(commentary)
The very first sensor I played with in a lab over two decades ago wasn't even capable of a VGA-sized image (480x640). Today, we've got DX sensors at 18mp (Canon) and 24mp (Nikon), and FX sensors that go higher in pixel count. One of the omnipresent marketing trends in digital cameras has been "now with more pixels," to the point where we have a 41mp smartphone from Nokia.
But what I want to write about today is the ubiquitous 24mp sensors now in all Nikon’s current DX cameras. 24mp DX is a lot of pixel density. Will the pixel madness ever end?
Yes and no. Yes in the sense that more pixels present very little visible gain for most users above where we’re at today. No in the sense that camera makers don't seem to have figured out anything else to market.
Now, before I go on, I need to repeat something I've written before: I'll always take "more sampling" if given it, all else equal. Essentially it's a bit like statistical polls: the more samples you take, the more accurately you can predict to a full population. The more data we collect, the more likely that our data accurately represents reality. Thus, given the choice between shooting with a 12mp D700 or a 36mp D800, I'll take the D800. First, sensor technology improved in the interim between those two cameras, so it's not "all else equal," but rather "newer is (slightly) better." Second, 36mp gives me more meaningful acuity to reproduce small detail and edges. It was a no brainer to trade in my D700 for a D800, given the work I do.
The same thing is true for DX: given a choice between shooting with a 12mp D300 or a 24mp D7100, I’ll pick the D7100, for the same reasons I’ve just given in the FX example.
But we're now into a slightly different land for 95% of the shooters in the world than for the 5% (or smaller) group of perfectionists to which I belong.
Every technology eventually hits a "good enough for most people" point. In audio, it was when most people couldn't hear a meaningful difference and convenience won out. In computers it was when most people didn't notice that the things they do every day (word processing, email, web browsing, etc.) didn't speed up, and price won out. In photography it's when most people can't see a difference, and convenience and size will win out.
With DX, we're at the "good enough" point, I think. I'd defy most people, under their normal shooting conditions, to be able to distinguish visually between a 16mp D7000 and a 24mp D7100 in their output. That's not to say that there aren't benefits to the newer camera, but rather that most people won't see those benefits in their intended uses. To some degree, I'd argue that we were already there with the 12mp D90. But with each iteration now I think more and more people will be coming to the same conclusion: the extra pixels aren't showing up as a huge benefit. (With one caveat: the heavy croppers. I'll get to them in a minute.)
Why that difference isn't really showing up is a large intersection of things. Back in 2003 I tried to predict the point at which we'd hit the "can I see a difference" mark. I calculated that it would be 24mp for DX. What went into my calculations and assumptions were a lot of things. First, there's diffraction. Once we have pixel sizes where diffraction overlaps most shooting apertures (and remember, most of our DX lenses are f/5.6 at their long focal length, and we don't really have much that's as fast as even f/2.8), we tend to not get any huge resolution gains at the sensor. Even removing the AA filter doesn't necessarily make a big difference, as the diffraction impacts can be higher than the AA impact at many "normal" shooting apertures.
The lenses themselves can resolve more than the sensor can record, but as the sensors grow in resolution we start to see exactly how the lenses perform. Whereas the old 6mp and even 12mp sensors hid a lot of the imperfections of the 18-200mm lens, today the 24mp sensors clearly show its weaknesses. Weaknesses that are enough to keep most people from seeing the few gains. After all, if the center does resolve a bit more but the edges start to show how blurry they really are, would most users think they’ve gained something?
But the biggest factor really is how photos are being used. I’m on a Retina MacBook Pro as I write this. That’s a whopper of a display, actually, with the potential for 1800 x 2880 pixels being shown. But what’s that in megapixels? Oh, only 5.2mp. So here I’m using one of the highest resolution displays available and best case is I’m seeing one fifth of the 24mp my DX camera can capture.
Okay, what about printing? Running an Epson inkjet at about the maximum number of dpi I’d ever consider sending it (360), I’m still getting a 17” print on the long axis. Using my rule of thumb about how large you can get a decent print from a sensor on an inkjet (with careful post processing), we’re at a 32” print for 24mp. Be reasonable now: how often have you printed at 17”, let alone 32”? And remember, those of you reading this are probably at the top of the photo enthusiast chain. The moms and pops and new couples and others buying most of the DSLRs never get that far. The biggest reason they bought a DSLR is speed of focus, the second biggest being the need to pull far away things in closer.
Which brings me to the croppers. I get hammered with emails every time I start writing about cropping, so I’m putting my helmet on and buckling up my protective vest, too.
Cropping tends to be for one of two reasons: you didn’t see it right in the field (improve the photographer!) or because you didn’t have the lens necessary to frame the subject right (buy the right lens!). But how much “crop factor” do you need for these things?
The “crop to fix composition” shooter probably doesn’t need an excessive amount of extra pixels. You can actually test yourself on this if you’d like: review a set of images you’ve taken and measure the pixels in the final image. If you’re shooting at 4000 x 6000 (24mp) and most of your final images are between 3500 x 5000 and 4000 x 6000, then you’re probably doing this type of cropping. So what’s that really work out to? You have a 24mp camera that you’re using more like a 16mp camera at worst case. At this level, you’d need a 36mp camera or so to get actual 24mp images guaranteed all the time. But do you even need 24mp? See my comments about printing.
If you’re cropping more than that, then it’s not that you’re not quite getting composition right, you’re simply using the wrong lens. However, here there’s another solution: use a camera with a smaller sensor! Again, let me illustrate: if you’ve got a DX camera (1.5x crop) that’s 24mp and you’re ending up with 3000 x 4000 images (12mp), you’re probably better off with a Nikon 1 (2.7x crop) camera and the right lens.
Indeed, a lot of folk get this backward. What you want is a large sensor camera for your wide angle work, a smaller sensor camera for your telephoto work. Consider this combo: a Coolpix A (16mp 28mp equivalent) and a V2 with the 30-110mm lens (14mp 85-290mm equivalent). Or move upscale and use a D800 with a top Zeiss wide angle and a D7100 with the 200-400mm for telephoto. Do you really need more pixels if you have either of those combinations?
Long story short: if you’re an extreme cropper for distant objects, you’re probably using the wrong sensor size and lens combo.
But back to my original point: most of us have what we need these days in terms of pixel count. Any more resolution we get from a sensor at this point is gravy, since we’re not actually taxing the pixels we’ve got today. Again, a magazine double-truck (two page spread) is within the realm of a 16-24mp sensors with pixels being buried at the equivalent of 300 an inch. I have no problem pushing a D7100 to ISO 3200 for a double-truck, and even ISO 6400 probably isn’t a problem with my post processing. So what more would I need?
I found some old (1950s) four foot wide prints recently, taken with a variety of 35mm to MF cameras. They don’t have a huge degree of acuity or detail to them. Less than I’d expect from a top end digital camera blown up that big today. I could see some lens issues, too. We may be getting a little picky about our pixels, therefore.
Don’t get me wrong. I’m only talking about pixel count here, not pixel integrity. I’d love to get the last stop or two we can out the underlying photosites. More dynamic range and less noise simply makes it easier to deal with getting an image ready for view the way I want it to look. But do I need more than 24mp DX? No.
Someone at Sony gets this right. The Sony RX-10, for example, is a smaller sensor than DX, but the lens out front is f/2.8 constant aperture. That lens at 200mm equivalent is making up for the small sensor size. This is one of the things that bothers me about Nikon’s DX lens lineup: we have one old f/2.8 lens, an f/1.8 prime, and…well, that’s it. To get low light performance out of these 24mp DX sensors that is exceptional (as opposed to good), we end up putting other non-Nikon lenses out front: the Tokina 11-16mm f/2.8, the Sigma 18-35mm f/1.8 or 17-50mm f/2.8, for example.
So where can you get more DX performance? Simple: the lens. Sharper lenses with fewer flaws, faster lenses that let you use lower ISO values. Any f/2.8 lens gains you a half stop from the kit lenses right off the bat. At the telephoto end of a zoom, the constant f/2.8 gains you two stops. That’s a lot of visible difference, whereas some more pixels won’t be.
Thus, here’s my tip for the day: instead of going for 24mp over the 16mp you’ve got, or wishing for more than the 24mp you can get today, think about the lens out front first. You can impact your DX results far more by judicious lens choice than you can buying megapixels now.
Let’s assume for a moment that the best optical performance of a lens is one stop below maximum aperture. Your kit lens at many focal lengths will thus need to be at f/8, which is beginning to be diffraction impacted on a 24mp DX sensor. But the f/2.8 lens is at f/4, which isn’t. So not are you potentially gaining two stops of ISO, you’re also avoiding the worst of the mechanism that takes away some of your resolution gains.
Bottom line: look carefully at your lens choices if you’re pushing up the megapixel count ladder.
