Nikon D3S ISO Quality Tests


By Olin Lathrop, 24 March 2011

I recently acquired a Nikon D3S camera, so am trying to get a good feel for its capabilities and tradeoffs.  One feature of this camera that is supposed to be particularly outstanding is its low light capability, which means good quality at high ISO equivalent settings. 

This type of image sensor can trade off sensitivity with noise.  Better sensors will allow higher sensitivity at the same noise, or less noise at the same sensitivity.  These tests are intended to get a feeling of the noise versus sensitivity choices of the D3S, to provide guidance on what ISO settings produce acceptable results in different situations. 

The camera was set on a tripod with a 50mm fixed lens set to f/8.  The ISO sensitivity was varied by one f-stop (factor of 2) in successive pictures, with the camera automatically adjusting the shutter speed to compensate.  The shutter speed varied from 1/30 second for the ISO 200 picture and was faster for successive pictures.  All pictures were taken within a minute of each other on 9 January 2011 outdoors during the day under heavy overcast conditions.  These conditions did not change noticably over the set of test pictures.  The full frames are 4256 x 2832 pixels in resolution, which is the maximum the D3S can produce.  The highest quality mode was used. 

Two images are shown for each test picture.  The left image shows a 160 x 120 piece of the original picture, pixel replicated 2 x 2 to make the pixels easier to see on the screen.  In other words, the image on the screen is 320 x 240 pixels, but shows a blowup of a 160 x 120 region of the original picture.  To put in perspective how small a portion of the original picture that shows, consider that the original picture would about 25½ times wider at that scale.  The purpose of this image is to show the raw pixel noise. 

The right image of each pair was made by first resizing the original picture to 1024 x 681 pixels, then extracting a 320 x 240 region from it.  The full image at that scale would be 3.2 times wider.  The purpose of this image is to show the effects of the pixel noise at roughly the largest reasonable size for a picture in a web page.  Put another way, spacial resolution can be traded off to get lower noise, and this image gives some idea of the result for about a 4x shrink in each dimension.  This will give some guidance of what ISO settings are acceptable if the full frame picture is intended ultimately for display in a web page or similar resolution. 

Full frame at 640 x 426 pixels for scale reference

ISO 200

ISO 400

ISO 800

ISO 1600

ISO 3200

ISO 6400

ISO 12800

The results are for everyone to interpret for themselves.  However, my own conclusions are:

Nice job, Nikon.