UHS-I / SD cards investigation

[OlRivrRat]

UHS-I/SD.mo Benchmark on 64GB SanDisk ExtremePro in SL1





https://photos.app.goo.gl/8s9d36uXVhxsiAyp1

[alpicat]

I tried Danne's build from 17th April for the EOS M and it works nicely. I'm using a Sandisk extreme pro 95mb/s 64gb card, which goes up to 55 mb/s write speed max. Seems a bit low for this card but still allows continuous recording.

Here's a test video I've done with it:

[IDA_ML]

This SD-interface overclocking feature is amazing!  I have been using it on a regular basis to record more than 60 GB of footage at 12-bit lossless, 2520x1080 resolution and 24 fps where I get continuous recording with sound on the 100D.  In my opinion, this is one of the most revolutionary developments of Magic Lantern since its introduction almost 7 years ago.  Here is a another test video that I shot on the 100D:

https://vimeo.com/266871797

If the development could be set forth towards implementation into all ML-supported cameras and improving stability, this will be well worth the effort keeping in mind the clean moire-free video and its superb quality at 5x-magnification, comparable to the video quality of high-end professional RAW-video cameras.

[Danne]

At least hardcode memory patches upon startup, maybe on 100D and eosm to start with? Is it safe enough? Seems like it. I also tested the hardcoded version on a slow card and it was skipping patching so that is good if running fast cards and suddenly changing to a slow card.

[IDA_ML]

Danne, I tested only with the SanDisk Extreme Pro (95 MB/s), since I do not have a slower card at the moment.  Your latest hardcoded build that allows up to 42 fps works very well if I do not switch modes too often.   As soon as I switch from 5x-magnification to Movie Crop Mode for example, the 100D freezes, I have to pull battery out, restart camera, reload modules, run the overclock test and only then I can continue working. 

As far as safety is concerned, I cannot comment on that but I never noticed a problem.  Even after a crash, all my previously recorded clips remained perfectly intact on the card.   I never lost a clip, unless the crash happened during recording.  This happened only once or twice and I cannot reproduce it but the reason may have been a battery level close to running out of power where supply voltage may fluctuate and drop below critical level for short periods of time during recording.   

Overall, I am very satisfied with the overclocking feature.  Again, I think, it's revolutionary!

[alpicat]

Thanks @IDA_ML - your instructions on how to set this up on the 100D were really helpful and got me up and running on the EOS M.

I've also only tested the Sandisk extreme pro 95 MB/s on eos m and in my case haven't suffered a crash or any problems with footage yet, though I'm only shooting 24fps. Maybe for safety it would be nice to make a list of cards that work fine with this build... and recommend that those cards be used?

@Danne I agree it would be nice to focus on the 100D and eos m first. They're the cheaper cameras and people seem to be having success with them.

I'm particularly interested in the eos m because it's the only magic lantern camera on which we'll be able to use a speed booster soon (Viltrox seem to be planning to release a proper EF-M to EF focal reducer in the summer). That means that in 5x zoom mode, it will be possible to go wide angle using aps-c lenses + speed booster.

If we want to go further, in the "crop_rec on steroids" thread, @theBilalFakhouri spoke about bypassing the resolution limitation in the 5x zoom mode. See here: https://www.magiclantern.fm/forum/index.php?topic=19300.msg197870#msg197870
Now with SD interface overclocking, reaching those higher resolutions seems more feasible? It would be nice to be able to shoot closer to 16:9 in 2.5k. He was testing on a 700D and one of the resolutions he was able to get working was 2520x1386 at 23.976fps, which sounds incredible.

[Danne]

@alpicat
Did you try any changes with adtg module to increase resolution? Big post and sure would take some time and fiddling to get it right.

[IDA_ML]

If we want to go further, in the "crop_rec on steroids" thread, @theBilalFakhouri spoke about bypassing the resolution limitation in the 5x zoom mode. See here: https://www.magiclantern.fm/forum/index.php?topic=19300.msg197870#msg197870
Now with SD interface overclocking, reaching those higher resolutions seems more feasible? It would be nice to be able to shoot closer to 16:9 in 2.5k. He was testing on a 700D and one of the resolutions he was able to get working was 2520x1386 at 23.976fps, which sounds incredible.

I fully agree with you, Alpicat.  A resolution ot 2520x1386 would be a dream come true for the 100D which has the same senzor size and resolution as the 700D.  I have been using the 7D at 2520x1200 resolution a lot lately and believe it or not, the 120 pixel larger vertical resolution of this camera, compared to the 100D, makes a huge difference in the overall vision of the video.  A 16:9 vision at 2520x1386 would be so much better ... But I am dreaming again.

I have a question for you.  You mentioned the speed booster for increased view angle with EF-S lenses on APSC-sensor based mirrorless.  Would it be possible to use such speed boosters for the same purpose but with 1,6x crop sensor DSLRs operating in the LifeView (video recording) mode, with full-frame lenses, of course?  If not, why not?  And if yes, are such speed boosters available?  Sorry for the stupid question but I have never used and even seen a speed booster sofar.

[alpicat]

@alpicat
Did you try any changes with adtg module to increase resolution? Big post and sure would take some time and fiddling to get it right.

I haven't tried any of this since I don't have any technical knowledge. However, will research this adtg module and see if I can understand it. Looks like there's quite a bit of info about it in the crop_rec on steriods thread. @theBilalFakhouri seemed to get quite far with this on his 700D, and wondering if what he discovered would also apply to the EOS M and 100D, since they're pretty similar cameras... although maybe this discussion is a bit off topic in this thread!

I have a question for you.  You mentioned the speed booster for increased view angle with EF-S lenses on APSC-sensor based mirrorless.  Would it be possible to use such speed boosters for the same purpose but with 1,6x crop sensor DSLRs operating in the LifeView (video recording) mode, with full-frame lenses, of course?  If not, why not?  And if yes, are such speed boosters available?  Sorry for the stupid question but I have never used and even seen a speed booster sofar.

If you're shooting with the full aps-c sensor area, adding a speed booster would give the EOS M the same effective field of view and DOF as a full frame sensor, whilst increasing the lens aperture by 1 stop. At the moment you can get cheap Chinese speed boosters on ebay which have poor optics and no electronic contacts. However, Viltrox told me in an email that they will be releasing a proper speed booster for EF-M mount in around 3 months or so. That's exciting. It'll turn the EOS M into a full frame camera.

[IDA_ML]

It'll turn the EOS M into a full frame camera.

Well, my question was actually about APSC-sensor based DSLRs.  Can a speed booster turn them into full-frame cameras when operating in Life View with full-frame lenses?  Do such speed boosters exist and if yes, does anyone on this forum have any experience with them? 

[alpicat]

Well, my question was actually about APSC-sensor based DSLRs.  Can a speed booster turn them into full-frame cameras when operating in Life View with full-frame lenses?  Do such speed boosters exist and if yes, does anyone on this forum have any experience with them? 

Sorry I misread your question. Unfortunately that's optically impossible. The flange distance (between sensor and EF mount) on EOS DSLRs is 44mm so there's no space for a speedbooster. You could hack one and put it inside the camera housing, but you'd have to take your camera apart and remove the mirror mechanism. On the other hand, EF-M mount flange distance is only 18mm (as it's mirrorless), so you can add a speedbooster in front of that mount.

[Walter Schulz]

I tend to disagree. With same logic applied you may prove teleconverters are impossible to use on Canon cams. IMO you mix up two concepts: Extending flange distance without optical elements (extenders, bellows) and inserting an optical element (teleconverter, speedbooster, mount adapters with lenses) into optical path.

It's another question if there is a company seeing a market for such an adapter. My 2 cents: Nope.

[alpicat]

@Walter Schulz, that's a very good point!

[Licaon_Kter]

Keeping an eye on this too

[ArcziPL]

Hello, does anyone have a Samsung Evo Plus microSD card and did or could do the benchmarking? It seems to be a promising alternative. Generally it is also able to achieve approx. 90MB/s bulk write speeds but at much lower cost than SanDisk Extreme Pro. As an example: 128GB Samsung Evo Plus is at Amazon equally priced to 64GB SanDisk Extreme Pro (39€ vs. 38,99€). Both are sold by Amazon, so the risk of counterfeit should be equally low.

Funnily, according to some sources, the transfer drops by roughly a half when using the provided MicroSD -> SD adapter. If another adapter can overcome this bottleneck and if similarly high speeds can be achieved with an EOS camera it for me unknown so far. It will decide if it's really an alternative. If yes, it would be a bargain.

[Sapporo]

Hello, does anyone have a Samsung Evo Plus microSD card and did or could do the benchmarking? It seems to be a promising alternative. Generally it is also able to achieve approx. 90MB/s bulk write speeds but at much lower cost than SanDisk Extreme Pro. As an example: 128GB Samsung Evo Plus is at Amazon equally priced to 64GB SanDisk Extreme Pro (39€ vs. 38,99€). Both are sold by Amazon, so the risk of counterfeit should be equally low.

Funnily, according to some sources, the transfer drops by roughly a half when using the provided MicroSD -> SD adapter. If another adapter can overcome this bottleneck and if similarly high speeds can be achieved with an EOS camera it for me unknown so far. It will decide if it's really an alternative. If yes, it would be a bargain.

Samsung PRO 32GB here. 41MB/s default with 6D. 65MB/s overclocked.

[kye]

Hello, does anyone have a Samsung Evo Plus microSD card and did or could do the benchmarking? It seems to be a promising alternative. Generally it is also able to achieve approx. 90MB/s bulk write speeds but at much lower cost than SanDisk Extreme Pro. As an example: 128GB Samsung Evo Plus is at Amazon equally priced to 64GB SanDisk Extreme Pro (39€ vs. 38,99€). Both are sold by Amazon, so the risk of counterfeit should be equally low.

Funnily, according to some sources, the transfer drops by roughly a half when using the provided MicroSD -> SD adapter. If another adapter can overcome this bottleneck and if similarly high speeds can be achieved with an EOS camera it for me unknown so far. It will decide if it's really an alternative. If yes, it would be a bargain.

I have Samsung Evo Plus 256Gb microSD card (in the adapter) and 700D running the crop_rec from 10th March.  I am willing to do some testing but haven't got a clue how to install the UHS module, so would need a lot of hand-holding!
I also have a few adapters that I could try if someone is able to help me set it up.

[Walter Schulz]

@kye:
First question is if there is a penalty involved using SD-card adapters. If you own a decent cardreader you can run Crystaldiskmark (Windows) or Blackmagic Disk Speed Test (MacOS, OS X) to find out.
Don't forget sd_uhs.mo is able to kill memory cards.

[kye]

@kye:
First question is if there is a penalty involved using SD-card adapters. If you own a decent cardreader you can run Crystaldiskmark (Windows) or Blackmagic Disk Speed Test (MacOS, OS X) to find out.
Don't forget sd_uhs.mo is able to kill memory cards.

Ok, figured it out!

BM Disk Speed test in Transcend SD card reader:
Samsung EVO Plus with no adapter (directly in microSD slot in card reader) ~83MB/s
Samsung EVO Plus with Red Sandisk adapter ~83MB/s
Samsung EVO Plus with Black Sandisk adapter (from Sandisk Ultra card) ~83MB/s
Samsung EVO Plus with Black Kingston adapter ~83MB/s

Looks like no differences between those adapters anyway.

Log from sd_uhs.mo in 700D:

===================
Before the hack: r:27MB/s w:31MB/s  W:27MB/s R:38MB/s    [best 31MB/s]
SDR50 @ 96MHz  : r:38MB/s w:33MB/s  W:31MB/s R:38MB/s    [best 33MB/s]
SDR50 @ 96MHz  : r:38MB/s w:33MB/s  W:31MB/s R:37MB/s  [best 33MB/s]
SDR50 @ 80MHz  : r:33MB/s w:29MB/s  W:29MB/s R:32MB/s  meh [best 33MB/s]
SDR50 @ 80MHz  : r:33MB/s w:29MB/s  W:29MB/s R:32MB/s  meh [best 33MB/s]
SDR50 @ 120MHz : r:42MB/s w:32MB/s  W:32MB/s R:45MB/s  [best 33MB/s]
SDR50 @ 120MHz : r:42MB/s w:32MB/s  W:32MB/s R:45MB/s  [best 33MB/s]
SDR104 @ 96MHz : D0 D0 r:37MB/s w:33MB/s  W:31MB/s R:37MB/s    [best 33MB/s]
SDR104 @ 96MHz : D1 D1 r:38MB/s w:33MB/s  W:31MB/s R:37MB/s  [best 33MB/s]
SDR104 @ 80MHz : D0 D0 r:32MB/s w:29MB/s  W:29MB/s R:32MB/s  meh [best 33MB/s]
SDR104 @ 80MHz : D1 D1 r:32MB/s w:30MB/s  W:29MB/s R:26MB/s  meh [best 33MB/s]
SDR104 @ 120MHz: D0 D0 r:42MB/s w:32MB/s  W:33MB/s R:45MB/s  [best 33MB/s]
SDR104 @ 120MHz: D1 D1 r:41MB/s w:32MB/s  W:32MB/s R:45MB/s  [best 33MB/s]
SDR104 @ 132MHz: D0 D0 r:40MB/s w:32MB/s  W:32MB/s R:43MB/s  [best 33MB/s]
SDR104 @ 132MHz: D1 D1 r:40MB/s w:32MB/s  W:32MB/s R:42MB/s  [best 33MB/s]
SDR104 @ 160MHz: D0 D0 r:56MB/s w:36MB/s  W:47MB/s R:42MB/s    [best 36MB/s]
SDR104 @ 160MHz: D1 D1 r:42MB/s w:47MB/s  W:46MB/s R:42MB/s    [best 47MB/s]
SDR104 @ 160MHz: D2 D2 r:55MB/s w:37MB/s  W:47MB/s R:41MB/s  meh [best 47MB/s]
SDR104 @ 160MHz: D3 D3 r:56MB/s w:46MB/s  W:47MB/s R:41MB/s  [best 47MB/s]
SDR104 @ 160MHz: D0 D0 r:43MB/s w:46MB/s  W:47MB/s R:42MB/s  [best 47MB/s]
SDR104 @ 160MHz: D1 D1 r:46MB/s w:46MB/s  W:47MB/s R:42MB/s  [best 47MB/s]
SDR104 @ 160MHz: D2 D2 r:44MB/s w:46MB/s  W:47MB/s R:42MB/s  [best 47MB/s]
SDR104 @ 160MHz: D3 D3 r:45MB/s w:47MB/s  W:46MB/s R:40MB/s  [best 47MB/s]
SDR104 @ 160MHz: D0 D0 r:41MB/s w:48MB/s  W:47MB/s R:42MB/s    [best 48MB/s]
SDR104 @ 160MHz: D1 D1 r:42MB/s w:45MB/s  W:47MB/s R:42MB/s  [best 48MB/s]
SDR104 @ 160MHz: D2 D2 r:56MB/s w:47MB/s  W:47MB/s R:42MB/s  [best 48MB/s]
SDR104 @ 160MHz: D3 D3 r:42MB/s w:46MB/s  W:47MB/s R:43MB/s  [best 48MB/s]
SDR104 @ 160MHz: D0 D0 r:42MB/s w:47MB/s  W:47MB/s R:41MB/s  [best 48MB/s]
SDR104 @ 160MHz: D1 D1 r:42MB/s w:48MB/s  W:45MB/s R:41MB/s  [best 48MB/s]
SDR104 @ 160MHz: D2 D2 r:43MB/s w:47MB/s  W:46MB/s R:41MB/s  [best 48MB/s]
SDR104 @ 160MHz: D3 D3 r:43MB/s w:46MB/s  W:46MB/s R:42MB/s  [best 48MB/s]
Best: D0 D0 r:42MB/s w:46MB/s  W:47MB/s R:40MB/s  [best 48MB/s]
Best: D0 D0 r:42MB/s w:45MB/s  W:47MB/s R:40MB/s  [best 48MB/s]

Done.
Please run THOROUGH tests before using!!!

Edit: forgot to mention my Samsung card didn't come with an adapter.
Here's the equipment from this test:




Edit 2: ran it again and it seemed to change its mind on which settings were the best?  Not sure if this is just the test results varying normally?  (the BM tests would jump around between about 80-83MB/s so I assume it's just what happens with throughput testing.

Before the hack: r:38MB/s w:35MB/s  W:32MB/s R:38MB/s    [best 35MB/s]
SDR50 @ 96MHz  : r:38MB/s w:34MB/s  W:34MB/s R:38MB/s  [best 35MB/s]
SDR50 @ 96MHz  : r:38MB/s w:34MB/s  W:33MB/s R:38MB/s  [best 35MB/s]
SDR50 @ 80MHz  : r:33MB/s w:29MB/s  W:29MB/s R:33MB/s  meh [best 35MB/s]
SDR50 @ 80MHz  : r:32MB/s w:30MB/s  W:30MB/s R:32MB/s  meh [best 35MB/s]
SDR50 @ 120MHz : r:45MB/s w:40MB/s  W:39MB/s R:46MB/s    [best 40MB/s]
SDR50 @ 120MHz : r:45MB/s w:38MB/s  W:39MB/s R:43MB/s  [best 40MB/s]
SDR104 @ 96MHz : D0 D0 r:38MB/s w:33MB/s  W:34MB/s R:35MB/s  meh [best 40MB/s]
SDR104 @ 96MHz : D1 D1 r:37MB/s w:34MB/s  W:34MB/s R:37MB/s  meh [best 40MB/s]
SDR104 @ 80MHz : D0 D0 r:33MB/s w:30MB/s  W:29MB/s R:32MB/s  meh [best 40MB/s]
SDR104 @ 80MHz : D1 D1 r:32MB/s w:30MB/s  W:30MB/s R:32MB/s  meh [best 40MB/s]
SDR104 @ 120MHz: D0 D0 r:46MB/s w:40MB/s  W:38MB/s R:42MB/s  [best 40MB/s]
SDR104 @ 120MHz: D1 D1 r:46MB/s w:36MB/s  W:40MB/s R:41MB/s  meh [best 40MB/s]
SDR104 @ 132MHz: D0 D0 r:43MB/s w:38MB/s  W:36MB/s R:42MB/s  [best 40MB/s]
SDR104 @ 132MHz: D1 D1 r:43MB/s w:33MB/s  W:33MB/s R:43MB/s  meh [best 40MB/s]
SDR104 @ 160MHz: D0 D0 r:58MB/s w:45MB/s  W:45MB/s R:55MB/s    [best 45MB/s]
SDR104 @ 160MHz: D1 D1 r:55MB/s w:47MB/s  W:47MB/s R:57MB/s    [best 47MB/s]
SDR104 @ 160MHz: D2 D2 r:57MB/s w:47MB/s  W:47MB/s R:57MB/s    [best 47MB/s]
SDR104 @ 160MHz: D3 D3 r:53MB/s w:47MB/s  W:47MB/s R:57MB/s  [best 47MB/s]
SDR104 @ 160MHz: D0 D0 r:58MB/s w:47MB/s  W:47MB/s R:57MB/s    [best 47MB/s]
SDR104 @ 160MHz: D1 D1 r:56MB/s w:47MB/s  W:47MB/s R:58MB/s    [best 47MB/s]
SDR104 @ 160MHz: D2 D2 r:57MB/s w:46MB/s  W:46MB/s R:54MB/s  [best 47MB/s]
SDR104 @ 160MHz: D3 D3 r:57MB/s w:45MB/s  W:46MB/s R:54MB/s  [best 47MB/s]
SDR104 @ 160MHz: D0 D0 r:55MB/s w:47MB/s  W:47MB/s R:57MB/s  [best 47MB/s]
SDR104 @ 160MHz: D1 D1 r:56MB/s w:44MB/s  W:43MB/s R:55MB/s  [best 47MB/s]
SDR104 @ 160MHz: D2 D2 r:57MB/s w:42MB/s  W:46MB/s R:54MB/s  meh [best 47MB/s]
SDR104 @ 160MHz: D3 D3 r:58MB/s w:47MB/s  W:47MB/s R:56MB/s  [best 47MB/s]
SDR104 @ 160MHz: D0 D0 r:56MB/s w:44MB/s  W:43MB/s R:57MB/s  [best 47MB/s]
SDR104 @ 160MHz: D1 D1 r:54MB/s w:48MB/s  W:46MB/s R:45MB/s    [best 48MB/s]
SDR104 @ 160MHz: D2 D2 r:55MB/s w:45MB/s  W:38MB/s R:55MB/s  [best 48MB/s]
SDR104 @ 160MHz: D3 D3 r:56MB/s w:47MB/s  W:46MB/s R:55MB/s  [best 48MB/s]
Best: D1 D1 r:55MB/s w:47MB/s  W:46MB/s R:58MB/s  [best 48MB/s]
Best: D1 D1 r:57MB/s w:43MB/s  W:47MB/s R:46MB/s  meh [best 48MB/s]

[Danne]

I have put up a hardcoded 160MHZ 700D build in my repository to be tested. It´s the same as for 100D and eosm. If anybody tests this, please feedback here what´s working or not. Fried cards or not etc...
On top of downloads page:
https://bitbucket.org/Dannephoto/magic-lantern/downloads/

[kye]

I have put up a hardcoded 160MHZ 700D build in my repository to be tested. It´s the same as for 100D and eosm. If anybody tests this, please feedback here what´s working or not. Fried cards or not etc...
On top of downloads page:
https://bitbucket.org/Dannephoto/magic-lantern/downloads/

I think I might have stuffed up..  do I need to be in photo mode when I run the module?  I've just run it in there and gotten much better results than I posted before!   

Also, should I replace your previous module with the hard-coded one?  What's the difference?

Also also, with the non-hard-coded one do I have to run it every time I turn on the camera?  or does it remember the best settings from last time?  I tried a couple of tests and couldn't get a straight answer.

Sorry for all the questions...  hopefully I'm helping more than getting in the way 

[Danne]

Good questions. This topic has started and the module is tested atm. If it's here to stay, who knows.
This version has some early code based on some code erased in a1ex earlier code. As you say, run the module in photo mode or with raw_rec turned off. Switch back to movie mode and you should be able to record at higher speeds. You have to run this set up every time you restart your camera unfortunately. Maybe if A1ex thinks it safe we could have the patch hard coded upon start up...

[Walter Schulz]

To verify overclocking is working in movie mode: Load Bench.mo, too, run sd_uhs in photo mode, switch to movie mode and run card benchmark (debug tab). Numbers will be lower in movie mode, though.

[ArcziPL]

Samsung PRO 32GB here. 41MB/s default with 6D. 65MB/s overclocked.

Ok, figured it out!

BM Disk Speed test in Transcend SD card reader:
Samsung EVO Plus with no adapter (directly in microSD slot in card reader) ~83MB/s
Samsung EVO Plus with Red Sandisk adapter ~83MB/s
Samsung EVO Plus with Black Sandisk adapter (from Sandisk Ultra card) ~83MB/s
Samsung EVO Plus with Black Kingston adapter ~83MB/s

Looks like no differences between those adapters anyway.

Log from sd_uhs.mo in 700D:
Before the hack: r:27MB/s w:31MB/s  W:27MB/s R:38MB/s    [best 31MB/s]
(...)
SDR104 @ 160MHz: D1 D1 r:42MB/s w:47MB/s  W:46MB/s R:42MB/s    [best 47MB/s]

Sapporo, kye, thank you very much for sharing your results! So there is no problem with adapters, Samsung Evo Plus is a decent card but only twice more expensive Samsung Pro is on par with SanDisk Extreme Pro when using with 700D. Good to know!

[Tullen]

Well, my question was actually about APSC-sensor based DSLRs.  Can a speed booster turn them into full-frame cameras when operating in Life View with full-frame lenses?  Do such speed boosters exist and if yes, does anyone on this forum have any experience with them?

Actually I already made a thread about it. I have a speedboaster, an extra 50D and some ideas about how it can be done, just haven't got to it yet. https://www.magiclantern.fm/forum/index.php?topic=13957.msg134615#msg134615