What is the Maximum write speed for 5D3?

Started by theBilalFakhouri, September 09, 2022, 11:43:18 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

theBilalFakhouri

Quote from: vastunghia on September 09, 2022, 11:12:47 AM
Just ran a new test, with Global Draw disabled I get continuous rec. and

  • Reported write speed: ~89 MB/s CF, ~60 MB/s SD, total ~148 MB/s
  • Actual write speed (calculated from final MLV file size): 90.1 MB/s CF, 60.7 MB/s SD, total 150.8 MB/s
So reported write speed looks pretty reliable to me, at least on my 5D3.

Also, I do not understand if my results are in line with your expectation. When you say "in LiveView" you mean with Global Draw enabled, right? So cannot compare these with your 135 MB/s I guess.

Thanks for your clarifications

Sergio

Thanks for running the test.

To check write speed accurately with RAW video recording:

-Make sure your recording is continuous, at least record 4 minutes, this way we can ignore buffer impact on write speed or at least minimize it
-Share your FPS, number of recorded frames, MLV and M00 files sizes in bytes
-If you want to use Lossless, make sure your scene is 100% static in a controlled environment

LiveView and Global Draw are two different things. Pretty sure you already know what LiveView means. When I say in LiveView, Global Draw can be either disabled or enabled, for my test it was disabled.
I will redo my tests again to double check, also I will use 1.1.3 version instead of 1.2.3, because all my tests were using 1.2.3.

Walter Schulz

We need file size in bytes because there is a common miscalculation.
In ML calculations 1 MByte = 1024 x 1024 Byte = 1,048,576 Byte. This is how file size in Windows is displayed, too.
In macOS 1 MByte = 1000 x 1000 Byte.

Therefore some 4.8 percent difference may occure just by the way your OS is handling file sizes.
Difference will get worse in GByte (7.37 percent) and TByte (almost 10 percent).

vastunghia

Thank you Bilal.

Referring to the test I performed earlier:

SETTING:

  • Camera (5D3 1.1.3 with ML build Nightly.2022Apr07.5D3113) was mounted on tripod and faced towards building with sky in clear sky conditions (definitely static)
  • I recorded 1' 17" -- then I stopped it, I thought it was enough
  • I used the 3.5K 1:1 centered x5 preset, 14-bit lossless, 3584x1500, 25 fps, with card spanning, more hacks (lvface + aewb) and one more hack enabled, SD overclock @160, Global Draw disabled
  • I exposed as close as possible to the right, just before triggering the dreaded automagical stop

RESULTS:

  • 1932 frames
  • file.MLV (CF): 6965015552 bytes
  • file.M00 (SD): 4692122624 bytes

MATH:

  • 150.8 MB/s (with unitary prefixes, i.e. 1000 Bytes per kB etc)
  • 143.9 MB/s (with binary prefixes, i.e. 1024 convention, thanks Walter for reminding me of this issue)

Regarding the impact of buffer I think that 11.7 GB (pardon, 10.9 in binary fashion ;)) of footage should be enough?

Sergio
5D3 for video
70D for photo

theBilalFakhouri

Thanks!

But, I would like to see numbers when you record 4 minutes at least.

vastunghia

Quote from: theBilalFakhouri on September 09, 2022, 12:36:24 PM
But, I would like to see numbers when you record 4 minutes at least.

Sure. But will have to wait. Not feasible today.

Just to understand: trying to minimize buffer impact further?
5D3 for video
70D for photo

theBilalFakhouri

Quote from: vastunghia on September 09, 2022, 12:50:01 PM
Just to understand: trying to minimize buffer impact further?

Yes, and to make sure you have continuous recording. Already said that in first post.

vastunghia

Ok, in the end I managed to find 10 minutes of spare time. So here we go again.


SETTING:
Same as above


RESULTS:

  • 9016 frames (I stopped the recording as close to 6 minutes, i.e. 9000 frames, as I could)
  • file.MLV (CF): 33413230080 bytes
  • file.M00 (SD): 22046281728 bytes

MATH:

  • 153.8 MB/s (with unitary prefixes, i.e. 1000 Bytes per kB etc)
  • 146.7 MB/s (with binary prefixes, i.e. 1024 convention)

It turns out I was able to expose even better to the right than before without hitting the cam limits 8)

As a matter of fact, these real-life results are lower bounds of the theoretical max write speed.

Looking forward to reading your comments. Of course would be happy to share original 52 GB MLV files for anyone willing to double check ;)

Sergio
5D3 for video
70D for photo

Walter Schulz


theBilalFakhouri

Thanks!

I think I will have to make tests using 1.1.3 firmware. Will check it soon, maybe later today or tomorrow.

ShittyWebsite

CF: 64gb SanDisk Extreme 120mb/s
SD: 64gb SanDisk Extreme Pro 170mb/s
Build: lastest anamorphic (with the lv hacks improvements)

I couldn't get anything higher than:
CF 70mb/s
SD 50mb/s
(Killing global draw)

I can't get 3.5k 14bit lossless or 12bit lossless continuous
10 bit is continuous, manually stopped 4 minutes later (118mb/s)

I can't copy the files right now, I'll do later


Many people are getting higher speeds, guess my CF is too slow

Walter Schulz

If you benchmark your card using a decent cardreader you should see write rates around 70 MByte/s.

Grognard

Quote from: vastunghia on September 09, 2022, 12:28:49 PM
  • I used the 3.5K 1:1 centered x5 preset, 14-bit lossless, 3584x1500, 25 fps...
How did you get 25 fps? With fps overdrive enable?

Grognard

Quote from: ShittyWebsite on September 09, 2022, 07:43:05 PM


I can't get 3.5k 14bit lossless or 12bit lossless continuous
10 bit is continuous, manually stopped 4 minutes later (118mb/s)


Your CF is the issue. 12 bits is continuous and 14 bits "almost" in every situation sometimes it could stop but is rare.

vastunghia

Quote from: Grognard on September 09, 2022, 07:58:32 PM
How did you get 25 fps? With fps overdrive enable?

Actually, I'm pretty sure I didn't have to do anything at all. 25 fps is the default frame rate with the 3.5K 1:1 centered x5 preset.

S
5D3 for video
70D for photo

ShittyWebsite

Quote from: Grognard on September 09, 2022, 08:01:56 PM
Your CF is the issue. 12 bits is continuous and 14 bits "almost" in every situation sometimes it could stop but is rare.

What CF do you use?
I guess the SanDisk Extreme pro would be better, I'm just not sure how better would be, if it's just less than 10mb/s maybe it's not worth

Without card spanning I can get a bit higher cf speeds (85mb/s or something)

Walter Schulz

We have write benchmark numbers (in cam) beyond 117 MByte/s for Extreme Pro 64 GB (1066x).

theBilalFakhouri

Same crop mode as vastunghia test, 160 MHz, 15-5-2022 Danne build, 1.1.3 firmware:

-Small hacks enabled
-Two more hacks lvface + aewb
-One more hack enabled
-Kill GD used
-ML Global Draw off
+ Preview set to Frozen LV (this wasn't mentioned by vastunghia, but I guess it was used?)

My CF: Sandisk Extreme PRO 64GB 160 MB/s UDMA7
My SD: Sandisk Extreme PRO 64GB UHS-I U3 170 MB/s

I ran six CF + SD benchmarks: 1 , 2 , 3 , 4 , 5, 6 (using custom bench.mo to turn on the hacks)

Average write speed ~145 MB/s in LiveView!
So yeah, vastunghia was correct :D

Leszek Wysocki

https://www.flickr.com/photos/147176311@N08/52352074115/in/dateposted-public/

On FW 1.23 I have SD 75.6 MB / s + CF 78.9 MB / s
CF - Kingston Canvas Focus 128GB
SD - SANDISK Extreme PRO 256GB (170MB/s)

PS. Test in photo mode, and from what I can see you do a test in video mode, so mine does not bring anything.
5D Mark III 1.2.3 + Canon EF 24-105 mm f/4L IS USM
CF Kingston 128GB Canvas Focus
crop_rec_4k_mlv_snd_isogain_1x3_presets_2022May15

Walter Schulz

Just benchmark numbers won't be sufficient in this thread. We need RL recording numbers for 4 minutes (and up)!
Then calculate
Add both byte numbers and divide by 1048576
Divide this number by recording duration in seconds.

Would be fine to give some details about recording mode, settings. See Bilal's last post.

We have established a new limit at around 145 MByte/s. And it seems only possible with firmware 1.1.3.

vastunghia

Quote from: theBilalFakhouri on September 12, 2022, 02:03:24 AM
-Kill GD used
+ Preview set to Frozen LV (this wasn't mentioned by vastunghia, but I guess it was used?)

Negative, in fact did not mention these two in my post.

Quote
So yeah, vastunghia was correct :D

I guess seeing is believing ;D

Had a busy week-end, in which I needed a reliable build to do some magic shooting -- so still haven't found the time to test the new trick. Hope I'll be able to do that soon.

Ciao

Sergio
5D3 for video
70D for photo

vastunghia

So I repeated same test as before with new sd_uhs module by Bilal, both @160 and @240 (6 minute footage in both cases).

Results:

@160:
SD: 58.6 MB/s
CF: 88.7 MB/s
Total: 147.3 MB/s

@240:
SD: 68.4 MB/s
CF: 77.2 MB/s
Total: 145.6MB/s

NB: for the avoidance of doubt -- all reported speeds in binary prefixes, i.e. 1024 Bytes per kB etc.

Of course difference in total write speed is not relevant and for sure it is due to not perfectly controlled recording environment.

No particular issue with this build. However, I just limited myself to these two (nearly) controlled-environment tests so far, will need time to test general stability in real life.

Would love to repeat these tests with 1.2.3. Bilal, did I get it right that apparently these speeds are achievable only with 1.1.3? And if so, any clue why?

As a more personal note, sounds like a confirmation for me that I should stick to 160, as this makes the filling rates of my two cards similar (CF being twice the size of SD). Oh, well.

Let me also remark that reported write speed *is* reliable in my tests, apparently contrary to prevailing belief.

Thank you

Sergio
5D3 for video
70D for photo

theBilalFakhouri

Quote from: vastunghia on September 12, 2022, 09:58:53 AM
Negative, in fact did not mention these two in my post.

I don't think I am able to reach ~145 MB/s without Frozen LV.

Quote from: vastunghia on September 12, 2022, 03:42:32 PM
Would love to repeat these tests with 1.2.3. Bilal, did I get it right that apparently these speeds are achievable only with 1.1.3? And if so, any clue why?

I think 1.2.3 can reach same result as 1.1.3 + a bit slower, maybe ~3 MB/s slower because of LiveView implementation changes in 1.2.3.




I will do more tests later.

vastunghia

Quote from: theBilalFakhouri on September 12, 2022, 07:45:39 PM
I don't think I am able to reach ~145 MB/s without Frozen LV.

Weird.

Thanks for the feedback, will definitely try 1.2.3 asap, though that won't happen before Friday I'm afraid.

S
5D3 for video
70D for photo

theBilalFakhouri

@vastunghia Could you run "Memory benchmark (1 min)" on your 5D3 using 1.1.3 firmware in Play mode?

vstrglv

5D3 1.1.3 with Danne build crop_rec_4k_mlv_snd_isogain_1x3_presets_2022May15.5D3113 + Bilal patch
3.5K 1:1 centered x5 preset (3584x1730, 23.976fps), 240MHz,180sec
My CF: Sandisk Extreme PRO 256GB 160 MB/s UDMA7
My SD: Sandisk Extreme PRO 128GB UHS-I U3 170 MB/s(made in China,NoBP2019...)
-Small hacks enabled
-Two more hacks lvface + aewb
-One more hack enabled
-Kill GD used
-ML Global Draw off

1. 10-bit lossless
-Preview -Framing
SD-10591223kB
CF-11267126kB
((10591223+11267126)*1024/1048576)/180=118.59MB/s

2.12-bit lossless
-Preview -Framing
SD-11129842kB
CF-12693299kB
((11129842+12693299)*1024/1048576)/180=129.25MB/s

3.14-bit lossless
-Preview -Frozen LV (Framing is not continuous)
SD-12451466kB
CF-14141708kB
((12451466+14141708)*1024/1048576)/180=144.28MB/s
Canon 5D3,1.1.3; Canon EOS M,202,  CF-SanDisk Extreme PRO,160MB/s, 256GB, SD-SanDisk Extreme Pro, 170MB/s, 128GB.