15mp RAW file ISO 25600 Comparison. D4, D800E, 5D3, D600, 650D, D7000

[An Exception]

Downsampling is notorious for getting rid of hot pixels. Removing over half of the pixel count is bound to make things look better than they are at 36mp 100% crop. A D4 is only 16mp to start with, so thats by far the "truest" representation of ISO performance.

I ticked the option to down-sample all the files to 15mp raws. The purpose of this is to end up with raws that have the same output resolution as each other. This will give a more accurate representation of the noise, and better simulates what you would get in the real-world if you made 'X' size prints from each of the camera's.

Obviously, see the above quote.
This is why if your going to compare camera's noise performance, you first normalise the output size. Comparing pixel level noise of camera's with vastly different resolutions doesn't tell you anything about the quality of the final output (which is the important bit).

 

[An Exception]

To be honest, all this really shows is that between the 5D3/D600/D800, you shouldn't be basing your decision on noise performance xD

kd

It shows you shouldn't be basing decisions on ISO performance. Dynamic range and pattern noise is a different story (there is a big difference).

 

[James J]

My mistake then, didn't read that bit! Nice to see we are on the same page (just reading different bits )

 

[DJ_Switch]

Whilst interesting I'd rather have seen an iso6400 comparison as out of all those cameras only the D4 and 5D Mk3 have iso25600 as part of the camera's native iso range.

 

[An Exception]

What the camera's manufactures label as 'Native' ISO is irrelevant.
The D800 apparently does not achieve ISO 6400 by analogue amplification. It can only get to ISO 1600 by analogue, to get to 6400 it uses digital gain to increase the exposure 2 stops.

What is relevant is the quality of 'X' ISO, not what the manufacturer decides to label it, as the native/extended labels are arbitrary.

 

[jonneymendoza]

5d3 still king, nikon are crap.

There i said it just to upset the "OP" lol

 

[Andy90]

All this confirms is that I still wouldn't shoot at 25600 ISO....

 

[DJ_Switch]

What the camera's manufactures label as 'Native' ISO is irrelevant.

If a manufacturer is not willing itself to stand by the quality of the output enough to include it in it's official iso range for marketing that would make it look better to the consumer then I think it is worth taking note of. The results may still prove favourable over another company's official range however it is certainly not irrelevant that they chose to omit it from the offical range. It's no huge point but it's certainly worth noting when comparing these. No doubt there wouldn't be much difference between the order seen here at iso6400 however I'd still find it interesting.

 

[2StepSteve]

FIGHT!



Seriously though, this illustrates that I'd rather not use ISO 25600 on any brand/model.

 

[An Exception]

All this confirms is that I still wouldn't shoot at 25600 ISO....

I would and do. Would much rather some grain than destroy atmosphere/create distraction with flash.

 

[D.P.]

If a manufacturer is not willing itself to stand by the quality of the output enough to include it in it's official iso range for marketing that would make it look better to the consumer then I think it is worth taking note of. The results may still prove favourable over another company's official range however it is certainly not irrelevant that they chose to omit it from the offical range. It's no huge point but it's certainly worth noting when comparing these. No doubt there wouldn't be much difference between the order seen here at iso6400 however I'd still find it interesting.

All that is really says is that Canon have a weaker threshold on noise.

It really is irrelevant. The more recent canon crop sensor supposedly offer higher ISO ranges than the slightly older cameras, but the sensors are all nearly identical and offer the same noise performance.

What ultimately matters is the Signal to noise ratio and the quantum efficiency of the sensor.

 

[James J]

5d3 still king, nikon are crap.

There i said it just to upset the "OP" lol

Its not the gear, its the photographer

 

[alexthecheese]

Again, probably wishing because I want to buy a D800 one day, but it does look better to me than the 5D3 - particularly noticeable under the eyes. Yay.

 

[NickK]

You daylight boys have it easy for noise!

If you want to take shots in dark situations think about the following:
a) temperature of the sensor (let it cool in DSLRs) or stick cooling on it
b) noise/heat caused by the LCD and the CCD amp being on (used to read the image from the CCD sensor itself into voltages that are then turned into bytes of your image).

Techniques for DSLRS:
a) multiple shots of the same target to remove random noise (strengthens the signal to noise ratio) - called stacking
d) dark shots (lens cap on) to remove the read noise (that appears as a static pattern on the sensor) - called darks (used for preprocessing the images before stacking above.

By subtracting the noise in post processing you'll improve the noise.

Fast moving targets are hard - however if you're taking a photo of a church with a DSLR then it may be better to run at ISO 400 or 800 with multiple long exposures then take some darks (same length exposures with the lens cap on and stack them) then stack taking out the noise using both "lights" (the images) and "darks" (the noise).

 

[An Exception]

Again, probably wishing because I want to buy a D800 one day, but it does look better to me than the 5D3 - particularly noticeable under the eyes. Yay.

The D800 is about on par with the 5D3. The D800E has a slight edge (AA filters reduce light hitting the sensor).

 

[D.P.]

You daylight boys have it easy for noise!

If you want to take shots in dark situations think about the following:
a) temperature of the sensor (let it cool in DSLRs) or stick cooling on it
b) noise/heat caused by the LCD and the CCD amp being on (used to read the image from the CCD sensor itself into voltages that are then turned into bytes of your image).

Techniques for DSLRS:
a) multiple shots of the same target to remove random noise (strengthens the signal to noise ratio) - called stacking
d) dark shots (lens cap on) to remove the read noise (that appears as a static pattern on the sensor) - called darks (used for preprocessing the images before stacking above.

By subtracting the noise in post processing you'll improve the noise.

Fast moving targets are hard - however if you're taking a photo of a church with a DSLR then it may be better to run at ISO 400 or 800 with multiple long exposures then take some darks (same length exposures with the lens cap on and stack them) then stack taking out the noise using both "lights" (the images) and "darks" (the noise).

Read noise over normal exposures is almost non existent, over long exposures it can be significant and is which by default DSLRs will do a dark frame subtraction on long exposures.


Frame stacking is entirely pointless for most common uses of a DSLR. The cameras has to be mounted on a tripod and the subject static. If you are on a tripod then you can be at base ISO and just take a long exposure.

It used to be slightly more common a long time ago when sensor noise at base ISO was measurable but these days the sensors are essentially noise free over standard luminance values. Shadows can display noise when pushed but then you need to be thinking of doing multi-frame HDR with different exposures and not stacking multiple frames of the same exposure.

 

[Logsi]

I would and do. Would much rather some grain than destroy atmosphere/create distraction with flash.

A picture lasts forever but a flash lasts for a mere split second.

I kinda thought the 5D3 had the most detail preserved but there is some weird digital artifacting that is such a common Canon trait under the chin next to the neck.

 

[James J]

You'd need to create the atmosphere with the flash though, which requires time to set up. Thats not always possible, especially for "split second" moments. The worst place to stick a flashgun for portrait photography is on the camera's hotshoe...

 

[Johnny|lucidcomposure]

Yes you are wishing that.

You think the 5D3 has less noise than the D4? Oh dear...

Oh dear?

Quote me again where I said the 5D had the better noise handling? I have never given a flying "$@% about noise. I actually quite like it. I have often found some grain adds a whole load to pictures in some instances.
I said which I personally thought looked the nicest. I think the 5DIII had the best sharpness, contrast.

You might look graphics and charts and think what you like. I saw a picture of a face which has far better colour, depth and contrast around the lips, nose and eyes.

So again, in my opinion I think the 5DIII looks the nicest.

Based on all that chart stuff the D800E looks a clear winner to me but I don't print and mount charts. I print pictures though so I only really give a hoot about that.

 

[An Exception]

A picture lasts forever but a flash lasts for a mere split second.

I kinda thought the 5D3 had the most detail preserved but there is some weird digital artifacting that is such a common Canon trait under the chin next to the neck.

Destroying ambience of candle light with flash is sacrilege imo. But more importantly, the pictures look nothing like how people remember how it actually was.
If you use flash that's cool, I do to if there is no other option, but different togs have different priorities depending on how they work.