Menu

Show posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Show posts Menu

Messages - Volumetrik

#1
The scenes at 0:19 and 0:20 are fantastic. They look straight out of a movie, well done!
#2
I made a detailed thread about this a couple of years ago. Still relevant!

https://www.magiclantern.fm/forum/index.php?topic=24933.0

#3
Quote from: reddeercity on May 22, 2021, 05:32:00 AM
With CS6 AE On my 2013 MacPro Hackintoss  i7 3770k (overclocked to 4.7GHz)  16GB ram , on After Effect I was averaging 3-4 frames/s @ 1080p/24fps
That was from the (4x2TB disk) 6TB Raid5 box thought  ATTO PCIe-x4 SaS card to the SSD
I had .75GB per thread of the CPU (x8) in the multi threading configtion  , it only limited by how much ram you have 
AE loves ram the more the better & the faster it works.
I hope i can get more with the new computer  :D

Interesting, I'll try messing in the multi threaded option in AE. I have 24 threads, so I guess I'll do 1.25gb/thread.

I'll try the 1080p/24 with my current rig : 5900x + 32GB Dual Rank 3733Mhz CL16 + SN850 Pcie4.0 SSD to see what frame rate I get. It would be interesting to compare our results. Maybe I can download your MLV so we can render the same file directly to compare ?
#4
Glad to have you back ! Congratulations on the house ! That's a pretty big machine you got there. After a lot of research, I figured that the SN750 NvMe drives from WD were the top option considering price. I have a 1TB SN750 and it's blazing fast. I also have a regular 1TB SATA SSD for my ML scratchpad files. The Samsung NvMe 970 EVO plus use more power than the SN750 and can heat up quicker. There are frequent deals for SN750's for 150-155$ CAD and I think that's a steal for that kind of performance and storage.

You'll see that MLVFS is slow for 5.7K ana and 3.5k files (5D3) because it looks like it's a single-threaded process and is outputting low amounts of data per sec. Also using almost 100% of one CPU thread. However, regular 1080p, I had no issues with scrubbing. If you dump the MLV contents via MLVAPP instead and access the folder with DNG's, even 5.7K and 3.5k DNG's load up fine and have no scrubbing issues with a SATA SSD.

I became frustrated with the lack of GPU accel on after effects even though I much prefer ACR's rendering of raw files vs Resolve. Even with a 3900x (12 cores & 24 threads), the export was 1-2 fps with 50% core utilization in AE. In resolve though, I could export ridiculously fast at 24-30fps with GPU accel on my 3070 and made things so much easier like scrubbing and color correction. Curious to see what you think about that
#5
I'd have to agree on that ! 5D Mark II can do it pretty well too, but the aliasing on fine detail like distant trees can be distracting.

Plus on the 5D Mark III you can card span on the SD+CF for even longer recordings on full frame 1080p and other modes as well.
#6
Thank you ! Very useful script. I was thinking about implementing something similar but with a copy functionality that takes any drive named EOS_DIGITAL and creates an agglomerate folder with the CF+SD and then calls for MLVFS. Your script is an excellent starting point.
#7
Still keeping that 4K crop w/ Contrast detect AF. No 10-bit in sight. This could have been a software update...
#8
I've had the same problem where it was zoomed in all the way. A restart solved it, but I think that my use of the x10 preview aid from the half shutter press triggered it. The recording of the latest build is rock solid, but sometimes navigating in liveview, I have to take my time and go a bit slower than usual and that seems to not trigger the ''zoom all the way'' issue.
#9
Is there any need for a new 5D3 ?

I've seen here in Montréal, good condition, low shutter-count 5D3's going for 1.3-1.5k cad. I snagged mine for 1000 cad with a non-functional 24-105 that I repaired and later sold, but I consider that a fluke. My 5D3 is super beaten up, rough around the edges and with 107k+90K shutter actuations, and still works super well. I don't think getting a new or open-box 5D3 is worth it in my opinion, when you can save 500 to 700 cad getting a low shutter used 5D3. If you resell it, you get 1.5k anyways, so why incur the $500 depreciation ?

Can't comment on the 24-105, I only repaired it to resell it, but it seemed big and heavy for my use.

5D3 ML video is absolutely fantastic as you know, but the live-view is only truly usable on 1080p. Other modes you have framing live view which then becomes much harder to focus with for hand held work +run & gun.

#10
@Danne @Bender@arsh

Confirmed on my end ! Everything is working well, no compression errors on 10-Bit or 12-Bit.

Over 5 minutes for both bit-depth !! Danne you're the man !
#11
Quote from: Bender@arsch on August 14, 2020, 12:47:49 AM
@Volunters
Interestingly, there is no pop out console while recording?
which firmware do you use?
I use 113 and can rec. up to 125mb/s with liveview and 131mb/s without (UHD mode).

In my case it never stops recording, but I see multiple compression error on frame *

Easily you can close your lens cap for less writespeed, but if you don't see the console message it will not really work.
but maybe you can enable console in ML menu under scripts.

1.1.3 Latest Danne build Jul15 with stock SD_UHS @160Mhz + Spanning +Raw Zebras + Global draw + Digic peaking

I did get the console message, but it kept on recording. (forgot to mention sorry) The compression error message was only for the frame 206 which was blank as shown. I agree that this is not ideal and it partially obscures the liveview, but the recording was okay and was fully available to view in MLVApp afterwards.

So in a sense, I did (kinda) reproduce your error for the compression, but I could not replicate the ''failed'' portion of it. The recording works, albeit for that singular blank frame.

Also, to think of it, I did a day-long shoot with the 5DIII in June and I did get that error (frame compression + console appearing on liveview, but still recording) recording 3K 1:1 present in 2:1 aspect ratio at 10-bit non spanning only on CF recording.

Quote from: has_abandonment_issu on August 14, 2020, 01:08:23 AM
Bender@arsch I get a similar error when recording higher framerates (either 1920 @ 60fps or 1050 @ 48fps) while using card spanning. I am currently using a 95mbp/s sandisk 32gb card. I haven't tested the error aggressively so I'm not sure if it conflicts with other settings.

I also got errors using my 32GB SDHC 95MB/s SanDisk. I think the newer SDXC 170MB/s cards work better in our use-case.
#12
Can't comment on battery differences because I only have OEM canon batteries.

For the cards, the first one you linked is a CFast 2.0 card. Not compatible with the 5DIII.

You need a CF card UDMA 7 rating to get the most out of it. You can't get faster CF cards than 160MB/s anyways.
#13
First test using your settings did a minute with no corruption issues. Had one frame (206) that had a compression error. Frame demonstrated in the screenshot below. Write speed was at only 89.35 MB/s.



Second test with 16:9 10-bit stopped at 12 seconds because it was writing at 120.14 MB/s. I think this one also has one compression error, however I can't recall the frame and I can't see it by playing through the footage.



You can see the file sizes for the second test highlighted. You can also see the sizes for the 1 minute test at 2.35 AR just above.

From my hardware, camera and settings, I have to stay around the 110MB/s to get full continuous footage. That figure might be different for others reading this.
#14
Not to spam your thread @Danne, but I could not reproduce the problems you're having Bender.

3K 1:1 Preset with 2:1 aspect ratio. 160Mhz SD_UHS + Card spanning + Raw Zebras + Digic Peaking

SanDisk Extreme Pro 64Gb 170Mb/s + 1066X Lexar 128Gb CF



I am actually having close no no issues. Thanks so much Danne, with the SD overclocking I can afford to have everything with 14-Bit lossless !
#15
Quote from: Eugenia on August 12, 2020, 12:44:13 AM
And no, I don't accept replies to the type of "this is just open source". ML is an 8 year old project by now. Some form of software maturity should have been evident by now. Also: I've been using Linux since 1999. So I'm not some newbie who doesn't understand OSS. But I'm also --again-- not a loyalist to anyone. I'm a realist, and a very practical person. If something doesn't work adequately, I'm not going to give it props just for being open source. That's NOT a feature in my opinion. It's a political aspect of it, but not a software feature that makes work get done. I rather pay $50 and have it work better.

You illustrated your own solution.

pay $50 and have it work better

No one forces you to use open source software created by a team of volunteers, on cameras which were not initially intended to have any of these features, on a forum where information is freely available to learn.

Your tone comes off like someone who did not receive what they paid for.

Did you pay for any of this ?
#16
Quote from: GullRaDriel on August 09, 2020, 11:19:19 PM
I'm currently testing with a Lexar Professional SD3XCII 10 128GB x1000 with a theoric write speed of 150MB/s . When I'll test I'll also do it with the CF, it will be a Lexar Pro UDMA7 1066x with a theoric write speed of 160MB/s.

Wait and see.

I think you may be referring to a discontinued model of that card. It just so happens that I have a 64GB version of said card.

Can you confirm that you're talking about this one ?


If so, I have tested it multiple times with card reader benchmarks and in camera with & without SD_UHS. My conclusion was that I found it very unstable performance wise. Sometimes I could get 118MB/s (SD_UHS 160Mhz + Spanning) writing for minutes at a time on the 5D3, other times my SD UHS-II card reader benchmarks were sub-standard, like 40MB/s and dropping.

Bad performance, to the point where I was seriously doubting the card's legitimacy. I bought it second hand with the 128GB 1066x 170MB/s Lexar card which runs flawlessly, so can't be sure.

For the reasons stated above, I don't use that card and I don't recommended for our use case.
#17
Quote from: has_abandonment_issu on August 08, 2020, 06:23:15 PM
I see that there's a lot going on here with the writespeed calculations.

You also asked what happens when I try to use SD overclocking. I am using a 95mbps Sandisk Extreme Pro 32gb SDHC card. I have a 128gb 170 coming in the mail on Monday.


I have that same card (32Gb SDHC 95MB/s). It's around 6 years old and I have the same issue. My 64Gb SDXC 170MB/s gives me no problems.

So it would seem that the SanDisk Extreme Pro SDXC 170MB/s cards work great with the SD_UHS patch, but not the SDHC (32gb or less) counterparts.

I just came back from shootings some clips and I was using 3520x1320 14-Bit lossless and 1920x2340 (5.7k) 14-Bit lossless all continuous with green seconds counter. The preview was going from black & white to color depending on the complexity of the scene. I was using ETTR methods where when I saw raw (green) zebras, I backed out 1/3 of a stop and then recorded.

I'm on the Jul15 firmware 1.1.3.
#18
When I use Global draw & RAW Zebras in combination with card spanning + SD overclocking, I get around 112-115 MB/s while recording.

I set my resolution to get me 110 MB/s and I'm usually covered for continuous recording in pretty much all scenarios.

Extreme Pro 64gb 170MB/s + Lexar 128gb 1066X CF card.
#19
Premiere cannot edit really edit the DNG sequence (raw). You totally could use premiere to grade the footage, but the debayering of the DNG's in premiere is atrocious. I suggest you try importing the DNG sequence in premiere and after effets. You'll see that the difference is night and day in terms of colour accuracy, edit-ability and control. The price you pay however, is that after effects renders extremely slowly. I suggest you do a bit more digging on the raw post processing section of this forum to get a better idea.
#20
Quote from: Danne on July 28, 2020, 10:48:00 PM
All modules are autoloaded except for bench.mo. I also plan on adding adtg_gui.mo as an optional module.
Do note that my builds comes from a fork from another experimental build. This is actually experimental on top of another experimental build.


That explains it ! Thanks, I do feel dumb now. From my understanding of the replies on this thread, the card spanning module is in the raw video section ? I cannot find it, even in the advanced menu. Could only find one post mentioning that it was in the Raw video tab, but I don't see it there.

EDIT : Found it. You need the CF card in as well for it to appear. Sorry for bothering.
#21
Not sure if this has been addressed before, but the only module I see when I install the Jul15 build is the Bench.mo module.

I am using the 113 version and I re-installed ML, formated, low level, tried an other card, reset to ML defaults, even created the SETTINGS folder and made a SD_UHS.EN file. I can see and load all the modules with the nightly exp 2018Jul22 build, but for some reason, this build I can't load or see any other module than bench.mo.

A lens is installed, only the SD card is in and the camera is in movie mode.

#22
Quote from: masc on July 09, 2020, 09:39:44 PM
h.265 has burned in white balance and is already debayered and lossy compressed. RAW is untouched sensor data.

To bounce off of what masc said, you're trading the lossy compression of H265 for file size. Raw video, regardless of 10, 12 or 14 bit (even lossless) will produce much larger file sizes than H265. The workflow is also different for RAW and H265. Since H265 is very efficient in compressing the image, editing H265 files have to be GPU accelerated to be edited and will require very large computing power to playback and scrub through. Especially with 4k60 stuff. Raw DNG sequences like our MLV's can be accelerated and scrubbed through with much less computing power.

Finally, you could deliver straight out of camera with H265, but with RAW, you'll always need to post-process it before delivering it.
#23
Thank you ! Honestly a) to d) could be pretty quick to do as the bit-rate factor has already been addressed in the other thread. For pure quality purposes, there's only four ways in which the 5D2 can give different results : Full-sensor readout (like full size liveview low-fps modes), line skipped readout (like 4k anamorphic modes), cropped sensor readout (2k and 3k crop modes) and full-frame modes where it's pixel binned. I have not tried the 48p modes, so I can't comment on those just yet. The continuous and non-continuous tests are very good ideas as I often wonder those things myself.
Quote from: SKecorius on June 04, 2020, 05:01:19 PM
Thank you, your tests are great source of information!

Long time ago, I wanted to see some condensed/one place info on what are the best settings for 5DII for: a) highest quality continuous not cropped recording; b) highest resolution (cropped) continuous recording; c) highest quality highest resolution non continuous recording for both cropped and not cropped; d) 48p and anamorphic modes; e) dual iso.

There  are some limited tutorials online, but all are scattered and not necessarily works for everyone all the time.. I feel this is beyond the scopes of your post, but with your systematic and super organized approach, I believe it could end up in super useful material that would be both interesting and helpful for 5DII users.

Either way, I am excited to see the comparisons between II and III.

All the best!

I'll include some aliasing inducting scenes in the test for sure ! It will be interesting to see the 3x1 vs 3x3 binning in action on the same scenes.

Quote from: allemyr on June 04, 2020, 10:28:10 AM
Thank you,

You can also test outside brick walls and roof tiles from a distance can sometimes show antialiasing quality pretty easy. Please do some 14bit test if you like.
#24
Good idea ! I just printed an ISO 12233 test chart. I'll include that in the test. I'll get my sharpest lens, the 100 2.8L macro.

I'll make a separate thread when I'm done and I'll link it here.

Quote from: yourboylloyd on June 03, 2020, 11:19:37 PM
I'm interested in using the full sensor on both cameras to record. I would like to see the difference between aliasing. If you can do one at your lens's sharpest aperture that would be awesome.

If you could:

  • On the 5D2, shoot at 1856x1248 in 10bit 23.976fps. That should use the full sensor as far as ML could use it
  • Compare that to the 5D3, shoot at 1920x1280 in 10bit. I would like to see if the anti-aliasing and small push of extra resolution makes it better.
#25
Yep, I'll upload those !

1080p full frame shootout between the two will be very interesting.