OM-D E-M5 Photographic Dynamic Range and Read Noise

[bclaff]

Thanks to excellent data collection by 'Anders W' I have complete Photographic Dynamic Range (PDR) and Read Noise data for the OM-D E-M5.
See:
http://home.comcast.net/~NikonD70/Charts/PDR.htm#OM-D%20E-M5
and
http://home.comcast.net/~NikonD70/Charts/RN_ADU.htm#OM-D%20E-M5_12

--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

 

[Detail Man]

Good to see these Olympus OM-D E-M5 test results published. Many thanks to Bill and to Anders !

 

[Timur Born]

Thank you. How to interpret the little jaggy hill between ISO 320 and ISO 400?

 

[Anders W]

Well, that was really quick Bill! Little more than an hour between data delivery and published report and two to three hours for our combined efforts. Now that sets an example for DxO to follow.

Glad to hear that you found the data to be in order. The general shape of the graphs look very much in line with my expectations although I have yet to digest the precise implications of the DR numbers in more absolute terms.

Thanks for taking the trouble to do this for the E-M5, the first MFT camera on your site as far as I can see. Hope you will continue to add other new MFT bodies as they appear.

Anders

 

[Anders W]

Thank you. How to interpret the little jaggy hill between ISO 320 and ISO 400?

Uneven read noise development between ISO 200 and ISO 400. At the intermediate ISOs (250, 320), read noise doesn't decline much (as measured in electrons), which means that DR falls. But at 400, read noise drops to nearly half of the orginal ISO 200 value, which suffices to bring DR up a bit again to a value not far from the one at ISO 200. See Bill's read-noise graph as well as this post for details:

http://forums.dpreview.com/forums/read.asp?forum=1041&message=41988325

Perhaps Bill can shed further light on the possible technical reasons for this somewhat strange read-noise curve. Two-stage amplifier or what? But why then this pattern only at the very beginning of the ISO range?

 

[Jerodequin]

How should we interpret the read noise charts? (the DR charts seem quite self explanatory) but wouldn't a lower score be better for read noise? (this can't be the case though as the EM5 plot is lower than even the D800) - could someone please explain how to read and compare when considering the read noise charts?

Thanks!

 

[Anders W]

How should we interpret the read noise charts? (the DR charts seem quite self explanatory) but wouldn't a lower score be better for read noise? (this can't be the case though as the EM5 plot is lower than even the D800) - could someone please explain how to read and compare when considering the read noise charts?

Thanks!

As Bill points out at the bottom of his read-noise page, these figures cannot easily be compared across cameras since they are not adjusted for gain or area. So what you can primarily get from this chart, without further information and calculation, is a view of how read noise develops across ISOs. When read noise is practically constant as measured in electrons, there will be a linear upwards slope as is the case here from ISO 400 on. When read noise as measured in electrons falls rapidly, as it does when you go from 200 to 400, there will be a much weaker upwards slope. For the strange behavior of the read noise at ISO 250 and 320 relative to 400, see my reply to Timur above.

 

[brunobarolo]

Please allow a layman's stupid question: The photographic dynamic range E-M5 graph (at least for ISO 800 and above) is very close to the ideal 4/3 graph. Would that mean that the laws of physics don't allow much improvement any more for dynamic range and thus shadow noise at high ISOs in future 4/3 sensors?

Or what does "ideal 4/3" mean?

 

[Timur Born]

Yes, I remembered that thread and also read the ADU explanation for "jaggy" curves at the bottom of Bill's page. I just wondered how you would interpret that single "disturbance in the force".

 

[Anders W]

For those interested, I noticed that there is also a diagram for E-M5 read noise as expressed in electrons rather than ADUs.

http://home.comcast.net/~NikonD70/Charts/RN_e.htm

I am not yet sure whether this graph is a preliminary or a final version, but it certainly looks very reasonable to my eyes.

 

[bclaff]

Please allow a layman's stupid question: The photographic dynamic range E-M5 graph (at least for ISO 800 and above) is very close to the ideal 4/3 graph. Would that mean that the laws of physics don't allow much improvement any more for dynamic range and thus shadow noise at high ISOs in future 4/3 sensors?

Or what does "ideal 4/3" mean?

Not a stupid question at all.

I don't measure or report based on "real" ISO but on the manufacturer stated ISO.

In this case, if you apply the DxOLabs numbers, the curve is shifted about 1/3EV to the right of where it "should" be. That's why it's so close to the ideal line.

When comparing cameras using my PDR curves, especially across brands, you should probably not take differences of less than 1/2 EV too seriously.
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

 

[bclaff]

Timur Born wrote:

Perhaps Bill can shed further light on the possible technical reasons for this somewhat strange read-noise curve. Two-stage amplifier or what? But why then this pattern only at the very beginning of the ISO range?

Some "high end" Canon models use a two-stage amplifier but most manufacturers push/pull the raw data in their firmware. (FWIW, Nikon always uses a single amplifier.)

From http://home.comcast.net/~NikonD70/Charts/RN_e.htm#OM-D%20E-M5_12 it would appear that E-M5 intermediates are pushed; so ISO 250 and ISO 320 are pushed from ISO 200.

Contrast this with a Canon camera http://home.comcast.net/~NikonD70/Charts/RN_e.htm#EOS%207D_14,OM-D%20E-M5_12 where it seems clear that intermediate ISO are either pushed or pulled from the nearest "whole" ISO.

The effect is more pronounced at the low ISOs but is present at all intermediate ISOs. The reason stems from how much of the read noise is at the photosite as opposed to how much is being added by the Analog to Digital Converter (ADC). As the value in electrons levels out the effect is harder to see.

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

 

[brunobarolo]

Please allow a layman's stupid question: The photographic dynamic range E-M5 graph (at least for ISO 800 and above) is very close to the ideal 4/3 graph. Would that mean that the laws of physics don't allow much improvement any more for dynamic range and thus shadow noise at high ISOs in future 4/3 sensors?

Or what does "ideal 4/3" mean?

In this case, if you apply the DxOLabs numbers, the curve is shifted about 1/3EV to the right of where it "should" be. That's why it's so close to the ideal line.

--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

And that "ideal line" is fixed by the laws of physics? Which would mean, at least regarding high ISO DR, future 4/3 sensors will never perform substantially better than the E-M5 sensor performs today?

 

[compositor20]

The graph for maximum theorical DR for 4/3 sensor size worried me..

That means we at m43 ar at our maximum at high iso...

Does it means we won´t have better performance than om-d at High ISO ?

Photon noise could be better at least... but I think that om-d is already at 57 to 58 % at QE. only just 0.5 EV to get better results... although that would make iso 12800 usable and iso 6400 good for day to day as iso 3200 already is on OM-D.

 

[bclaff]

And that "ideal line" is fixed by the laws of physics? Which would mean, at least regarding high ISO DR, future 4/3 sensors will never perform substantially better than the E-M5 sensor performs today?

The graph for maximum theorical DR for 4/3 sensor size worried me..

That means we at m43 ar at our maximum at high iso...

Does it means we won´t have better performance than om-d at High ISO ?

Photon noise could be better at least... but I think that om-d is already at 57 to 58 % at QE. only just 0.5 EV to get better results... although that would make iso 12800 usable and iso 6400 good for day to day as iso 3200 already is on OM-D.

As has been mentioned already, The E-M5 line cannot be compared directly to the ideal line because "true" ISO is about 1/3EV lower than the stated ISO. So think of the E-M5 line as about 1/3EV to the left of where it shows.

That said, the ideal lines are computed assuming a perfect world with only photon noise. Even so, at each sensor size I have tested there is at least one camera that approaches the ideal line at high ISO. There's room for improvement but not dramatic improvement.

As photographers the improvement we are most likely to see is that better sensors get into less expensive cameras! Another related improvement are Auto-Focus (AF) systems that perform in lower light (like Nikon's move from f/5.6 to f/8 on some bodies).

In general we are at the point that is you need much more DR than your current (modern top of the line) camera then you may have to go to a larger format to meet that need.
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

 

[brunobarolo]

Thanks, Bill, for clearing up that question.

 

[Anders W]

And that "ideal line" is fixed by the laws of physics? Which would mean, at least regarding high ISO DR, future 4/3 sensors will never perform substantially better than the E-M5 sensor performs today?

The graph for maximum theorical DR for 4/3 sensor size worried me..

That means we at m43 ar at our maximum at high iso...

Does it means we won´t have better performance than om-d at High ISO ?

Photon noise could be better at least... but I think that om-d is already at 57 to 58 % at QE. only just 0.5 EV to get better results... although that would make iso 12800 usable and iso 6400 good for day to day as iso 3200 already is on OM-D.

As has been mentioned already, The E-M5 line cannot be compared directly to the ideal line because "true" ISO is about 1/3EV lower than the stated ISO. So think of the E-M5 line as about 1/3EV to the left of where it shows.

That said, the ideal lines are computed assuming a perfect world with only photon noise. Even so, at each sensor size I have tested there is at least one camera that approaches the ideal line at high ISO. There's room for improvement but not dramatic improvement.

As to the assumptions that go into that perfect world, I take it you assume that Bayer is perfect. If we assume a world where all the light rather 1/3 of the light (approximatively) can be used by each pixel, we'd end up with another ideal, wouldn't we?

As photographers the improvement we are most likely to see is that better sensors get into less expensive cameras! Another related improvement are Auto-Focus (AF) systems that perform in lower light (like Nikon's move from f/5.6 to f/8 on some bodies).

In general we are at the point that is you need much more DR than your current (modern top of the line) camera then you may have to go to a larger format to meet that need.
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

 

[ginsbu]

That said, the ideal lines are computed assuming a perfect world with only photon noise. Even so, at each sensor size I have tested there is at least one camera that approaches the ideal line at high ISO. There's room for improvement but not dramatic improvement.

As photographers the improvement we are most likely to see is that better sensors get into less expensive cameras! Another related improvement are Auto-Focus (AF) systems that perform in lower light (like Nikon's move from f/5.6 to f/8 on some bodies).

In general we are at the point that is you need much more DR than your current (modern top of the line) camera then you may have to go to a larger format to meet that need.

I hope we'll also see manufacturers respond by including (genuine, not extended) low ISOs to offer increased DR. I don't know what the tradeoffs might be in doing that, though.

Thanks for your presentation and explanations.

 

[Timur Born]

Leaves the (unanswerable) question why read noise gets so much worse below (manufacturer) ISO 400 and thus results in the flattening of the DR curve instead of keeping it linear. A linear curve would have left us all dropping our jaw in awe because of the even better DR.

 

[Detail Man]

Leaves the (unanswerable) question why read noise gets so much worse below (manufacturer) ISO 400 and thus results in the flattening of the DR curve instead of keeping it linear. A linear curve would have left us all dropping our jaw in awe because of the even better DR.

One could also pose the question as, "why does the read noise get so much better at ISO=400?". Engineering design and development is made of blood, sweat and tears - as opposed to serendipity