UHS-I / SD cards investigation

[70MM13]

yeah, no more benchmark for now!
i'll just record and look at the readout

i thought this would be more convenient - ha!

the card is definitely fine.  i just copied a video to/from it on the pc and it worked perfectly, and the read speed is definitely super fast!  1GB file copied before the speed could be determined

[Danne]

Formatting card with both cards in is probably not good. Guessing but probably it tries to find stuff on cf card.
Also formatting sd card when patched in cam seems even worse. Reports here with broken cards as result.

[70MM13]

there was no cf card inserted at the time, but you're right, formatting while patched was probably a bad idea

[Danne]

I guess then that patch automation will interfere upon reinstall. Should look into fixing that.

[70MM13]

excellent guess!
i have never done this procedure before under any circumstances, but it did seem like something went wrong during the automated magic lantern reinstallation.

sorry, i don't want to repeat the "test"!

[Levas]

Too curious, had to check.

I'm using the first sd_uhs module I uploaded yesterday. (based on older automatic version of Danne's sd_uhs module, no pauzes no messages on screen)
Now go into Canon menu, select format card, keep ML.
And go 

Ok, works, card got formatted, ML got restored and camera restarts and behaves normal, can record raw video with ~80Mb/s.
So the automatic version without pauzes and messages seems safe to have enabled while formatting card in cam.

Although A1ex did mention somewhere, that during the patching process, there must be absolutely no card access, otherwise you could end up with card corruption.
So Danne is probably right that it went bad during formatting and using patch 2 on the 5d3 

[70MM13]

bravery!

[Danne]

I undefined restoring after format for the 5D3 and put up new builds.

Regarding eosm I heard conflicting reports about sandisk extreme pro 95Mb/s cards that don´t work with latest patch. masc reports not working, jiphop on the other hand that it works. So far sandisk extreme pro 170Mb/s seems stable.

Sorry guys, very little time to put into this atm...

[Levas]

Idea for the future of sd_uhs module, a submenu with different presets.
Option 1 = classic sd-uhs settings
Option 2 = Patch 1 settings
Option 3 = Patch 2 settings

Choice is saved with ML settings and patch is automatically done at startup.

[Danne]

Cool. That shouldn´t be too hard to do.

[names_are_hard]

If you wanted to get fancy, you could store the verified acceptable speeds for a given card on the card itself.  Then, when the module is activated, read the settings for the patch from the card.  Since it's likely that different cards fail at different speeds.  Have a benchmark mode that tests then records the best settings onto the card.

[Danne]

Not that easy. Different reports with 95Mb extreme pro cards. One working the other didn´t.

@Levas. Here´s a little something selecting and storing different patches:


Commit:
https://bitbucket.org/Dannephoto/magic-lantern_jip-hop/commits/aa513d18485cc6d20d4d19771b902c880c38a8fd

Eosm build updated, usual location.

EDIT: Updated again so it starts with 160Mhz patch upon install.

[names_are_hard]

Yeah, that's what I guessed.  The clean way is to measure each card to determine what works, then record the measurements onto the card itself, so you know what works for next time.

[Danne]

But even so, measuring includes trying to patch and then run code that sometimes works, sometimes not. Testing which also includes some safety numbers canon reverts to reducing card to around 20Mb/s etc requiring restarting cam. Many variables and buggy to say the least. But I get your point.

[masc]

Thanks for the build, Danne. With my SanDisk Extreme Pro 64GB 95MB/s and EOSM I get the following:
- 160MHz


- 192MHz

[Danne]

Yeah. No luck with that card. Jiphop got it better. Well good with reports.

[names_are_hard]

Yeah, it's always a lot more work to do it cleanly.  It's a suggestion only, since I won't be doing the work

If you start the testing with the slowest speed, it doesn't sound too hard to detect if the "safety feature" kicks in, and if so record the prior speed as "probably good".  Then maybe power cycle the cam.  Might be more reliable if you make the benchmark longer, too.

[Danne]

Thanks but I am not doing it either. Not worth it imo. But Levas maybe want to get his hands dirty.
Besides. All is wip. Numbers will probably change again soon.

[theBilalFakhouri]

Thanks for the build, Danne. With my SanDisk Extreme Pro 64GB 95MB/s and EOSM I get the following:
- 192MHz

Yeah. No luck with that card. Jiphop got it better. Well good with reports.

Just tested Danne sd_uhs.mo @ 192 MHz on 700D, same problem, commenting out these lines solved the problem, please re-test with these changes

Code: [Select]

    /* power-cycle and reconfigure the SD card */
    //SD_ReConfiguration();
   
    /* enable SDR104 */
    patch_hook_function(sd_set_function, MEM(sd_set_function), sd_set_function_log, "SDR104");
    //SD_ReConfiguration();
   
    if (sd_overclock == 1) memcpy(uhs_vals, sdr_160MHz, sizeof(uhs_vals));
    if (sd_overclock == 2) memcpy(uhs_vals, sdr_192MHz, sizeof(uhs_vals));
    if (sd_overclock == 3) memcpy(uhs_vals, sdr_240MHz, sizeof(uhs_vals));
Comment out both SD_ReConfiguration();

[Danne]

New build up according to BilalFakhouri suggestion. Maybe works better @masc?
https://bitbucket.org/Dannephoto/magic-lantern/downloads/crop_rec_4k_mlv_snd_raw_only_2020Jul04.EOSM202.zip

[743v04]

Tested working in 170MB/s mode with the Sandisk Extreme Pro 128GB 170MB/s version on EOSM. Ran the 2.5k preset in an overexposed frame for a solid 30 seconds or so. I can run a benchmark if anybody would be so kind to let me know how to do that 

[theBilalFakhouri]

I can run a benchmark if anybody would be so kind to let me know how to do that 

Load Bench.mo module, restart the camera, now you have "Benchmarks" in "Debug" Menu press it --> Card Benchmarks --> Quick R/W benchmark (1 min), then press PLAY button to get the best results, feedback

[743v04]

Thanks theBilalFakhouri but unfortunately it looks like the build Danne provided does not allow for additional modules to be included. Adding the bench.mo from the EOSM nightly build seems to not allow any of the modules to load correctly. I saw Walter Schultz's mention of another method but I don't see that option available either. I guess I will just wait for now 

[theBilalFakhouri]

Ah, Danne has removed bench.mo from his build

Grab bench.mo from this build:
https://builds.magiclantern.fm/jenkins/job/crop_rec_4k/80/artifact/platform/EOSM.202/magiclantern-crop_rec_4k.2018Jul22.EOSM202.zip

[743v04]

Each of these were done with the 192MHz mode enabled using the same 128GB Sandisk Extreme Pro 170MB/s card:








And for reference, here is a benchmark done with the build I had been using prior to this newest one with that same card:



Solid improvement, amazing work!