SD card speed test - playtime ;)

[Peter E]

Advertisement (gone after registration)

Hi,

 

Today I've played a bit with my camera

 

I've tested a Sandisk 16Gb 280Mb card(formatted in slot1) and performed a speed test regarding frames/sec. I've hoped it was faster with DNG only than my 600x Lexar pro 16Gb but it wasn't.

I also hoped to get more then 35 DNG files in high speed but nope.. ;(

 

Those are the results:

- DNG only, 35 frames and it stops shooting. It needed just almost 4 seconds for it, so we get an average of 8,75frames/sec.

- JPEG large only, it keeps going until 81 frames after it's slows down and I get an average of 11,25 frames/ seconds.

 

It seems that the speed and total amount of images in high speed are not depending on the SD card, it all has to to with the internal buffer of the camera...

[Jared]

Did you try the same test in Slot 2? I think there has been some confusion over which of the slots is UHS II and which is UHS I. I still wouldn't expect a huge difference, but perhaps one or two more frames before the buffer fills up with the faster card? Assuming the Maestro II processor isn't so busy with creating DNG's that it can keep up with the read/write speeds of the card...

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[Jared]

FYI, I got very similar results to you using a 64GB Lexar Professional 2000x 300MB/s UHS-II card.  Not necessarily faster or slower, but slightly different behavior, and slightly odd as well.

 

When shooting DNG alone, my buffer filled up in just 2.75 seconds and captured just 25 frames before it slowed way down to one frame every four seconds or so.  It took about 44s to clear the buffer (write out all 25 frames).  That's 9fps shooting DNG's which is a touch faster than yours, but possibly within my (or your) measurement error.  It seemed strange that it filled up so fast and captured only 25 frames, so I ran the test a couple more times and got between 30 and 32 frames.  As my battery began to run down (three bars rather than the full four), the frame rate seemed to drop a little, from 9fps to 8fps.  Nothing drastic, but it was definitely lower.  Interesting.  Not certain it's battery related, of course.  I'll charge it up again and see.  

 

When shooting DNG + JPG my buffer filled up in 3.5 seconds but captured 35 frames instead of 25.  It took about 60s to clear the buffer.  Again, that's about 9fps shooting DNG + JPG.  

 

When shooting JPG alone, the buffer filled after 10.7 seconds and I got a total of 108 frames, so that's almost exactly 10 frames per second, a bit below the advertised 11 fps.  

 

When I had two cards inserted rather than one and had the camera set to mirror files on both cards, the shooting rate for DNG's, DNG's + JPG's, and JPG's alone was much slower--more like five or six frames per second, so don't try to mirror files if you need a fast shutter rate.

 

All of this is purely academic for me.  Anything over 5fps is overkill for me, and I'd rather choose a rate that allows continuous AF update in any event.  Other users needs will vary, of course.  

 

- Jared

[Jared]

Ignore my last post.  It wasn't a battery issue at all.  The variability I was getting with DNG's and with JPG's in terms of frame rate was not remembering to set a manual shutter speed.  Duh.  You're not going to get the same FPS at 1/45s as you do at 1/1000s.  I'll try again tomorrow and report back.

 

- Jared

[Peter E]

Hi Jared,

 

I only used slot 1 as this is the one for UHS II but I'll test slot 2 too.

I've tested with a M-lens only.

 

It's strange that you've got only 25 DNG frames before slowing down, I'll get everytime 35 frames.

Also the difference between the JPG speed, I have 81 frames vs 108 frames that you measured before slowing down...

Are you shooting JPG-L too?

 

For my use it's interesting to have this high speed, and it would be better if I can capture more dng files than 35 in this mode but I'm happy with the results.

 

 

I did the tests at 1/2000

 

 

Regards

Peter

Edited January 14, 2016 by Peter E

[Jared]

I wouldn't trust my results till I can try again tomorrow.  Not certain why I am getting (at least sometimes) fewer than 35 frames before the buffer fills.  I'll see if I can narrow that down.  I noticed no difference between Slot 1 and Slot 2 tonight, but, again, since I hadn't manually set the shutter speed I don't trust my results.  I'll keep you posted.

 

- Jared

[CheshireCat]

Advertisement (gone after registration)

You're not going to get the same FPS at 1/45s as you do at 1/1000s.

 

 

Why not ? Even at 1/45s the shutter should not be the bottleneck for the SL.

 

I don't have the SL, but I understand that you can't get the max fps if you are shooting with focus priority.

Edited January 14, 2016 by CheshireCat

[Jared]

No, I had autofocus turned off, so that's not an issue. And slower shutter speeds definitely slow down your total frames per second since the camera is spending some time with the shutter open, not able to re-set the first curtain.

 

I'm getting 10.4 frames per second with JPG only, about 9.3 frames per second with DNG only, but still a range of anywhere from 25 to 35 DNG shots before the buffer is full--3 to 4 seconds. Don't know why the number of shots to fill the buffer isn't more consistent.

 

- Jared

[Peter E]

Hi Jared,

 

Did you format you're card inside the camera?

 

Peter

[忘不了]

SL boot and storage speed does not compare to the Canon and Nikon Leica, must want to occupy the market with the Japanese Learning

[Jared]

Hi Jared,

 

Did you format you're card inside the camera?

 

Peter

 

 

No, mine were formatted in SD Formatter on a Mac.  I could try formatting in the camera instead, but honestly I'm happy with the performance of the SD cards in terms startup speed, and I don't really care whether my image buffer can handle 35 frames or 30.  Just not an issue for my shooting.  Either is plenty.  Now, I WOULD be a little more interested in finding a way to reduce the amount of time it takes to write out the contents of the buffer to the card, but it seems that I'm already at the point that the Maestro II processor is the bottleneck, not the cards.  This is based simply upon the fact that even my 90MB/s and 95MB/s cards seem to be about as fast from a practical standpoint as the new Lexar cards.  Maybe this will improve in future firmware releases for the SL as Leica finds time to optimize the code.  Maybe it won't.  If I really cared, I'd just shoot JPG's for action work that required hundreds of frames.  The benefits of the smaller files in terms of workflow when dealing with hundreds of images are substantial enough that I probably wouldn't be using DNG for that kind of work anyway.

 

- Jared

[Jared]

SL boot and storage speed does not compare to the Canon and Nikon Leica, must want to occupy the market with the Japanese Learning

 

For some reason that I'm not aware of, all the mirrorless cameras seem to have substantially slower startup times than DSLR's.  Not sure why.  It's not a Leica specific issue, though.  The Sony A7r II is around 1.5 seconds.  The Canon EOS M3 is about 2.5 seconds.  The Olympus E-M10 is just under 1 second.  The time of 1.5 s for turn-on that you can get out of the SL is among the leaders for mirrorless.  The bigger issue for me would be the fact that Leica has historically had inconsistent results from one brand and model of SD card to another.  Not certain why that is--I'm not aware that other manufacturers struggle so much with compatibility. 

 

As far as write speed... It's hard to know how much of the time required is for processing the DNG's as opposed to actual write time, but I seem to be getting about 30 MB/s out of the camera as it clears its buffer.  The D800, as a standard of comparison, clears its buffer a little bit faster--about 35MB/s with fast cards.  The Nikon D4, a fast performance oriented camera, is about 42MB/s.  None of these is anywhere near the 90 MB/s write speed that a good UHS-I card is capable of, let alone the 250 MB/s that a UHS-II card can handle.  I suspect the time to clear the buffer is much more about the image processing itself--converting the truly "raw" output from the imaging chip to DNG's--not about the time to store the data on the card.  Again, it seems that Leica is getting reasonable results here to other similar cameras.  Nothing that will blow you away, but 30 MB/s is in line with Nikon and Canon.

 

Do you have any idea why mirrorless startup times seem to be so much slower than DSLR's?  There may be a good reason for it, but I've never heard it.

 

- Jared

[CheshireCat]

Do you have any idea why mirrorless startup times seem to be so much slower than DSLR's?  There may be a good reason for it, but I've never heard it.

 

 

There is no good reason. It is just that the firmware is crap.

[Jared]

There is no good reason. It is just that the firmware is crap.

Really? Across all brands? Because this isn't just a Leica feature. Even Canon mirrorless cameras have much slower startup times than their DSLR cousins. It seems like there must be more to it than just sloppy coding.

 

- Jared

[pico]

I suspect the time to clear the buffer is much more about the image processing itself--converting the truly "raw" output from the imaging chip to DNG'

 

Isn't the DNG truly raw binary anyway? When we read the data for post-processing it is the computer's software that interprets it and as such it does not require great speed.

 

None of these is anywhere near the 90 MB/s write speed that a good UHS-I card is capable of, let alone the 250 MB/s that a UHS-II card can handle.

 

Those speeds are advertised by the vendor, and are max READ speeds, not Write.

.

Edited January 17, 2016 by pico

[pico]

There is no good reason. It is just that the firmware is crap.

 

Oh, stop sugar-coating it, Cat!

[scott kirkpatrick]

Isn't the DNG truly raw binary anyway? When we read the data for post-processing it is the computer's software that interprets it and as such it does not require great speed.

 

The "raw" data in the DNG file is processed at least twice.  First each pixel's output is looked up in a table for the particular lens to obtain the value corrected for vignetting and color shifts.  There may be other corrections applied at the Bayer pixel level, but those are not disclosed. Then there's a pass for lossless encoding to shrink the files a bit.  (It doesn't seem adaptive -- every file is the same size.)  Then to wrap it up as a DNG there's a small ocean of metadata and an embedded thumbnail that goes inside before the actual picture data.  Encoding any jpeg and the thumbnail might also be a rate-limiting step.  

[CheshireCat]

The "raw" data in the DNG file is processed at least twice.  First each pixel's output is looked up in a table for the particular lens to obtain the value corrected for vignetting and color shifts.  There may be other corrections applied at the Bayer pixel level, but those are not disclosed. Then there's a pass for lossless encoding to shrink the files a bit.  (It doesn't seem adaptive -- every file is the same size.)  Then to wrap it up as a DNG there's a small ocean of metadata and an embedded thumbnail that goes inside before the actual picture data.  Encoding any jpeg and the thumbnail might also be a rate-limiting step.  

 

 

This is not accurate.

 

To estimate the "extra" DNG overhead, we must only take into account operations that are not required by JPEG-only.

Therefore we can safely rule out vignetting correction and other processing also required by JPEG.

We can also rule out metadata as they are also embedded in the JPEG and string processing takes basically zero processor time.

 

As far as computational requirements are concerned, JPEG requires a more complex compression than DNG lossless compression (and file size does change depending on entropy).

But we can safely rule out DNG compression, because as far as I know, the SL cannot compress DNG (shame).

 

Yes, the DNG also embeds a JPEG thumbnail, but it is low-resolution, hence much lighter on the processor than a full JPEG.

 

Now, what is the bottleneck here ?

The buffer is filled faster than it can be flushed to SD, so it can be the processing or the SD write operation.

I would rule out the processing, as DNG files do require less processing than JPEG, yet shooting DNG is slower.

So it must be the write operation.

If you think about it, what really makes the difference between DNG and JPEG is the file size. A lot more data to be written when saving a DNG.

And if faster speed SD cards have the same performance of lower speed ones, then this means the SD card controller in the SL cannot take advantage of fast cards.

[jaapv]

There is no good reason. It is just that the firmware is crap.

Of all cameras that the poster mentioned? Note that he is talking about mirrorless in general, not your favorite flavour of firmware (Leica)

[Advertisement]

Hallo,

Look here: Leica L-Mount

[CheshireCat]

Really? Across all brands? Because this isn't just a Leica feature. Even Canon mirrorless cameras have much slower startup times than their DSLR cousins. It seems like there must be more to it than just sloppy coding.

 

Really, I can't see any good reason. I guess that since mirrorless cameras are not targeting pros, no one told the developers "The damn thing needs to be ready to shoot in less than 300 milliseconds".

 

My "crap firmware" comment apart , achieving this is far from trivial because also the embedded OS needs to be carefully optimized for this, and the camera "application" must be able to handle different tasks (including error handling) in parallel. Sequential processing makes things much easier, but slower, and we are talking about expensive pro cameras here.

Edited January 17, 2016 by CheshireCat