5D3: 1.1.3 vs 1.2.3 Slow Mo Record Times

[spnsir]

Hey all,

Just for reference. Here's an unscientific look at how many seconds I can record for:


https://www.dropbox.com/s/tkwhcxypjqmfjru/Screenshot%202014-09-28%2023.41.08.png?dl=0

As you can see, 1.1.3 has a very slight edge on 1.2.3 at 60fps, and at 50fps it is a significant difference. Based on prior tests I can record for 2.5 minutes at 50fps on 1.1.3 MLV. All settings were optimal settings with the same GD options turned on for both (cropmarks, raw zebras-off while recording). All shot on a Kb 1066x 64GB card.

[beauchampy]

Interesting..

Out of interest (and if you can be arsed to check) what do you get on 1.1.3 at 50p with a 16:9 ratio?

[noix222]

i also find that version 1.1.3 is faster... i am still using nightly from 10th august. Still the fastest for me with no bugs at all. 

[spnsir]

Interesting..

Out of interest (and if you can be arsed to check) what do you get on 1.1.3 at 50p with a 16:9 ratio?

If I recall correctly, I get about 9 seconds at that ratio at 50p. At 60p I get 4 seconds.

[a1ex]

i also find that version 1.1.3 is faster... i am still using nightly from 10th august. Still the fastest for me with no bugs at all. 

What bugs did you find on 1.2.3, that are not present in 1.1.3?

[Walter Schulz]

He said he found no bugs in this 1.1.3 build (10.Aug.2014). Speed is his main concern, I think. And I would not expect anyone claiming to use an rather old build to say something about bugs in a version he/she is not using.
Come on, a1ex ...

Have you been able to reproduce anything about existing/nonexisting speed differences between 1.1.3 and 1.2.3?

[a1ex]

He said it's faster than 1.2.3 (he was comparing the two versions).

[Walter Schulz]

Spnsir came up with numbers.
Would it be of any use to do some tests with a newer 1.2.3 build? If he wants and has time, of course.

Have no 5D3 at hand, sorry.

[a1ex]

No, the build versions are OK.

The slowdown might be from the dual monitor support code (LiveView is quad-buffered in 1.2.3, even on built-in LCD, but 1.1.3 is triple-buffered).

[spnsir]

Would it be of any use to do some tests with a newer 1.2.3 build?

That is the newest 1.2.3 nightly

[Walter Schulz]

Sorry, of course you're right. Forgot 5D3 1.2.3 has a very different schedule compared to the rest of the bunch. Sticks out here:
https://builds.magiclantern.fm/jenkins/

[beauchampy]

If I recall correctly, I get about 9 seconds at that ratio at 50p. At 60p I get 4 seconds.

Nice. I consistently get 6 or occasionally 7 seconds on 1.2.3

[legreve]

I concur... the 1.23 builds are pretty much useless in terms of 50/60 fps....

Last year I made a music video that was shot entirely in 50fps 1920 with 1.85 aspect, and there was zero issues with continous recording.

I tried just today with the 1.23 and a march 17 2014 build and I couldn't get past 6 seconds.

So... I guess I have to go back to the old firmware.

[rocnoxious]

hi all
i get continuous 50p with the 24 jan version 1.1.3 - 32 gb san disk 160mb/s
nothing since can record more than 2 min (1.2.3 a lot less)
just one bug - if you run out of cf card, the last mlv file will NOT be saved)
hope this helps

[DeafEyeJedi]

Thanks @Audionut -- Benchmark speed test with new KB 128gb 1066x...

1.1.3


1.2.3


Chime in your thoughts regarding this particular KB card -- is it worth to keep?

[a1ex]

To double-check, can you run the same test in playback mode, with the same card, on both 1.1.3 and 1.2.3?

And while you are at it, can you also run the benchmark in the 30p and 50/60p modes? (on both firmware versions)

[DeafEyeJedi]

sure will do @a1ex...If I understand correctly for the playback mode -- is this the one that you watch RAW/MLV files on the live view or you meant that for something else?   

[a1ex]

Much better to use the one where you watch still photos, not videos.

[DeafEyeJedi]

currently BM'ing them per your request -- still unable to put it into 'playback mode' so could you try to be a bit more detailed on that one if possible?

thx

**EDIT**

My fault I think what you meant was for me to have LiveView enabled under Global Draw which states in 'Playback mode' is this correct?

[a1ex]

How do you call this button?

[DeafEyeJedi]

bro I knew that was the button for playback mode but I guess my question was how to actually BM during playback mode only because when I try going back into ML menus by pressing the 'trash button' it asks me to delete picture/video or not...

sorry to be such a foolish nut in here but willingly to help. my apologies, Alex.

**EDIT**

Actually I just randomly decided to press playback right the second after I start the BM process... was this what you were referring to this whole time?

because there was the playback mode in the background...

[DeafEyeJedi]

and I wonder why the write/read speed is SIGNIFICANTLY different!!!

I mean holy smokes...

but then how come we don't get these massive numbers for RAW/MLV shooting?

[Audionut]

IIRC, CPU cycles are needed for data throughput.

Playback mode doesn't require many CPU cycles, raw recording does.  More CPU for raw recording, less CPU available for data throughput.

[DeafEyeJedi]

Thanks @Audionut -- that makes sense. However, I'm still worried that I may have gotten a cheesy version of KB 128gb 1066x in my hands...

Here's the BM comparisons re: KB 128gb 1066x per @a1ex's request (THANKS for clarifying the Playback mode trouble that I was having earlier):

1.1.3 (Playback mode)


1.1.3 (24p)


1.1.3 (30p)


1.1.3 (50p)


1.1.3 (60p)


____________________________________________________________________________________________

1.2.3 (Playback Mode)


1.2.3 (24p)


1.2.3 (30p)


1.2.3 (50p)


1.2.3 (60p)


Obviously RAW/MLV write speeds are slightly over the edge in 1.1.3... my question is do you guys think I should keep this KB 128gb 1066x card?

Thanks again everyone!

[DeafEyeJedi]

Well guys I think I have some interesting news (whether its good or bad) but I decided to run a test to see how long I can record MLV in 23.97p 1080p (MLV Audio ON) for....

I got slightly just over 24 minutes of continuous recording until the 'eXfat Full' pop-up window which indicates that I've ran out of space as to be expected.




Guess I should be OK about keeping the KB 128gb 1066x card... right?

**especially after the fact that I actually enabled an option in ML menu to 'save reserved space' to prevent loss of footage once the card fills up. LOVE it!

**EDIT**

Next test will be 30p & 50p/60p continuous recording times w this same KB card...

[a1ex]

So, you have got:
- roughly the same result in playback mode
- consistently slower results in 1.2.3 in LiveView mode

So far, all the tests indicate a higher overhead in LiveView for 1.2.3 (probably because of dual monitor support).

This is really ugly - I want to get rid of the 1.1.3 branch, but I guess it's not a good idea if the difference in recording times is significant.

Same with RAW vs MLV...

[DeafEyeJedi]

@a1ex I guess that tells me that my card is somewhat not up to par as it should be? Although I ran into an interesting surprise with a pattern occurring during testing in 25p/1080 MLV (w Audio enabled) it gives me a pop-up window indicating 'Card full?' at the 7:55 (7 minutes and 55 seconds) mark... not sure if it's this is related to the limitations of the KB card that I'm using or it's due to the stress of using 25p/1080 mode?

Tried this all four times in a row (twice w card warm-up @ 16mb & twice w-out) and all came out with pretty much the same results...

My gut is telling me to return the KB 128gb 1066x for two 64gb 1066x (but then again I would compensate for the practical size of 128gb.

Argh!

Off I go with testing in 30p/1080 -- to be continue...

[Danne]

+1 for keeping 1.1.3 and raw.

[jose_ugs]

+1 for keeping 1.1.3... even if its just for the slow mo video. I have also experienced a huge difference with 1.2.3 and went back, just for the 50/60fps performance

[DeafEyeJedi]

Thanks @Danne & @jose_ugs for the +1 inputs and that was exactly what i have decided to do last night.

Stick w 1.1.3 & use RAW over MLV (unless Audio is neccessary) regardless of the 50/60p options.

ALSO I'm still on the fence with this KB 128gb 1066x card -- chime in your thoughts?

SJ

[DeafEyeJedi]

Also @noisexx -- I was only able to locate aug 9th 2014 and aug 20th 2014 nightly builds to test the speed writing for 1.1.3... Where did you find your aug 10 2014 build from?

Unless it's more ideal to stick w the sept 26 2014 build (only because of the full_res silent module)

Thanks in advance!

[Danne]

You should try out some limits by finding sweet spots with your card. Mlv with sound 25fps I think was mine with my card 1920x1080(It has been a while since trying). I think I can get my hardware in camera run up to around 110mb tops in raw 3x mode on my card but often my card has to be warmed up and more often it will cruise around 104-105. These are of course not exact measures but I,m confident that my card isn,t the limitation but hardware. (transcend 1000x 128gb card). Also I tend to use mostly 3x crop and dual iso with liveview turned of with raw to get some really good aspect ratios. Slowmo I think I was around 16:9 1728 x (can,t remember) live view turned of to get continouos. Kudos a1ex for keeping the 1.1.3 and raw alive.

[DeafEyeJedi]

Just ran a test with this KB 128gb 1066x card that was pushing 110.6mb's recording in 3x crop mode 25p 2.5k x 1090 MLV (hacked preview & card warm up) for continuous over 3-4 minutes without skipping on 1.1.3 -- Not bad at all!

[Danne]

Sounds like a good card. You shot in mlv? Should be even longer recording times with raw then?

[DeafEyeJedi]

Correct on RAW @Danne will type up a stat sheet of all different fps settings (w and w/o hack) for time varies comparisons regarding KB 128gb 1066x hopefully soon!

[paisleybrew]

Hey guys

I know I'm messaging in an old thread but I have a question that I'd like clarity on.

Is 1.1.3 best for write speeds with mlv and raw? Including audio? I'm a bit confused if 1.1.3 is better for just raw or mlv as well.

Thanks

[jose_ugs]

I speak from experience, and i believe it's mentioned in this thread: MLV (spanned) works a lot better (stable) on 1.1.3
This alone (and a few other) could be the main reason we'll end up having different branches 113, 123, etc...
I don't know what cards you got on you, but 1.1.3 is probably going to work better than 1.2.3.
Nuff said ?

[DeafEyeJedi]

Agreed.

[paisleybrew]

Thanks!

We have Lexar x1066 128GB and 256GB cards. Not sure if that makes a difference.

Could you elaborate what you mean in terms of spanned? Good to know that it is more stable on 1.1.3 as I recently installed that version. I'll have to do some proper tests to be sure though. What advantages does 1.2.3 have, then? HDMI out?

[beauchampy]

There's an option in the raw menu to enable 'spanning'. This will write some raw frames to the SD as well as the CF. So effectively you get the write speed of the CF + SD.

You simply pull the files into the same folder and convert as normal afterwards.

Thanks!

We have Lexar x1066 128GB and 256GB cards. Not sure if that makes a difference.

Could you elaborate what you mean in terms of spanned? Good to know that it is more stable on 1.1.3 as I recently installed that version. I'll have to do some proper tests to be sure though. What advantages does 1.2.3 have, then? HDMI out?

[KMikhail]

Apparently, your camera has to bee freed from other activities in order to take full advantage of spanning - GD should be fully OFF, hacked and it should be 1.1.3 with no hdmi output used.

[jose_ugs]

Apparently, your camera has to bee freed from other activities in order to take full advantage of spanning - GD should be fully OFF, hacked and it should be 1.1.3 with no hdmi output used.

Why's that? Most of us use spanning in order to use GD and peaking or zebras (for example) and have it all working smoothly...

[KMikhail]

Cause level of load of processing scales writes to cards. I can get continuous 1920*1080 @ 30fps with CF+SD, only when I disable pretty much everything, including rec blinking on LV. It is impressive (compared to what was otherwise), but not suitable for everything. When GD is on, combined CF+SD isn't that much different from CF alone (both fastest available). The definite advantage of CF+SD is max recording time - not terribly better, but always there.