With tiny, high pixel count chips noise is always going to be an issue, and to a large degree this is more a test of the effectiveness (both measurable and visible) of a camera's noise reduction system. Designers have to balance the desire to produce smooth, clean results with the need to retain as much detail as possible (if you blur away the noise, you blur away image detail too).

The gray panels show signs of chroma noise at ISO 200 (though not to an extent you're likely to notice in a real-world scene), which is pretty well controlled unto ISO 1600 at which point a light wash of green and magenta is starting to appear. Noise reduction kicks in increasingly aggressively from ISO 200 upwards, with less and less detail appearing in every shot. Despite this, luminance noise becomes a major problem at ISO 1600. We'd recommend avoiding the ISO 1600 and 2000 modes.

ISO 64 ISO 100 ISO 200 ISO 400
100% Crops
ISO 800 ISO 1600 ISO 2000

Noise graph

Indicated ISO sensitivity is on the horizontal axis of this graph, standard deviation of luminosity is on the vertical axis.

The Luminance Gray line on the graph dipping between ISO 100 and 200 is a dead giveaway that noise reduction is being applied more aggressively at ISO 200 and this is borne out by the loss of detail we saw in the crops further up the page. The other thing we saw is that noise reduction between ISO 200 and 800 was blurring much of the detail out of the image, which explains how the noise graphs are able to stay so flat up to ISO 800. After that point noise shoots up though, having looked at the ISO 1600 images, we don't really need a graph to show us this.

Low contrast detail

What the crops and graph don't show is the effect of noise reduction on low contrast fine detail such as hair, fur or foliage. An inevitable side effect of noise removal is that this kind of detail is also blurred or smeared, resulting in a loss of 'texture'. In this test the crops below show the effect of the noise reduction on such texture (fur) as you move up the ISO range.

ISO 64 ISO 100 ISO 200 ISO 400
100% Crops
ISO 800 ISO 1600 ISO 2000

As the noise graphs suggest, noise reduction isn't really an issue at the lowest two settings but kicks in quite aggressively as low as ISO 200. Much of the fine tonal distinction is lost - a problem that just gets worse at ISO 400 - and, by ISO 800, all the fine detail has been smeared into indistinct mush. At ISO 1600 and 2000 there is virtually no detail to see for all the noise. The luminance noise gives the impression of there being more detail than the ISO 800 image but the distinction between tones has gone. Splodgy chroma noise and grainy luminance noise do nothing to improve an underlying image devoid of almost any detail. The loss of contrast in the ISO 2000 sample is dramatic.

Electronic image stabilization

Unlike its larger, more expensive, brother, the P50 does not feature optical image stabilization - relying instead on "electronic vibration reduction." The theory is that the camera tries to calculate what the effects of vibration would have been on the image, then correct for them (Imagine the opposite of Photoshop's 'motion blur' feature). Unfortunately, the feature is not only fairly ineffective - it only will only take effect in very specific circumstances - but the results can be pretty unpleasant. Here is a real-world example.

100% crop 35mm equiv., F2.8

eVR can either be set to "Off" or "Auto." If it is set to "Auto," the effect is applied where is can be. When set to "Off," the camera still assesses whether it can apply eVR and provides that option when reviewing images. Here are some sample images - the first is the most extreme shot that eVR tried to cope with in Auto mode, the second is the worst taken without eVR that still gave an eVR option. The final image is the second image with eVR applied.

eVR ON eVR OFF Applied afterwards

Essentially eVR doesn't seem any more effective than trying to sharpen the image using an unsharp mask in post-processing. Of the 300 real-world images we'd taken with the P50 when we conducted these tests, eVR had been, (or could be), applied to fewer than a quarter of all images.