60Mbps VBR, this is limited by the High Level 4.1 specification, High Level 5.1 (All-I) can reach 300Mbps theoratically.

The card is a Sandisk Extreme 60MB/s CF, we ran a benchmark that showed this card is capable of about 39MB/s maximum write in 5D3, which is about 290Mbps

yes 122Mbps was stable, at ISO12800 it's all noise though...
 
200Mbit/s+ ALL-I would be amazing! No need for ProRes.

Do you think there's any chance of changing the colour space to 4:2:2?
 
200Mbit/s+ ALL-I would be amazing! No need for ProRes.

Do you think there's any chance of changing the colour space to 4:2:2?

Yeah I think 150Mbps CBR ALL-I would be a good balance between quality and file size.

Currently I don't think there is any chance of achieving 4:2:2, that would mean completely rework the H264 encoder.

But if we can find some top-notch russian hackers/video engineers to help us, it could be possible...
 
from all those numbers, you posted, the one that would make me most interested is IPB 60Mbps VBR: that should look very, very nice (and also: a1ex knows this, but just in case he forgets... if you get access to the RAW data from the sensor, call me, I'm writing a high performance cinemaDNG encoder for ARM processors...)
 
Aside from raising the Mbps, have you guy's been able to increase the resolution of the codec? Or is that something is too low level?

Just out of curiosity (this is a little off topic), but do you think you will be able to over-crank the 5D3 significantly at 1080p?
 
Aside from raising the Mbps, have you guy's been able to increase the resolution of the codec? Or is that something is too low level?

Just out of curiosity (this is a little off topic), but do you think you will be able to over-crank the 5D3 significantly at 1080p?

Well there is always possibility, but who knows if we can dig that deep... some initial tests with overcranking bricked a 5D3, then luckily unbricked.

The high level 5.1 codec ALL-I uses support 1080p up to 120fps and 4K*2K up to 30fps, it's theoratically achievable but I think it would be very hard to do it on 5D3...
 
Wow if the all-I codec is able to record 4k*2k I wonder if it would be possible to hack that into the camera, then again there could be heat issues and with only at certain bit rates it may not be the best 4k but it would probably look nice and crisp at 1080p. I know the digic 5+ chip is someing like 17 times faster than the digic 4 chip so you would think it would be fast enough to read out all that info. Does anyone know when any beta testing for magic lantern will be released.
 
doubt it
and digic 5 is 17x faster at some very specific tasks - for most stuff, it's said to be around 3x faster (which is already a huge improvement, but not 17x)
 
Is there any chance you could improve or change Canon's sharpening algorithm? Or add in another one? Everyone is doing post-sharpening because it's so bad (seems like USM, big halos, etc.). A better in camera sharpening algorithm would be huge. There would be less need for super high bit rate codecs in this case, too...

I couldn't care less about 1080p/60fps... 1080p from the 5d 3 looks like 720 at best anyway and no one uses this feature for professional production; slow motion is a toy or extremely special-purpose at best. Just rent something better if you really need slow motion! An improvement to sharpening would be HUGE.
 
Is there any chance you could improve or change Canon's sharpening algorithm? Or add in another one? Everyone is doing post-sharpening because it's so bad (seems like USM, big halos, etc.). A better in camera sharpening algorithm would be huge. There would be less need for super high bit rate codecs in this case, too...

I couldn't care less about 1080p/60fps... 1080p from the 5d 3 looks like 720 at best anyway and no one uses this feature for professional production; slow motion is a toy or extremely special-purpose at best. Just rent something better if you really need slow motion! An improvement to sharpening would be HUGE.

"No one uses this feature for professional production"? Feature film "Act of Valor" and "Red Tails" all used 5D2 and other Canon DSLRs, and no one complains about them being soft.
 
Back
Top