UHS-I / SD cards investigation

[Walter Schulz]

Pretty sure it's not fake, fake cards write speeds maxes out at 20-25 MB/s

Source?

Also my friend got same SD card version from same local store nearly the same time, his card appear to maxes out @ 55 MB/s on PC, on camera with the overclock it gives the full speed.

Which proves what?

I think only some of Sandisk Extreme PRO UHS-I U3 95 MB/s cards have this issue, probably I will get the maximum write/read speed on PC using a Sandisk card reader or on another OS.

Source?

[theBilalFakhouri]

Source?

I had three fake Sandisk Extreme PRO cards before, all of them had ~20-25 MB/s write speed (or a bit more), source is coming from my own tests
Also the fake cards didn't accept the overclock.

I am not sure if there are fake Sandisk Extreme PRO cards which have same performance as authentic ones (at least on camera). If this is the situation here, that's would be another story . .

Which proves what?

The local seller have imported many affected cards in the same time, probably Sandisk had bunch of these affected cards at that time.

Source?

Idea from a mind, I am guessing.

[z56667895]

Any setting errors cause slow writing speed?

My RAW recording can only record about 2 seconds in any mode

I don't know what caused it
My card is Sandisk Extreme PRO 170MB/s 128G

No problem testing the memory card on the computer

[theBilalFakhouri]

This card appears to accept 240 MHz overclock (tested on 700D by a user interacted with me on FB)



-No overclock:
Write 35.2 MB/s
Read 43 MB/s

-160 MHz:
Write 52 MB/s
Read 74 MB/s

-192 MHz:
Write 58 MB/s
Read ~84 MB/s

-240 MHz:
Write ~62 MB/s
Read 106 MB/s

-All previous tests was in play mode, we didn't test write speed using card reader on PC.
-We recorded many clips, it didn't switch to 48 MHz, seems like a stable card with 240 MHz overclock.

[Walter Schulz]

There is a bar code on the rear of the blister pack and a string like "SDSQUSC-128G-ANCMA" above. This is the only identifier giving valid information about type and revision. Are you able to retrieve this string? Would be helpful to get a grip on working cards.

[Danne]

This card appears to accept 240 MHz overclock (tested on 700D by a user interacted with me on FB)



-No overclock:
Write 35.2 MB/s
Read 43 MB/s

-160 MHz:
Write 52 MB/s
Read 74 MB/s

-192 MHz:
Write 58 MB/s
Read ~84 MB/s

-240 MHz:
Write ~62 MB/s
Read 106 MB/s

-All previous tests was in play mode, we didn't test write speed using card reader on PC.
-We recorded many clips, it didn't switch to 48 MHz, seems like a stable card with 240 MHz overclock.

Behaviour on 5d3?

[theBilalFakhouri]

There is a bar code on the rear of the blister pack and a string like "SDSQUSC-128G-ANCMA" above. This is the only identifier giving valid information about type and revision. Are you able to retrieve this string? Would be helpful to get a grip on working cards.

That was:
SDSDUN4-128G-GN6IN

Yeah, it's better to list these cards in a spreadsheet in Google Drive or ML Wiki

Behaviour on 5d3?

That card was tested by a user with his 700D, not me, he doesn't have 5D3

[theBilalFakhouri]

I did some experiments lately with SD overlocking stuff,

After calling sdSoftReset the camera switch from 96 MHz to 48 MHz mode, and AccessMode from 7 to 3, resulting in 21 MB/s write speed, changing preset to 96 MHz makes the write speed ~36 MB/s instead of ~41 MB/s , changing AccessMode from 3 back to 7 makes the write speed ~41 MB/s (now we got back original write speed), from there I could overclock to 160 MHz without a problem, but doing an overclock to 192 MHz/240 MHz makes the camera switches back to 48 MHz (21 MB/s)

After some research I found something (probably related to Bus speed configuration) which changes its value from 0 to 1 after doing sdSoftReset and especially after hitting recording button (after doing write/read processes), I tracked what part of code does that in 700D ROM, and found out it's coming from sdReadBlk and sdWriteBlk, well let's force that "something" value from 1 to 0 using this code (for 700D):

Code: [Select]

patch_instruction(0xff74c814, 0x13a01001, 0x13a01000, "0xff74c814"); // sdWriteBlk
patch_instruction(0xff74cae0, 0x13a01001, 0x13a01000, "0xff74cae0"); //sdReadBlk
-->I could overclock to 192/240 MHz presets (after a sdSoftReset/after switching to 48 MHz) (didn't blow up the card or camera . . managed to record several clips and +50 GB of data, everything was stable)

I am not really sure what that "something" does, do you remember these initial 192/240 MHz presets (which were and still not stable):
https://www.magiclantern.fm/forum/index.php?topic=12862.msg228447#msg228447
https://www.magiclantern.fm/forum/index.php?topic=12862.msg228460#msg228460

Above patch forced these none-stable presets to work, and it didn't switch to 48 MHz . . but I experienced speed drops then --> Card can't be accessed a.k.a "Card full" issue, well we knew those presets weren't stable and we are not using them, but we know current 192/240 MHz presets are stable (on 700D/6D)

--> So from above experiment and results it might be there some Hope to force 240 MHz on EOS M, and 192/240 MHz on 70D, please try the following sd_uhs.mo (for 70D/EOS M):

Download: https://drive.google.com/file/d/1DGa2nhSdgu8tSY2Yy22tZQR3PmYpwB4i/view?usp=sharing

If it worked --> you will have stable 192/240 MHz overclock
If it didn't work --> you might experience speed drops while recording and emergency stops, then "Card full" issue might appear (maybe something else?)

-Do this test at your own risk
-If it worked, it's might not be the correct way to force these presets

Edit: Doesn't work well, not stable at 240 MHz for EOS M, don't use it.

[Danne]

Hehe, nice work man. Will test in the evening 

[masc]

I tried your module on my EOSM with SanDisk Extreme Pro 95MB/s 64GB:
240MHz: 21MB/s (R/W)
192MHz: 86MB/s (R) 79MB/s (W)

[theBilalFakhouri]

@masc

Hmmm, could you share a picture for "Memory patches" from "Debug" tab

[masc]

could you share a picture for "Memory patches" from "Debug" tab

This is booted with 240MHz setting.

[theBilalFakhouri]

It seems the patch didn't apply, will check why

[theBilalFakhouri]

Okay, it was a typo, updated the link, please try this:
https://drive.google.com/file/d/1DGa2nhSdgu8tSY2Yy22tZQR3PmYpwB4i/view?usp=sharing

[masc]

I tried again. With 240MHz setting, I get a console-like output when starting the cam. Recording 12bit 5K frtp works without problem. The benchmark is... strange...:
W 96.1MB/s
R 889.6MB/s
W 134.7MB/s
R 4785.0MB/s
LOL... really? First try had even bigger values.  (While the benchmark screenshot is all black.)
Recording 14bit lossless 5K frtp 1440x1860 was working with green indicator all the time.

[theBilalFakhouri]

Cool, it didn't switch . . but

These big values happens probably due to instability, could you try recording 1736x976 @ 23.976 FPS in 14-bit uncompressed, press Info to toggle to Canon Preview, try to record as long as possible, what is the reported write speed? is there an emergency stops? do you have "Card full" issue?

[Danne]

Also verify by checking the recorded files even after getting those wacky numbers. Shouldn't give you normal files but who knows what works with this hack .

[masc]

... could you try recording 1736x976 @ 23.976 FPS in 14-bit uncompressed, press Info to toggle to Canon Preview, try to record as long as possible, what is the reported write speed?

Sorry... stupid question...: how to record uncompressed? I use Danne's Feb build (I think it is not possbile here) - do I need the standard nightly for that? And where to find the "reported write speed"?

[Danne]

I think I turned off the uncompressed alternative in my version. But do recordings work in general with 240mhz patch?

[masc]

I think I turned off the uncompressed alternative in my version. But do recordings work in general with 240mhz patch?

Yes, worked fine. I recorded 14bit lossless 1440x1836 for over a minute, indicator always green. File worked fine in MLVApp. Did not see any pink frame or something else...

[Danne]

Interesting. And it kept working even after you were getting those wacky test numbers?

[masc]

Interesting. And it kept working even after you were getting those wacky test numbers?

Tried again for this: after the benchmark I get a "File create error".

[Danne]

I see. So maybe it´s the usual lucky first recording and then it breaks. Too bad I have no battery in my cam and charger at home.

[theBilalFakhouri]

Sorry... stupid question...: how to record uncompressed? I use Danne's Feb build (I think it is not possbile here) - do I need the standard nightly for that? And where to find the "reported write speed"?

Not sure how to do it from Danne build, use the official crop_rec_4k build:
https://builds.magiclantern.fm/jenkins/job/crop_rec_4k/80/artifact/platform/EOSM.202/magiclantern-crop_rec_4k.2018Jul22.EOSM202.zip

You can change bit-depth from mlv_lite submenu, maybe you need to use Movie crop mode (lower the resolution to 1736x976) because official build doesn't have mv1080 patch. I mean reported write speed by mlv_lite, like this:



I want you test 1736x976 @ 23.976 FPS in 14-bit uncompressed because it has fixed write speed which is 67.8 MB/s , and recording should be continuous only with 240 MHz preset, look out for speed drops during recording . .

[masc]

Tested a little more: recording as long as possible with 5K frtp setting from above:
1st file stopped after 11.8GB, 4272frames, 2:58min. No error visible in file.
2nd file stopped ("card full", while 36GB are still left) after 11.3GB. I can't open the file, because of error "Invalid blocksize '0', corrupted file".

Edit: 11,8GB*1024/178s = 67,88MB/s.