Messages

Perhaps impossible, but found out that my battery drains very fast during filming if the OIS (or VC as on Tamron) is turned on. At the moment I turn the thing off between every take, but sometimes I forget. OIS is always on when liveview is on even though Im not recording video. If outside liveview the OIS turns off after about 5 seconds of non use.

It would have been a great feature to have the OIS physically turned on, but only active during recording.

Is this benchmark good or bad?

I would say it looks very good. If you record some raw with the debug on you can see a buffer graph. Look if it is stable or very volatile. Try record a couple of times to the card, format it, and see if the speed is changing. If not I think you made a very good purchase. Is it 5Dmk3?
edit: (can be seen in the pic). It would be nice to see how fast the Sandisk 160MB/s is on a 5Dmk3)

as stated above:
http://www.magiclantern.fm/forum/index.php?topic=5391.msg113114#msg113114

The 5D3 full size sensor is both a blessing and a curse. I can tell you that most pro DP's despise it and much prefer the Super-35mm sized APS-C.

Interesting! Why?

you won't get dissapointed.

edit: maybe you just did ! sorry

I wouldn't go "safe" with the transcend. IMO the Transcend-cards are alot more variable in performance than any of my sandisk-cards (I also have a 400x Transcend that is slower than my sandisk 60MB/s) . I returned my 1000x Transcend in favor for a Sandisk because of the performance issue. It might have been something wrong with it but it was very volatile in the write-speed. The Sandisk gives me steady speed all the time. Also check out the VPG65 that the Sandisk and Lexar 1066x has: guarantee of a least 65MB/s writespeed! (VPG : video performance guarantee)

check my benchmark (sandisk 1066x vs transcend 1000x)
http://www.magiclantern.fm/forum/index.php?topic=5391.msg113114#msg113114

thanks

Alas, I personally really miss the matrix code from the old implementation, just resurrecting this with raw dng would be at least an improvement even if it took a while to take a pic.

Is there anywhere I can see the source for that code? Was it a module or can I just look at an old version of silent pic?

[Not very supported if you ask me]

Sandisk Extreme Pro 64GB 160MB/s (1067x)

87.1, 86.9, 86.9, 81.2, 78.9, 75.0, 69.2, 23.8

Sandisk Extreme Pro 128GB 100MB/s (667x)

76.9, 77.2, 70.4, 63.5, 65.9, 57.1, 54.4, 21.4

Transcend 64GB 1000x  - Not very supported if you ask me

59.1, 61.2, 63.9, (?), 46.6, 53.1, 52.5, 14.0

The mirror is just a thing that moves up and down (don't see how mechanical wear should affect it), but the real action is in moving the shutter blades (where timing is essential, since they determine the exposure time).

Ah okay, I just assumed they was connected, but of course they aren't.

For nearly static subjects you might even be able to exploit the line skipping property with super-resolution (that is, combine 2 or 3 frames into a higher-resolution one). It's on my list of things to be studied, but it requires some math skills quite a bit above mine. It's pure postprocessing though, no hacking required.

Interesting. What kind of math are we talking about here? The silent pic resolution is 3,744/3 ≈ 1250 in height? Does that mean 2 pixel of line skipping? So to capture all y pixels you would need 3 pics? Then the same for the skipped pixels at the x-axis? that would give us 3x3 pics = 9 pics to cover whole sensors all pixels?
Is it possible to do 1880x1250 readouts with different offsets? The burst in silent module is quite fast isn't it?

I have had both macs and pcs and I have to say I like the apple world better.

If you buy an iMac you get a solid solution that will give you what you need in performance. You also get a nice/neat desktop that looks nice. I had an iMac before i got my macbook air (that I currently do editing on in raw).

If you buy a PC you can get twice the performance for the same amount of money. If you build a hackintosh you get both - great OS together with moneysaving performance. There are tons of guides out there on how to buid a working hackintosh.

My next upgrade will be something that runs OSX. I have not decided if I will just take the easy route and cash up for a new iMac or if Im going to build a serious renderfarm- hackintosh. The nice thing about just buying a ready iMac is that you can concentrate on filmmaking instead of nerding in forums like this

  oh. trolling. I see. Usually I prefer Mbit to keep them apart

...buy better card.  160Mb/s is nothing and recording with this card is a suicide  !!!!  Do you see difference between Mb vs MB ?

?
@windh
I have the Sandisk Extreme Pro 64GB 160MB/s (1067x). This card is very fast but the 160MB/s is the read-speed. Actual write speed is around 150MB/s. The card has the VPG65 logo (Video Performance Guaranteed), which means that the card will always deliver write speeds above 65MB/s! In reality the card writes at around 70MB/s in my 5D2. In my USB3 card-reader it delivers the promised 150MB/s write speed. I think this card is very stable and I would definitely recommend it if you have the money.

I changed my Transcend 64GB 1000x to a Sandisk 128GB 100MB/s 667x. The Sandisk performed better with continuous recording @1856x842 2.20:1 with sound. Decided to change the Sandisk to a 1067x Sandisk Extreme Pro 150MB/s to get highest performance possible on the 5D2.

I made benchmarks of the all the cards, even the ones i returned. If anyone has some collection of ppm-files feel free to add these:  http://snk.to/f-cdul6p3u

Transcend 1000x 64GB - the first faulty card I had (maybe its not representative for Transcend)
Sandisk 128GB 100MB/s 667x - my first exchange, turned out to be a bit slow for highest setting on my 5D2
Sandisk Extreme Pro 64GB 150MB/s 1067x - Rock solid with very stable rates regardless of buffer settings etc.

Besides this, powersaving is one of those features that nobody notices when it's completely broken for months (it was not working at all since March 09, and I've fixed it the day before yesterday, when a related a related issue was reported). I could not find the exact issue reported in the tracker, and from this I can only conclude that nobody will miss this feature if I'll remove it.

Users feature requests are often only wanted but never used.
Users also tend to think within the box created for them, solving the wrong problems e.g. suggesting some complicated solution when the goal simply is to get as long photo interval as possible with minimal battery drain.

On the other hand, from a user perspective I only tend know what I need when I need it. I like using the silent pic to save the shutter (this might be my first wrong assumption) and being silent. When using it it feels nice to turn LCD off to save battery (maybe not saving so much then?). Going on a trip this summer I prepare for all the features I will use. Thats why I tested it and reported the error (I am NorthFilm).

I guess the interval with live view OFF is saving more battery (even with mirror moving?) - so then it would probably be better with a solution that just kept the mirror up.

[Conclusion:]

The nightly for Feb16th is the one I use for Paid Work. Rock Solid !
http://builds.magiclantern.fm/jenkins/job/5D2.212/206/artifact/platform/5D2.212/magiclantern-v2.3.NEXT.2014Feb16.5D2212.zip

I just tested the feb16th against the latest and I don't get much performance difference in MLV-recording w/w.o sound. I just tested one resolution that I know the card can't handle and I get the same amount of frames (within deviation of 2%). The card is a 60MB/s Sandisk which performs very stable in its write-speed.

Testing raw module at max 1856x1044 (the higher resolution is not available to compare against in later builds)

Build 16Feb
Best result (tried different settings etc)
126 frames@1856x1044 (16:9) fps 23.976 (the best out of 10 readings - all between 123-126) - stable

Build 25Apr
Best result (tried different settings etc)
I got one 74 frames - early stop (!) - had global draw on by mistake.
148 frames@1856x1044 (16:9) fps 23.976 (the best out of 10 readings - all between 126-148) not counting early stop

Conclusion: 16th Feb might be more stable, 25th Apr seem faster but gives a bit more unpredictable result.

More: Just tested the MLV + Audio 8khz (for sync) and got stable 126 frames @1856x1044. That is as fast as raw module in 16th feb.

Ah, thats unfortunately not possible what I know of (yet?). I have seen that question myself and the common answer here is to buy one of these VAF  filters that holds the mirror up. A cheap alternative is to just tape it.

I once asked for a higher resolution silent picture module, and got the answer that I should develop it . Apparently it is possible to take higher resolution silent pics by moving the crop window and so covering the whole sensor. There is no implementation of this yet (I made a feature request) but think the developers rather develop fun stuff than boring easy stuff. To mention is also the delay of (if I remember correctly) 2 sec between every crop image, which will screw up moving objects (clouds, trees etc). However, if the images overlap this could easily be corrected in post. Don't know about longer shutter times... that might not work very well.

Maybe there are other ways, but I'm no ML developer so...

Another way to get an estimate is to look in the ML-menu. While browsing resolutions you can see how many MB/s and even a frame/time approximation [this however seem to be non-working at latest build??] . This works only after you have recorded something as the estimate is based on previous recording speed.

Would be nice to see time instead of frames IMO (or both).

Hum. If not using the silent-picture you can just press the live view button to turn off live view. Don't forget to turn off image preview from canon menu. There you go - full resolution timelapse without LCD turning on. I tested this right now with build 24th April. Works.

Dont forget to turn off live view by just press the lv-button (thats why you still get preview even if preview is off).

@Doyle4
ad's build? Where do you find them? I thought it was only unified now... I just download the nightlies (probably tested a dozen since feb)
@reddeercity
About the 16th?. Is it faster? - What is the maximum resolution for continuous recording?

I can't really say anything about the speed of raw-recording (I do MLV w sound for syncing) between the different builds. I just control that it works.
I have successfully recorded continuous in 70MB/s and below on the 23th of April build. That means
1792 x 968   (1.85:1)   continouos with audio of 48.1kh (crop mode)
1856 x 928   (2:1)      continouos with audio of 48.1kh
@24fps
(Sandisk Extreme Pro 64GB 160MB/s 1066x).

If you think I can get higher than that I can try some other build?

I also tested everything with the HDMI out (to my field monitor) without problems.

Had no problem except some power save feature that I reported here (https://bitbucket.org/hudson/magic-lantern/issue/1969/5d2-interval-liveview-powersave#comment-9816912). A1ex was able to fix it and the 24th is working fine.

To shoot with mirror up:
load the silent module (this will enable silent pics from live view)
The silent picture settings will be found under camera. Just turn it on, and then use the intervalometer.
(unfortunately the resolution will only be 1880x1250 or 2152x1076 if you zoom in 5x/10x before taking the pic. The latter is to recommend if you don't want aliasing artefacts, but you loose the full frame image since this will only be a crop.)

To powersave:
There is a setting in "Powersave in LiveView"
just set the turn off LCD to 20min (or whatever), and if you do time-lapse with silent pics, the LCD will turn off immediately

This works in the 24th nightly (before that it was a bug with the powersaving).
Hope that helped.

I would like it to set Image quality to raw when Im in photo mode and change to - & Small JPG when in movie mode (to free buffer for the raw-filming. However, ideally I would like to not have to think about it (if there was no performance difference).

I just tried latest 12.2 version av AE and they have something called detail-preserving upscale. I can pixel peep and see a small advantage (more sharpness) DPU compared to AE bilinear scaling. Anyway - as Midphase say scaling just 103.7% does not make footage soft.
I will try more - maybe one can get away with more scaling if using this new scaler. It is also fast - I could play footage in 10fps on my mac bookair after scaling. Overall new AE is alot faster than 12.0.
/J

Ok, I can now confirm that this new Sandisk 160MB/s (1067x) gives me about 16s MLV + 48khz audio in 16:9 (1856x1044).
I also get continuous @ 2:1 (1856x948) + 48khz sound and below. Nice. (The old Transcend 64BG 1000x barely gave me continuous @2.20:1).

This card seem to never drop under 70M/s. I will do some more testing, but guess my 5Dmk2 is the bottle neck here.

yay!

Great!