[falconeye]

This statement would be true ...

I don't know what to make of this post.
Let's make it short: I disagree.

[richard balonglong]

Wouldn't it be nice if Pentax introduced 2 variants of their new ff ? Canon and Nikon have done this before.

A two sensor option would be intriguing... One aimed at those who have the consumer urge for the highest number of pixels and version two to be a high Dynamic Range variant

As a photographer working in two kinds of fields, editorial and commercial, what I want and need are two different FF Pentax bodies: one for high megapixel and one for incredible low light performance...

[Ratcheteer]

As a dreamer....

i would want a dslr with interchangeable shutter, interchangeable focus screens and full coverage pentaprism viewfinder, and the old fashion needle meter instead of the digital ones we get in the viewfinder just for nostalgia.

[Fenwoodian]

I'll take more pixels over DR.

If I am at a high contrast scene, I'll use HDR to create better DR.

[Ratcheteer]

If I am at a high contrast scene, I'll use HDR to create better DR.

whatever floats your boat at the end of the day it still comes down to personal preference

[biz-engineer]

I don't know what to make of this post. Let's make it short: I disagree.

You can disagree. Eventually you disagree or agree. By simplifying ideal models, we may come up with unrealistic conclusions. Practically, the read noise is not the sole noise contributor. Each source of noise contribute to a different extent based on ISO and incoming signal level. Practically, the A7s (larger pixels) outperform the A7r (smaller pixels) in terms of per pixel noise and dynamic range from ISO200 onward. Below ISO200 and in complete darkness, the DR advantage goes to the A7r that benefit from spacial averaging of read noise since there are more pixels being read. For all ISO values above 200, the A7s has better per pixel noise and DR, not only in the lab, but also in real world conditions (unlike measuring the noise level of a dark frame).

[Simen1]

I would find i hard to choose between a high resolution HR model and a high DR model. The HR model should be focused on still photography, slow frame rates and so on. The HDR model should have close to 12 Mp and high frame rate in order to be able to crop to 3840*2160 video. If the HDR model have as fast full readout as 120 fps it could take four exposures for every 30 fps frame output, meaning high resolution mode on 4K video. Maybe thats a to optimistic speed. 30 fps readout in 3840*2160 should do for a start. The 12 Mp camera should have astrotracer, longer exposures (5 min) and higher ISO settings then the HR model. If i get rich soon i will buy both.

[Gimbal]

One way of looking at the large vs small photon wells is the old rain bucket comparison. Common sense dictates that a large bucket will not fill up as fast as a small bucket when left out in the rain.

But that is only true if the larger bucket is deeper then then smaller one. If their depth is the same (as with image sensors where the only difference is pixel size) the large and the small bucket will fill up at the same time.

[Simen1]

That is correct Gimbal. Photon capasity per unit square area are almost the same for small and large pixels.

But there is more to consider. Read noise in all cases and thermal noise in really long exposures. These noise sources increases per unit square area when the pixels shrink. Thats why large photon buckets cannot be emulated good by binning smaller ones. Read noise is especially evident on high ISOs (low signal to noise ratio).

[Gimbal]

But there is more to consider. Read noise in all cases and thermal noise in really long exposures. These noise sources increases per unit square area when the pixels shrink.

Why would read noise increase due to a smaller pixel?
Or is it because you get read noise four times instead of one time if the size difference is four times (for example)?

[Simen1]

Yes, thats correct. Also read noise can increase if you demand the same frame rate as a sensor with fewer pixels because you need to A/D more pixels in the same time (MPix/s). That means higher clocked and hotter A/D converters. The heat adds noise.

[biz-engineer]

Noise is integrated into each pixel capacity. Commonly used thermal noise model is: k.T/C , k boltzmann constant, T temperature in kelvin, C = well capacity. For long exposures, pixel specific dark current leakage (that's why dark frame substract works for long exposure, for short exposures dark frame substract add noise due the random nature of the noise). Back to k.T/C , as pixel size inceases, the current source (from photon conversion) charges a capacity (designed by cell designer), sensitivity depends on both light sensitive area and capacity value. So we can have a large pixel with small "well" capacity, or a small pixel with large "well" capacity. It's a design decision. For a constant cell sensitivity value, larger pixels needs a larger capacity. At the same temperature of the sensor, larger capacity lead to lower noise. Cooling down the sensor reduces noise. For example, some imagers used for military, aerospace or x-ray medical imaging have pixels 10 x times larger than what you have on your Pentax camera, lower ISO, and the sensor is sealed into a cavity that is cooled down to -170C, so that to achieve much lower noise levels.

Back to camera sensors, unlike what Falk wrote in his blog, adding 4 small pixels (of 1/4th the size of a large pixel) does not provide a 6dB higher signal to noise ratio versus 1 large pixel, simply because if you have 4 times the signal of the smaller pixels, you don't have 2 times the noise (Falk statement) , you also have 4 times the noise because the capacity units of each smaller pixel are also smaller.

Then there is the read noise, which, when there is zero signal (black) is higher than thermal noise, but as soon as there is some signal (fine gradation of shade in a real photo), the shot noise rapidely becomes more significant than read noise. And of course, this is what we see on images, worldwide: for equivalent techno, smaller pixels create more noise on images, less DR per pixel.

In short, for same iso, larger pixels provide more DR and less high iso noise , larger sensors provide with smaller or larger pixels, deliver more IQ and DR. No need to align one page of mathematical equations to look smart to demonstrate with utter arrogance the opposite of reality.

[Gimbal]

At the same temperature of the sensor, larger capacity lead to lower noise.

That statement isn't obvious to me, why would a larger capacity lead to lower noise? (Unless you also feed the capacity with a stronger signal from a larger sensor, that is.)

[biz-engineer]

That statement isn't obvious to me, why would a larger capacity lead to lower noise? (Unless you also feed the capacity with a stronger signal from a larger sensor, that is.)

Thermal noise is independant from signal, defined by k*T/C , only depends on temperature and capacity.

[Stagnant]

In my experience (aps-c), more megapixels are an advantage in low light situations, because in postprocessing you can always downsize the larger file and carefully apply a certain amount of noise reduction. I did some 15 * 20 (cm) prints of photos processed that way and they were perfectly acceptable. I own Pentax K-3. Of course if you intend to print large, absolute low light performance is important.
Just my two cents.