Great news from Trammell Hudson from Magic Lantern about the recording rate of the Canon 5D MKII camera with tests achieving 66 MBps fairly stable and bursts of 74 MBps at 24p. If lifting the recording bit rate goes the way of the Panasonic GH1 camera we might see real changes within weeks, hopefully days. Nice thought but either way nice to see changes that are pushing the DSLR cameras along and making them better out in the field.
Get on board if you dare and help Trammell do what Canon should have done!

I have been neglectful of my favorite camera these past few months
and left it running the older 2.0.4 Canon firmware. Today I spent
the afternon locating the necessary symbols (thanks, patchdiff2!)
and have built a version that boots on the 5D Mark II with 2.0.8.

Attached is a version for folks to test. I put it through some
tests and the few things I tried seemed to work (autoboot, MBR
writing, zebras, int/ext mic switching, etc). I also fixed (?) a
bug in the bmp_load() code: by reading into an uncacheable region of
memory it seems to never get the bad results from the read.

Please give it a go and let me know if there are any outstanding
issues. Once we have it a stable we should make this one generally
available since all the new cameras have 2.0.8 pre-installed.

FYI: if you have a camera with a serial console, there does appear
to be a password on the port now.


h.264 qscale

The Canon function mvrSetDefQScale(int16_t *) (2.0.8 0xff990188) sets the quality parameter to the camera’s h.264 encoder. If the CF card is not able to keep up, the camera will display a buffer-level bar on the right side of the LCD and stop recording when it hits full. The parameter does not have any effect unless fixed bit rate encoding is selected via mvrFixQScale(int16_t *) (2.0.8 0xff9905d4).

The default q-scale is -8 in VBR mode and has a nominal 42 Mbps with the 2.0.8 firmware. The 24p tests were done with the camera on a tripod writing to a Lexar 400x UDMA 16 GB card with all Magic Lantern tasks disabled (no zebras, cropmarks, etc). qscales less than -12 occasionally stopped recording.

Q-scale FPS Mbps Notes
+8 30p 7 MBps Very low quality
-9 24p 32 MBps
-10 24p 41 MBps
-11 24p 51 MBps
-12 24p 59 MBps
-13 24p 62 MBps
-14 24p 66 MBps Best, sustainable in 24p mode?
-15 24p 68 MBps 2 buffer bars
-16 24p 76 MBps Only a few seconds


* 24p seems to have less issues than 30p
* Locked-off shots seem less likely to trigger buffer overflows; amount of motion?
* Auto-exposure changes seems to lead to immediate overflows: shoot manual exposure mode
* -14 seems to be the best that can be sustained with the card with all other ML tasks running.

Test firmware

* Download it from this ml-devel post

Check out Trammell’s work over at Magic Lantern.

Leave a Comment Here