Canon 100D / SL1

[OlRivrRat]

      @ML_IDA

   Had already tried to do as You described in 1180 & 1196 so either this .mo acts differently on diff' Cam's(unlikely) or there is some

SetUp Detail that is Diff' between Yours & Mine ~ Step 3 on 1180 does not happen as described >

With OC activated, the screen Does Not go dark but instead goes to Canon Settings Info Screen > and then camera performs a

Quick (1 to 2 Sec.) test During which the red LED blinks a few times. Upon pressing the Play button the message "No Image" appears.

   BenchMark provides this  >  https://ibb.co/gohoKH

                                 ORR ~ DeanB

[IDA_ML]

Probably you are using a later build that skips the full test and provides hardcoded overclocking of the SD card interface.  This is even better but I did not have it when I wrote the above post.  With the hardcoded build, after the 1-2 seconds activation of the OC function you are ready to go.  Check the benchmark speed and start shooting. 

[Levas]

Not sure how it is on 100d, but when I record 2560x960 resolution with SD_UHS module on 6D, I always use canon normal live view, not ML preview.
So I don't use the ML preview, which is rather slow and turns to black and white quickly.
I have my settings so, that I see canon live view by default and with half shutter press I get ML preview (which is perfect for framing, even use it quickly while recording, to check framing).
That way I get continuous in 14 bit lossless, with ML preview activated it's unpredictable how long it will record.

[Walter Schulz]

@IDA_ML: What benchmark results (1 min test) do you get in movie mode after activating overclock in photo mode and switching to movie?

[Danne]

Anybody got the framing function described by Levas working on a 100D?

[OlRivrRat]

   @ ML_IDA & Walter & TWEIMC

https://ibb.co/jR8mpH
https://ibb.co/dFzFic

[Walter Schulz]

The plot thickens. Got some strange numbers in movie benchmarks, too (+malloc errors). Waiting for IDA_ML to come around with some numbers.
@IDA_ML: Can you post/upload your ML\Settings contents?

[IDA_ML]

Waiting for IDA_ML to come around with some numbers.
@IDA_ML: Can you post/upload your ML\Settings contents?

Here you go:





This speed was achieved in photo mode with a build from 21. 04. 2018 that was sent to me by Danne.  It is not very stable, so if you want to try it, please ask Danne to give it to you.  In this moment, I have the camera in my hands and here is what I get:

When I turn video mode on, select RAW-video, 5x magnification and 2520x1080 (8...11bits lossless), at 24 fps I get more than 1 minute record time, maybe even longer but I stopped it.  I can set the fps in this mode at this resolution to maximum 36,752 fps and even record about 40 frames at this speed but I get a lot of maloc errors.

In the normal uncropped mode, I can set the fps to 42 and record between 130 and 700 frames at 1736x976 resolution at 8...11bit lossless dependent on how bright the scene is.  No corrupt frames at 42 fps.  If I try 45 fps, I get lots of corrupt frames.

Which ML settings exactly would you like me to post for you?

[Walter Schulz]

Thanks!
Installed Danne's build (it's not that difficult to find his repository). Problem here (and I suppose at OlRivrRat's site, too) occurs after switching to movie mode. Running benchmark shows overclock hack is no longer effective. We're back to standard timing again. Only very few attempts show numbers beyond 50 MByte/s.

[Danne]

Is this an issue with the other build which runs a1ex latest sd_uhs code?

[Walter Schulz]

Haven't found a build giving me higher transfer rates in movie mode for 100D in reliable ways.
Update: After running several benchmark runs with your build I got numbers all over the place (27-41 MByte/s write rate) without overclocking active. Debug menu order switching and Memory error a lot.
Will switch to experimental build now and run some tests again.

@IDA_ML: I asked for a benchmark run in movie mode. Your screenshot looks like photo mode with empty DCIM directory.

[IDA_ML]

@IDA_ML: I asked for a benchmark run in movie mode. Your screenshot looks like photo mode with empty DCIM directory.

Oh, I am sorry, I overlooked "movie mode".  I remember, I tried the benchmark test in movie mode once but that generated errors and instability all over the screen, so I never bothered to try that again.  By the way, did you try Danne's build from April 20-th which has the the overclock feature but not the the 42 fps one built in?  That build is very stable and works flawlessly even at 2520x1080 resolution.  I shot more than 100 GB of footage with it and never had a glitch.

[Walter Schulz]

Found tested one from 21st. Unable to get hold of 20th build. Found one dated 18th.

[IDA_ML]

I also would like to report a different bug with the 100D running the above April 20-th build.  It probably exists in much older builds too but I never noticed it before I tried infrared remote start/stop RAW video with the infrared remote controller RC1 from Canon.  This neat little device emulates the mechanical RC1 operation remotely, over an infrared channel.  Here is the problem:

The 100D works perfectly with both: the mechanical and the infrared RC1.  When activating the remote control function from the Canon menu, I can start and stop the built-in H.264 video without problems.  And this works from ML too.  When I activate RAW video, however, and set its trigger function to the start/stop option in the sub menu, the following happens:

1) In the "medium" position of the RC1 switch, when the start button is pressed, the blue timer icon appears in the top right corner of the screen indicating stand by.  So far so good.  When I press the button a second time to start the video, the 100D starts recording but it stops immediately, recording an unreadable 4,8 MB MLV file.

2) In the "front" position of the RC1 switch, the camera focuses, takes a still photo and stops.  The screen freezes and the camera does not respond to the RC1 any more.  When I reset it with a half shutter press, and repeat the procedure in #1, after the blue icon appears, I can switch the RC1 to the "front" position and press the button again.  In this case, RAW video starts continuous recording but there is no way to stop it - neither from the RC1, nor from the camera.  It does not stop even if the camera is switched off.  Just a battery pull stops video recording.

Since the above issue is reproducible on the 100D, I hope, it can be fixed, so we can use this extremely useful remote control function to start and stop RAW video with the infrared RC1 controller. 

[IDA_ML]

Found tested one from 21st. Unable to get hold of 20th build. Found one dated 18th.

Then use the 18-th one for your tests.  It behaves in the same manner as the 20-th one but the OC function is not hardcoded for that specific fast card (64 GB SanDisk Extreme Pro/95 MB/s).  You will just have to wait 1-2 minutes for the OC test to complete after you turn the camera on.

[OlRivrRat]

      @WeIMC

   Link to 4 (SoFar) Benchmark tests >

SL1 + 64GB SanDisk Extreme Pro 95MB/s +
magiclantern-Nightly.2018Apr21.100D101_sd_uhs_hardcoded_160mhrz_45fps

https://photos.app.goo.gl/8s9d36uXVhxsiAyp1

         Am only able to get about 5 Sec's of 2520x1080 @ 24FPS
 

[IDA_ML]

      
         Am only able to get about 5 Sec's of 2520x1080 @ 24FPS

You seem to be getting very inconsistent write speed data with your card.  Maybe, you are using a fake one.  I experienced a similar issue with a SD-card from Hama which was supposed to provide 80MB/s write speeds but sometimes it would write at 20, sometimes at 60MB/s and very often, while writing at 60 MB/s the speed would suddenly drop to 12 MB/s.  I stopped using it for video.

[Walter Schulz]

I have a Sandisk 128 GB Extreme Pro, tested with different programs and it's a genuine Sandisk.
I get very short recording times, too. Nothing near to your results.

OlRivrRat and I share the same problem: We get excellent (reproducable and consistent) write rates in photo benchmark but are unable to use anything beyond 50 MByte/s in movie mode. And I think it does not depend on SD-card itself.

[IDA_ML]

Are you guys sure that you are using "RAW video" and your sub menu setting is "8...11 bit lossless compression" and not 14 bit uncompressed? And are you sure that your fps rate is set to 24 and not 30 or 42?  Check that again?  If you are still getting too short record times then quite obviously, you are severely overexposing your scene or it is extremely bright, so compression does not work properly.  This can happen if you film text on a white background on a monitor screen or a light bulb for example.  Make sure, you use a normally lit scene - the street outside or your living room for example.  And expose properly - at 0.0 or +1.0.

When I get back from work tonight, I will test write speed in the movie mode if it lets me do it.  As far as I remember, the benchmark tests are not supposed to be made in movie mode.

Danne, could you possibly post here your April 20-th build that you sent to me?  This was the one with the hardcoded OC but without the 42 fps.  It is very stable and Walter Schulz and OlRivrRat could test with it.  I am sure, that build will make them happy.

[a1ex]

OlRivrRat and I share the same problem: We get excellent (reproducable and consistent) write rates in photo benchmark but are unable to use anything beyond 50 MByte/s in movie mode. And I think it does not depend on SD-card itself.

Another bottleneck, besides SD/CF write speed, is the memory bus speed. There are many image buffers around, besides the LiveView buffer that you see on the screen, including some higher-resolution ones; I believe not all of them are necessary for LiveView, but identifying and stopping them is not very straightforward.

One such workaround is the "Frozen LV" hack, which stops most of the LiveView streams, but keeps the raw one. Does that help with recording speed?

[Danne]

@IDA_ML

Code: [Select]

Danne, could you possibly post here your April 20-th build that you sent to me?Tbh, I can´t remember what build that was. Anyway, if anyone wants to play with it you can send them personally. I´m off doing some other stuff atm.
Basically I use code changes from Levas here:
https://bitbucket.org/daniel_fort/magic-lantern/commits/979c1fc3c37ac8913d8f6ca605889734bf100a67
I then put sd_uhs.mo version into a compiled version from this branch:
crop_rec_4k_mlv_lite_snd

Anyway, really like this module but I can understand why it would be hard to put into nightlies...

[IDA_ML]

Hallo Walter,

Here is the April 20-th build that I have been using with the results reported above:

https://we.tl/gt9YjOeKvt

The link will be active for 7 days.  I  archived it as it was on my SD card, so you have the config file with all my settings. 

I also ran the benchmark test in the movie mode with the above build and the Preview option set to Auto.  I get 56 MB/s write speed in that mode.  Continuous recording works at 2520x1080 resolution/24fps/8...11bit lossless.

I hope this helps.  Please report what you get.

[Walter Schulz]

Thanks! Looks much better now.
2nd test with 47 MByte/s continuous (> 1 minute). First one (same settings) broke after some 20 seconds. Will do some more tests to find out where the limits might be.

Edit: Consistent 54 MByte/s now! No hickups. That's something! Thanks again!

[Danne]

Seemed you right IDA_ML. This build was more stable than my latest. Exchanged the version on my bitbucket downloads section. Beware the 160mhz hardcoded setting. I have no idea how this version will affect slower cards. In other words, Risky business.

[OlRivrRat]

      @ IDA_ML

   Downloaded Your .rar from > https://we.tl/gt9YjOeKvt

Still can only get 5Sec's Max & Scene & Exposure has little effect.

Also, there is no "Settings" Folder in the "ML" Folder in that archive.

& Pretty Sure My SD is a SanDisk, but do suppose I could be mistaken.

& the BenchMarks are very repeatable dependent upon Settings.