The myth of the large sensor "high ISO advantage"

[Aur]

Horribly misleading illustration. What it should be showing is a larger number of smaller buckets covering the same area as a smaller number of larger buckets, and without the wasted space since that really isn't an issue in reality.

the only way the smaller sensor would have more buckets is if the smaller sensor would have a larger resolution, which is never the case

no matter how you turn the argument around, a larger sensor allows for larger pixels while maintaining similar resolutions

it's basically the same argument HTC made, larger pixels allow you to catch more light for the same resolution, a larger sensor makes it easier to do this, but it's not the sensor size causing the cleaner ISO, it's the size of the pixels

 

[Lab D]

the only way the smaller sensor would have more buckets is if the smaller sensor would have a larger resolution, which is never the case

So if I put a 200mm lens on a Canon 6D and take a picture of a building 1000 yards away, then put the same lens on a Samsung NX1, stand at the same spot, and take a picture of the same building...which image has the "larger resolution"?

Why do I think in daylight the building in the NX1 image would be significantly more detailed? Do you disagree?

 

[cainn24]

Horribly misleading illustration. What it should be showing is a larger number of smaller buckets covering the same area as a smaller number of larger buckets, and without the wasted space since that really isn't an issue in reality.

the only way the smaller sensor would have more buckets is if the smaller sensor would have a larger resolution, which is never the case

I was replying to your suggestion that larger sensors only have an advantage if they also have larger pixels, which simply isn't true.

The Nikon D7000 (APS-C) and the D800 (FF) have pixel sizes that are very similar. Yet the D800 maintains a noise advantage nonetheless.

Or DxO Labs if you prefer:



An E-M1 vs D7100 comparison provides another example.

 

[Allan Olesen]

The A7r probably uses the older sensor from the Nikon D800 while the a6000 probably got a completely new sensor.

So the 3% difference you have found can probably be adequately explained by the age difference. (We got a one stop improvement in 6-8 years, but it has slowed down now because we are closer to the theoretical limit).

 

[SKersting66]

Then you'll need to explain how a 36MP D800/D810 has better normalized SNR/DR than the 16MP D4s at low ISOs (even ones that both cameras share). That simply doesn't happen if the D800/D810 is collecting less light.

A "pixel" isn't a separate piece of silicon. It is an area on the silicon wafer that is separated from other areas by a mask. Within the masked area are the source followers. For a given generation of technology the size of the source followers is relatively fixed. So, as you increase the number of pixels you increase the required masked area.

The reason the D8xx keeps up with the D4 at lower ISO's is because of "oversampling" (normalizing) with *enough* information to generate a good image.

To take the rain analogy further; low light is like trying to capture a light sprinkle and bright light is like trying to capture a heavy downpour... you don't need much time in a downpour to get soaked. In fact, you can use higher ISO's combined with smaller apertures/faster SS's in good light with fewer negatives because there is still enough information collected. The high ISO/low light penalty is really primarily due to a lack of adequate signal (photons), not the amplification (ISO).

ISO 12,800

Link to image on flick'r if you want to see it full size

--
https://www.flickr.com/photos/skersting/

 

[Lab D]

The A7r probably uses the older sensor from the Nikon D800 while the a6000 probably got a completely new sensor.

So the 3% difference you have found can probably be adequately explained by the age difference. (We got a one stop improvement in 6-8 years, but it has slowed down now because we are closer to the theoretical limit).

So the A6000 uses the latest technology. So how does the old E-M1 with a smaller sensor match it for DR at almost every ISO?

 

[Aur]

Horribly misleading illustration. What it should be showing is a larger number of smaller buckets covering the same area as a smaller number of larger buckets, and without the wasted space since that really isn't an issue in reality.

the only way the smaller sensor would have more buckets is if the smaller sensor would have a larger resolution, which is never the case

I was replying to your suggestion that larger sensors only have an advantage if they also have larger pixels, which simply isn't true.

The Nikon D7000 (APS-C) and the D800 (FF) have pixel sizes that are very similar. Yet the D800 maintains a noise advantage nonetheless.

Or DxO Labs if you prefer:



An E-M1 vs D7100 comparison provides another example.

there are other things influeing ISO performance other than pixel size

such as the ability of sensors to interpret the signal and disregard the noise

"The intrinsic radiometric resolution of a sensing system depends on the signal to noise ratio of the detector. In a digital image, the radiometric resolution is limited by the number of discrete quantization levels used to digitize the continuous intensity value."

that's why you have like 0.3 megapixel sensors for hyperspectral imaging, they have a ridiculously good signal to noise ratio, they are so sensitive that they call tell you what material it is capturing, because of their ability to interpret the signal, that's what they're used for, recognising objects through light

however, that doesn't nullify the fact that larger pixel size allows you to capture more light, and should give you better ISO performance

 

[Allan Olesen]

The A7r probably uses the older sensor from the Nikon D800 while the a6000 probably got a completely new sensor.

So the 3% difference you have found can probably be adequately explained by the age difference. (We got a one stop improvement in 6-8 years, but it has slowed down now because we are closer to the theoretical limit).

So the A6000 uses the latest technology. So how does the old E-M1 with a smaller sensor match it for DR at almost every ISO?

Because the old EM-1 is a mystery. It has always been known to perform better than it "should".

 

[MediaArchivist]

So the A6000 uses the latest technology. So how does the old E-M1 with a smaller sensor match it for DR at almost every ISO?

Could 24Mp vs 16Mp have anything to do with this?

 

[SKersting66]

Lab D wrote:

So the A6000 uses the latest technology. So how does the old E-M1 with a smaller sensor match it for DR at almost every ISO?

It doesn't, you picked the only measurement that's close. It falls notably behind in SNR and Tonality (a couple bits difference is pretty significant).

 

[PerL]

I have been waffling back and forth between moving to a Sony A7 or A6000, and looking at DxOMark's low light sensor rating I think I realized something kind of interesting.

First of all, correct me if I'm wrong, but sensitivity across formats is not really equivalent, and smaller sensors are inherently more sensitive than larger sensors for a given DoF & shutter speed length. I.e. if you have the same ISO, aperture diameter & shutter speed in front of two different size sensors, the larger sensor will have a dimmer exposure due to the given volume of light being spread over a larger area.

With that in mind, FF has 1.2 stops/2.36 times more sensor area than APS-C. So correct me if I'm wrong, but if an APS-C sensor is rated at ISO1000 for some low light S/N metric, a FF sensor with equivalent performance per unit of area will be rated at ISO2360 for the same metric, correct?

Well when you apply that math to the A6000 and A7 something interesting happens. The A6000 is rated at ISO1350 for DxO's 30db low light S/N test, and the A7 is rated at ISO2248. You do some quick math and realize that if the sensors are supposed to be equivalent, either the A7 should be rated at ISO3186 or the A6000 should be rated at ISO950. That's half a stop!

What's the practical impact of that? If you have a low light scene and two equivalent lenses wide open on the two cameras (i.e. the 24 1.8 & the 35 2.8), the A7 will have half a stop more noise. Yes, you will be able to bump up to a "higher ISO", but that's meaningless- you will have to up your ISO to have an equivalent picture (same shutter speed and depth of field) on the bigger sensor as it has inherently dimmer exposures than the smaller sensor.

Does this mean I won't get the A7? Probably not though I might go for the A7R as it somehow claws back most of that missing half a stop. Plus my legacy lenses will work a lot better on the full frame sensor naked than on a crop sensor through a focal length reducer. So there are other considerations to make. But my point is it might be time to change how we look at "high ISO" performance, and create a new metric to normalize it to sensor area. To really compare apples to apples ISO performance between formats you have to use a correction factor. If my math is correct it is pretty much the crop factor between the two formats squared. I.e. Sony APS-C has a crop factor of 1.536 vs Sony FF, so to compare the two you have to either divide the FF ISO or multiply the APS-C ISO by 1.536^2 (the 2.36 I referenced before which is the ratio of the two sensor areas).

This has other implications as well. A little 1/2.3 sensor at ISO100 is operating at a sensitivity equivalent to ISO3500 on a full frame. No wonder dynamic range and color sensitivities are so much worse.

Anyways I just thought that was interesting.

You hear this over and over from small format supporters, but less DOF = better subject isolation and more creative options.

True, but there is more to this. (see below)

Greater DOF control is one of the main reason of getting a larger sensor.

OK, the keyword there is "control". But there is a problem with this thinking.

The"everything in focus" is the curse of phone cameras and one reason 35 mm film shots looks often more attractive and interesting than modern, sharp but flat looking, digital shots with small sensors.

I am sure we agree smartphones are very different from 1 inch sensors, M43, and APS-C and lumping them in the same group is disingenuous.

Its a sliding scale. For me it starts to get boring with a typical slow kit lens on APS-C.

But if you dont care much for this creative option, sure - the smaller the sensor - the better. "At the same DOF".

Here are some points you are missing. Larger formats do offer greater control but at a price. For example, as the OP mentioned if you try to get the same DoF with a FF camera as an APS-C camera, image quality can suffer slightly. Next, the shallower the DoF the more lightly one is to have focus errors or more noticeable focus errors (front/back focus). Finally, while it is possible to reduce the DoF with software, it is near impossible to increase it (except via multi-exposures). I can take any image and make it look like it was shot with a 1 stop wider aperture pretty easily and it would be near impossible to tell (above that it gets more difficult and things such as fine hair become problematic). I can't go the other way though.

Personally I am not fond of software created shallow DOF so I would not equal real shallow DOF with that.

So we are left with, which is better for the individual? If a FF owner shoots 70+% of the time wide open, then I would never suggest switching to a smaller format. If an M43 users is happy with the DoF the is getting and doesn't want anything shallower most of the time, then switching may only serve to reduce IQ (if same DoF is use).

I like to error on the side of caution. I'll use a slightly larger DoF and make sure focus is nailed. If later I want something a little shallower, I can easily make some simple changes (just as anyone would with softening skin). If focus is off or DoF is too shallow, there is little one can do.

Of course by using caution and a slightly wider DoF, it allows me to use a smaller/lighter lens.

 

[cainn24]

Then you'll need to explain how a 36MP D800/D810 has better normalized SNR/DR than the 16MP D4s at low ISOs (even ones that both cameras share). That simply doesn't happen if the D800/D810 is collecting less light.

A "pixel" isn't a separate piece of silicon. It is an area on the silicon wafer that is separated from other areas by a mask. Within the masked area are the source followers. For a given generation of technology the size of the source followers is relatively fixed. So, as you increase the number of pixels you increase the required masked area.

If you're suggesting that the fill factor is compromised to a degree that presents as noticeably inferior normalized SNR I find that to be a dubious claim. I think that microlenses overcome this problem well enough.

The reason the D8xx keeps up with the D4 at lower ISO's is because of "oversampling" (normalizing) with *enough* information to generate a good image.

It doesn't just keep up, it gets well out in front in terms of dynamic range (noise in the shadows).

The high ISO/low light penalty is really primarily due to a lack of adequate signal (photons), not the amplification (ISO).

For sensors of the same size but different resolutions I'm pretty sure that the penalty is indeed primarly related to read noise.

If there is further disagreement the PST forum might be the best place to continue this discussion as I basically just regurgitate what all the experts in there have to say about it (and make sure it gels with my own real-world experience of course).

 

[sportyaccordy]

The A7r probably uses the older sensor from the Nikon D800 while the a6000 probably got a completely new sensor.

So the 3% difference you have found can probably be adequately explained by the age difference. (We got a one stop improvement in 6-8 years, but it has slowed down now because we are closer to the theoretical limit).

So the A6000 uses the latest technology. So how does the old E-M1 with a smaller sensor match it for DR at almost every ISO?

You're comparing apples to oranges (what relevance does DR have in a thread about SNR?), and in any case the A6000 has an advantage in maximum DR about equal to its advantage in sensor size, which for ISOless sensors like the Sonys in both of these cameras is all that matters.

 

[cainn24]

however, that doesn't nullify the fact that larger pixel size allows you to capture more light

For a given unit of sensor area this is simply NOT the case. Be careful not to conflate per-pixel performance with performance per unit area. The latter is all anyone should care about in the context of photography where it is only the final image that matters.

, and should give you better ISO performance

You get better high ISO performance (comparing sensors of equal size but significantly different pixel counts) because lower resolution sensors generally produce less read noise.

In other words you're on the money when it comes to real-world results but you're wrong about the reasons that we see those results. Any theory you have has to fit the data, and yours doesn't.

 

[Lab D]

The A7r probably uses the older sensor from the Nikon D800 while the a6000 probably got a completely new sensor.

So the 3% difference you have found can probably be adequately explained by the age difference. (We got a one stop improvement in 6-8 years, but it has slowed down now because we are closer to the theoretical limit).

So the A6000 uses the latest technology. So how does the old E-M1 with a smaller sensor match it for DR at almost every ISO?

You're comparing apples to oranges (what relevance does DR have in a thread about SNR?), and in any case the A6000 has an advantage in maximum DR about equal to its advantage in sensor size, which for ISOless sensors like the Sonys in both of these cameras is all that matters.

Well, DR will determine the amount of shadow noise more the SNR18%. Also, I love the "max DR" argument. If you shoot at ISO100 the A6000 wins, BUT if you use too high a shutter speed and ISO changes to 120 or god forbid 200, then magically the E-M1 wins. The truth is, with real word use they are equal.

 

[Aur]

some pixel size information

iPhone: 1.5µm

Canon 1200D crop: 4.3µm

Nikon full frame D750: 5.9µm

 

[Allan Olesen]

(what relevance does DR have in a thread about SNR?)

A lot, actually.

DxO's DR curve is in reality a curve showing shadow noise at a given exposure of the main subject.

 

[Tom Wilson 62]

Good point re amount of enlargement. I come from the film era where the saying was "a good big one (i.e. medium format) will always beat a good small one". The maximum enlargement size I regulary have printed is 11 x 14 inches and I often think that DPReview (and other test sites) should compare different size sensors at different ISO's as 11 x14's as then, you would more accurately see what you would get in a real life situation. Regards. Tom

 

[ultimitsu]

First of all, correct me if I'm wrong, but sensitivity across formats is not really equivalent, and smaller sensors are inherently more sensitive than larger sensors for a given DoF & shutter speed length. I.e. if you have the same ISO, aperture diameter & shutter speed in front of two different size sensors, the larger sensor will have a dimmer exposure due to the given volume of light being spread over a larger area.

Nope.

no, he is right and you are wrong.

With a given scene/exposure settings the exposures will be the same between different sensors/formats. It doesn't matter if it's a cell phone or 8x10.

You should read carefully before responding. he said:

" if you have the same ISO, aperture diameter & shutter speed in front of two different size sensors"

That would indeed result in same volume of light spread over an larger area. whether that means it is a "dimmer exposure" is a separate issue due to different understanding of what exposure means. but the resulting image will have less brightness.

The advantage of larger formats is larger pixels. It's like trying to catch rain with teacups vs catching rain with large buckets. The large buckets will have more rain (light) per pixel and that's what gives it an advantage.

This incorrect thinking should have been eradicated several years ago, it is amazing how it still perpetuates in DPR. please tell us:

1, does D600 have better high ISO than D800?

2, does D300 have better high iso than D7000, who in turn has better high iso than D7100?

 

[zxaar]

The A7r probably uses the older sensor from the Nikon D800 while the a6000 probably got a completely new sensor.

So the 3% difference you have found can probably be adequately explained by the age difference. (We got a one stop improvement in 6-8 years, but it has slowed down now because we are closer to the theoretical limit).

So the A6000 uses the latest technology. So how does the old E-M1 with a smaller sensor match it for DR at almost every ISO?

Another reason not to buy m43 -









Sony offers much better option if DR is really that important. Fuji did few years ago. I am not sure why you did not buy it since you seem to sing the DR song quite often.



--
::> I make spelling mistakes. May Dog forgive me for this.