ydgmdlu
Veteran
[Note: This is cross-posted on the 5D forum as it also would be of keen interest to readers there.]
The rumored explanation for the withholding of the long-rumored 24p firmware for the 5D Mark II is that testers and engineers felt that the jello was too severe in the 24p mode for Canon to deem acceptable for release. I know that there are many who can't make sense of this claim, but it makes perfect sense to me. It all has to do with how the camera is recording video.
The video comes from the live view stream. So the flaws in the video should have a lot to do with how the live view stream is achieved. Each frame in the stream comes from a single pass of the electronic rolling shutter down the sensor. If the frame rate of the stream is 30 fps, then each pass of the rolling shutter takes presumably around 1/30th of a second to complete (probably a just a hair less to give the DSP time to scale and compress the image). That's not very fast at all, which is why jello appears so easily.
It could be made faster, but I'm guessing that Canon's engineers didn't feel that live view needed a faster read-refresh rate. After all, live view was only intended for previewing a still photo shot, and the 1080p video was a hack made on top of it. The sensors on dedicated camcorders are most likely engineered to have maximum read-refresh rates, specifically because proper camcorders aren't supposed to have jello, as much as it can be helped.
On the DSLRs, the different frame rates are probably achieved by simply retiming the rolling shutter's (read-refresh) clock. So for 24 fps, the clock is slowed to, say, 24 Hz (~1/24th second read time), and for 60 fps (~1/60th second read time), the clock is sped up to 60 Hz. That would be the easiest, least technically challenging thing to do. Based on this hypothesis, jello should increase or decrease as the frame rate decreases or increases (respectively). We have real-world proof of this: On the GH1, rolling shutter artifacts are virtually imperceptible in the 60p mode but much worse in the 24p and 30p modes.
The GH1 has a noticeably better rolling shutter all around versus the 5D, so the timing might be something like 48 Hz for 24p and 120 Hz for 60p. I'll bet that the main reason why the 5D has less jello than the D90 is because it's 30 fps (30 Hz) versus 24 fps (24 Hz). So if Canon does release the 24p firmware, the jello would probably be just as bad as the D90's. And this is likely the rumored technical reason why the firmware hasn't been released.
That brings us to the 7D. When the first sample videos were released, there were conflicting views on the jello; indeed, the jello seemed to vary from clip to clip. It was about as bad as the 5D on some clips, but in the fast motion clips in "Line of Sight," the jello was missing. Quite possibly, this is an indication that the 7D's rolling shutter (and clock) behaves a lot like the GH1's. In fact, I predict that that is exactly what we'll find once we get our hands on the camera.
So at the end of the day, and perhaps to state the obvious, the DSLRs' sensors just weren't designed for video. They're improving with every new generation, but we still have to ground our expectations. So if the 5D receives a 24p firmware, and if the jello proves as bad as the D90's, would you still find the 24p mode to be worthwhile?
The rumored explanation for the withholding of the long-rumored 24p firmware for the 5D Mark II is that testers and engineers felt that the jello was too severe in the 24p mode for Canon to deem acceptable for release. I know that there are many who can't make sense of this claim, but it makes perfect sense to me. It all has to do with how the camera is recording video.
The video comes from the live view stream. So the flaws in the video should have a lot to do with how the live view stream is achieved. Each frame in the stream comes from a single pass of the electronic rolling shutter down the sensor. If the frame rate of the stream is 30 fps, then each pass of the rolling shutter takes presumably around 1/30th of a second to complete (probably a just a hair less to give the DSP time to scale and compress the image). That's not very fast at all, which is why jello appears so easily.
It could be made faster, but I'm guessing that Canon's engineers didn't feel that live view needed a faster read-refresh rate. After all, live view was only intended for previewing a still photo shot, and the 1080p video was a hack made on top of it. The sensors on dedicated camcorders are most likely engineered to have maximum read-refresh rates, specifically because proper camcorders aren't supposed to have jello, as much as it can be helped.
On the DSLRs, the different frame rates are probably achieved by simply retiming the rolling shutter's (read-refresh) clock. So for 24 fps, the clock is slowed to, say, 24 Hz (~1/24th second read time), and for 60 fps (~1/60th second read time), the clock is sped up to 60 Hz. That would be the easiest, least technically challenging thing to do. Based on this hypothesis, jello should increase or decrease as the frame rate decreases or increases (respectively). We have real-world proof of this: On the GH1, rolling shutter artifacts are virtually imperceptible in the 60p mode but much worse in the 24p and 30p modes.
The GH1 has a noticeably better rolling shutter all around versus the 5D, so the timing might be something like 48 Hz for 24p and 120 Hz for 60p. I'll bet that the main reason why the 5D has less jello than the D90 is because it's 30 fps (30 Hz) versus 24 fps (24 Hz). So if Canon does release the 24p firmware, the jello would probably be just as bad as the D90's. And this is likely the rumored technical reason why the firmware hasn't been released.
That brings us to the 7D. When the first sample videos were released, there were conflicting views on the jello; indeed, the jello seemed to vary from clip to clip. It was about as bad as the 5D on some clips, but in the fast motion clips in "Line of Sight," the jello was missing. Quite possibly, this is an indication that the 7D's rolling shutter (and clock) behaves a lot like the GH1's. In fact, I predict that that is exactly what we'll find once we get our hands on the camera.
So at the end of the day, and perhaps to state the obvious, the DSLRs' sensors just weren't designed for video. They're improving with every new generation, but we still have to ground our expectations. So if the 5D receives a 24p firmware, and if the jello proves as bad as the D90's, would you still find the 24p mode to be worthwhile?
