I've been enjoying Marc Levoy's Lectures on Digital Photography and have reached this point.

enter image description here

Marc has broadly said:

  • Sensors have got better at reducing noise
  • But pixels are gotten smaller so there is more noise
  • These effectively cancel each other out.

Unfortauntely his chart stops at 2008.

I've been unable to find an updated version of this chart. Does the SNR still remain constant when we add sensors up to 2018? (not counting downsizing, which he goes on to)

  • 2
    \$\begingroup\$ Does he still imagine that today? I vote that we have less noise today. Pixels may be more numerous and smaller, but noise has gotten tremendously better the last few years (since 2008), easily allowing higher ISO now. The past could not even imagine that. \$\endgroup\$
    – WayneF
    Sep 25, 2018 at 16:19
  • 3
    \$\begingroup\$ Smaller pixel sensors have more noise, that's physics. Image processors are always getting better at covering up noise, that's math. Taken together, it's hard to quantify what the actual sensors are picking up these days, but i've still not seen lower measured noise than old 4-6mp DSLRs. \$\endgroup\$
    – dandavis
    Sep 25, 2018 at 16:28
  • 2
    \$\begingroup\$ Note that you're misinterpreting the conclusion from this slide, at least as I read it without hearing the talk. Even in 2008, for the same display area SNR was improving over time. \$\endgroup\$
    – mattdm
    Sep 25, 2018 at 17:06
  • \$\begingroup\$ Hi @mattdm - that's what I meant by "not counting downsizing, which he goes on to" it's a fairly interesting distinction he makes. \$\endgroup\$
    – user67208
    Sep 25, 2018 at 17:15
  • 1
    \$\begingroup\$ Right, what I mean is: "not counting how images are actually ever actually used" is a caveat which turns the conclusion on its head. \$\endgroup\$
    – mattdm
    Sep 25, 2018 at 17:26

2 Answers 2


This data was iffy then — not really enough data points, and the trendline is dubious:

xkcd curve fitting
Source: a very timely xkcd

That said, the company DxOMark does measurements of camera sensors all the time, designed to be resolution-neutral. Here's a chart of the "Sports" score, which is based on SNR, from all tested APS-C camera models from 2002 to 2018:


Given the cartoon above, I won't try to draw a line, but

  1. It's pretty clear that there's a slight upward trend
  2. That trend might not have been obvious in 2008, but it seems like it's probably actually also there.

If you look at overall scores for the same cameras, which include dynamic range and color depth, you'll see the same general sort of upward trend, although it's arguable that there's more growth to around 2010 after which it kind of levels off.

In practice:

  1. All cameras over the last decade do very well on these measurements.
  2. More megapixels don't seem to be hurting.
  3. There are cameras within any given year's cohort which would fit right in ten years later or ten years earlier. That is, waiting until next year is unlikely to give you a see-in-the-dark miracle.
  4. Don't worry too much about this. All of the cameras produce excellent results even in very, very little light.
  5. None of these measurements really matter to making good photographs.
  • 1
    \$\begingroup\$ Clearly, the trend is best shown by clicking "smooth lines" in Excel. \$\endgroup\$
    – scottbb
    Sep 25, 2018 at 17:43
  • \$\begingroup\$ It would be interesting to see the DxO chart with all of the lower end models that use 'recycled' sensors from their higher end stablemates (e.g. the basement entry level Canon T6 introduced in 2016 used basically the same 18MP sensor as Canon's top APS-C model, the 7D, did in 2009). I think the trend would be even more noticeable. \$\endgroup\$
    – Michael C
    Sep 25, 2018 at 18:18
  • \$\begingroup\$ @MichaelClark Click the link and try your hand at some of the selection filters.... \$\endgroup\$
    – mattdm
    Sep 25, 2018 at 18:27
  • 1
    \$\begingroup\$ With just the Canon 18MP (blue) and 24MP (orange) rehashes marked. imgur.com/a/ZvfBAiZ The 24MP has two versions, with and without Dual Pixel CMOS AF. \$\endgroup\$
    – Michael C
    Sep 25, 2018 at 19:00
  • \$\begingroup\$ The chart shown by the OP shows the per raw sensel noise. The DxO chart you are showing gives the pixel noise on a rescaled image. The upward trend you see is perfectly consistent with the slide shown by the OP, which states “for the same display area, SNR has been improving”. \$\endgroup\$ Sep 26, 2018 at 7:36

It seems that, in general, most camera makers have been content to trade most, but not quite all, of the gains they've made in terms of sensor efficiency in exchange for more megapixels with roughly the same overall performance with regard to signal-to-noise ratio as their older, lower resolution sensors had.

There are some notable exceptions in terms of specific models. But when that is the case, there are often two similar models offered, one with higher resolution that means lower SNR at the sensel (pixel well) level and the other with lower resolution but larger sensels with better SNR for each discrete photosite.

Take for example the first generation Sony α7 model line, which had three different versions:

  • The high resolution α7R has a 36.4MP full frame sensor
  • The balanced α7 has a 24.3MP FF sensor
  • The high sensitivity α7S has a 12.2MP FF sensor

The second generation of the α7 model line was similar, with 12.2MP, 24.3MP, and 42.4MP sensors inside the α7S II, α7 II, and α7R II, respectively. So far we've seen two models in the third generation: the 42.4MP α7R III and the 24.3MP α7 III.

Note that often manufacturers will continue to use sensors that were "cutting edge" when introduced in an upper tier model. Later models lower in their product range will get what is essentially the same sensor. Perhaps the classic example would be the 18MP APS-C sensor first introduced by Canon in their original 7D back in 2009. It has appeared in numerous models since, including the basement entry level Rebel T6/1300D that was rolled out seven years later in 2016.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.