Back by popular demand: Nex-7 vs OMD

Starting with some facts:

• Sony Nex-7: APS-C = 23.5 x 15.6mm, 24.7Mp (24.3Mp effective). 3:2 aspect ratio.
• Olympus OMD EM5: MFT = 17.3 x 13mm, 16.9Mp (16.1Mp effective). 4:3 aspect ratio.

(Numbers taken from dpreview reviews).

• So, using total pixel count (24.7), a Sony Nex-7 has 367 mm2, or 14.8 umm2 per pixel site
• And the EM5 has 225 mm2, or 13.0 umm2 per pixel site.

For all practical purposes, the EM-5 pixels are 10% smaller than the Nex-7!!

Another way to consider this:

• The EM-5 has a sensor which is 60% the size of the Nex-7, with 16.9Mp versus 24.7Mp.
• 16.9Mp/24.7Mp = 68%. 
• To put 68% of the pixels in 60% of the available space means a 10% smaller pixel site.

Ok, technically, the pixel site is limited by routing on the top layer, which is different in a 4:3 arrangement over a 3:2 arrangement, so at lower level detail there will be more specific details. Here I just suffice to compare the pixel size density, not the actual pixel site size.

Meaning I am comparing the 14.8 versus the 13.0. Whereas the actual pixel site may only be half that.

Just to compare two more numbers:

• Nex-5N has 23.5 x 15.6 into 16.1Mp effective (I assume 16.9Mp total) = 21.7 umm2 per pixel.
• RX-1 has 35.3 x 25.8 into 24.7 Mp (24.3Mp effective) = 862 mm2. Or 34.4 umm2 per pixel.
• A D800 increases the pixel count to 36Mp effective, or about 23.3 umm2 per pixel. 

RX-1 has best ISO response, with its largest pixel site. On a per pixel level, a Nex-5N can be expected to yield similar noise as the D800.

Similarly, back to our EM-5 and Nex-7, on a per pixel level, the EM-5 and Nex-7 can be expected to yield similar noise.

Per the reviewers, Olympus plays a simple trick with us, by overstating the ISO numbers in the EXIF. ISO noise is proportional to the area, so there is a 1/68%, or about a 1.5x factor in actual ISO involved.

• Example: if the Nex-7 reports ISO = 1000 for a given aperture, speed, lens FL, etc., the true ISO of the EM-5, based on sensor/pixel site size, should be 1000. Olympus reports it as 1500. It factors in the discrepancy in total pixel sites (and noise floor), but then must compensate in other ways (e.g. NR reduction?). 

At low ISO, typically ISO below 1200, per the Nex-7 experience, the noise is fairly low. Below 800 there is virtually no noise in the Nex-7. The same would apply for the EM-5. So - for low ISO, the Em-5 and the Nex-7 produce similar images, based on pixel site sizes, with the exception that the Nex-7 has 50% more detail (since it has 50% more pixels). 
But at high ISO, the Nex-7 is known to get very noisy and you are left with aggressive NR (and get the water oclor effect), or down sample, e.g. to Nex-5N resolution. The latter is quite workable, and we have agreed here that the Nex-7 and Nex-5N, both down-sampled to 16Mp (native for 5N) have a very similar noise performance, resolution, rendering.

• The Nex-5N scores better in low light because it will use ISO 3200 by default, whereas the Nex-7 will limit itself to ISO 1600. This is another issue. 

Back to the EM-5. It also has the same noise characteristics, and to perform at the noise level of the Nex-5N, it has to down sample at a similar ratio as the Nex-7, meaning that its image size would have to be reduced to 16/24*16 = 10.7Mp. 
Since the EM-5 does not use this down-sample technique, the only alternative is that NR is applied to hide the bad color pixels. Many algorithms exist - see P&S techniques  but they work only up to a certain ISO level before washing out all details (water color effect). Perhaps a motivation for overstating ISO levels?

--
Cheers,
Henry