aps-c vs. full-frame

An old friend of mine recently emailed me a common question about DSLRs that is worth reconsidering after the Photokina announcements:

What is your opinion on the future of APS-C vs. full-frame cameras?
Many say that the race for megapixel is pretty much dead, but the race for much higher ISO and less noise, etc (quality pixel) is still ongoing. With the price of full-frame camera bodies dropping to low $2K ranges, and some say most likely much lower in the foreseeable future (5-10 years), do you think many people with high end APS-C bodies/lenses will consider switching back to full-frame equipments?

This question actually dates back to the dawn of history (in the DSLR world, that would be 2003), but has returned with the slew of new full-frame cameras (Nikon D700, Sony α900, Canon 5D MkII) announced before and during Photokina last month. On the one hand, I don't think anything material has changed recently in the APS-C vs. Full-frame debate: the 5D retailed in the low $2Ks in early 2007, which is about what it still costs, and lenses are about the same price as well. Actually, I take that back: something radical has changed (hint: I've predicted it before), but it is peripheral to the APS-C vs. Full-frame debate.

Before I tackle that larger question, though, let me handle the specific ones:

Many say that the race for megapixel is pretty much dead, but the race for much higher ISO and less noise, etc (quality pixel) is still ongoing.

Actually, I disagree with both clauses. I see no signs of the megapixel race even slowing, let alone stopping, and I don't see any physical limitation that would prevent it from going on for quite some time. Just as an illustration, most pocket cameras clustered around 5 megapixels about three years ago, which, for my taste, was the best balance between resolution and pixel quality for a typical 1/2.5" sensor. As compacts are now rounding 12-megapixels without any slowdown in sales, that clearly was no deterrent to the race. But even that "so-2005" five-megapixel sensor has a sensel pitch of 2.22 microns, which if used on a typical APS-C-size sensor (a Sony sensor) would yield around 76 megapixels. The latest-and-greatest Canon 50D is only 15-megapixels, so you can see we have a long way left to go to even reach the point of diminishing returns, and consumers have clearly shown a willingness to go beyond that point.

On the other hand I think the race for pixel quality is nearing the end. Both Nikon and Canon have stated that their latest APS-C sensors have gapless microlenses. That is, assuming the microlenses are efficient (as normal lenses indeed are), they are already gathering all the light that is hitting the sensor. As noise in the majority of situations is dominated by photon noise (i.e., the amount of light being recorded), the only ways to further improve signal-to-noise ratio is to increase the efficiency of the microlenses, use more transparent color filters (something Canon already claims for the 5D MkII), or improve the quantum efficiency of the sensors themselves. While all of these are possible, they are by no means inevitable, and it's not even clear there is research in the pipeline that will lead to any improvements in these areas. Up 'til now, sensor manufacturers have just been increasing the light-gathering area of sensors, but now that that is essentially 100%, there's little left to be done.

(Note: that most of the reduction in noise we've been seeing in recent DSLRs is just that: software noise reduction. Removing noise in post-processing can never increase the signal-to-noise ratio, and, in my informal evaluations, usually degrades the signal.)

With the price of full-frame camera bodies dropping to low $2K ranges, and some say most likely much lower in the foreseeable future (5-10 years)

I've said before online that the idea that full-frame cameras will soon become far more affordable is a pipe dream based on a misunderstanding (or ignorance) of Moore's Law. In fact, if you look at the predictions I made over two years ago for Canon cameras, you'll see that, for the most part, both prices and technology have generally lagged even my pragmatic predictions.

Let me explain. We've been conditioned to think that all electronics inevitably and perpetually get cheaper, as if they're some kind of mortgage-backed security. While this is often true, it is for a very specific reason: every time the semiconductor manufacturers shrink their "process" (that is the size of the smallest circuit that can be etched) they can fit more circuitry into a given space. This gives them the option of either shrinking the component or adding more features or computing power in the same space. Shrinking the component is the major mechanism by which digital products become cheaper: the smaller the chip, the more can be made out of a given batch of raw materials, and thus the more that can be made for the same input costs. The Intel Atom is a good example of this: this is basically a modernized Pentium M chip, and thus is about half a decade behind the state of the art in CPU power, but because it is built on a modern process, it is a tiny fraction of the size, and thus less than a quarter of the cost of the original Pentium M. By comparison, a high-end CPU for home use, say the latest Core 2 Quad, is neither significantly smaller, nor significantly less expensive than that original Pentium M from 2003.

Now I'm not saying shrinking chips is the only mechanism for reducing cost in electronics manufacturing. After all, a current Core 2 Quad is the same size as but roughly half the price of a Pentium M Dothan Centrino package. At least some of this is because Intel has significantly lower margins these days than in the Pentium M days, while much of the rest I attribute to the integration of the other Centrino components into the chipset, both of which are non-repeatable methods of cost savings. But even if we take a 50% price drop at face value, that is far short of the drop that would be expected from Moore's Law (50% every 18 months). So, again, die shrinking is not the only means of cost reduction, but it is the mechanism by which electronics continually and regularly get cheaper and faster.

So what does all of this have to do with full-frame cameras? Well, a full-frame DSLR is basically the same as an APS-C DSLR in the same feature class (think Nikon D300 vs. Nikon D700), except it has a larger (full-frame) sensor, and a few upsized components to support that larger sensor. And as Canon has stated over and over again, the dominant cost of making a full-frame DSLR is the sensor itself. That sensor is a silicon chip, the same as a CPU, RAM, or flash memory, and made in the same way. The difference is a full-frame sensor dwarfs the largest chip found in your computer: a typical Core 2 Duo (costing around $100-$200) has a die size of 82 mm2, while the ginormous GeForce GTX 280 (cards costing around $650) is around 400 mm2. A full-frame sensor is 24x36 = 864 mm2. I've estimated before that a full-frame sensor costs close to $1000 for Canon to make. Even if that were cut in half and suicidal competition set in, that would preclude a sub-$1000 full-frame camera. It took Intel (the best semiconductor manufacturer in the world) five years to achieve less than a 50% cost reduction in comparably-sized CPUs - what are the chances Canon or Sony will do better? And who's to say runaway inflation (caused by printing trillions of US Dollars for bailouts) won't interfere before then?

do you think many people with high end APS-C bodies/lenses will consider switching back to full-frame equipments?

Well, first, it's important to point out that the vast majority of DSLR owners never owned a full-frame camera, film or digital. That means they have no vested interest in full-frame.

Other than that, I can only speculate that many will switch to full-frame, whether or not that is a rational thing to do. My opinion is that, for the vast majority of amateurs out there, full-frame cameras offer a poor value proposition with respect to APS-C equivalents, if for no other reason that the vast majority of amateurs (including myself) don't realize the full potential of an APS-C camera, let alone a more technically-demanding full-frame one. For those This value proposition has little to do with the cost of the camera itself (a Canon 5D is not much more expensive than the fast-selling Nikon D300), and mostly to do with the cost of the lenses. 20+-megapixel full-frame sensors put enormous demands on lens sharpness, and I don't think one in twenty lenses can take full advantage of these sensors. Image quality is only as good as the weakest link in the image capture chain, and with these sensors, that link is almost always the lens.

Not that I'm saying there is no image quality advantage to full-frame versus APS-C. But it's a lot more subtle than many imagine, and there are disadvantages as well, including increased vignetting and poor corner sharpness. Not to mention the added bulk and weight.

Nor am I saying full-frame lenses necessarily cost more than their nearest APS-C equivalents (although in Canonland they do tend to be somewhat more expensive). But the main reason most amateurs upgrade to full frame is to get improved image quality. Upon doing so, many are shocked by just how much the next step up in image quality costs. Here are some typical examples (the prices are old, but illustrate the general trend):

Canon EF-S 10-22mm f/3.5-4.5: $700
Canon EF 16-35mm f/2.8L: $1400

While the EF lens does have a significantly wider aperture, I wouldn't say it has significantly higher image quality. It is much heavier.

Canon EF 70-300mm f/4-5.6 IS: $550
Canon EF 100-400mm f/4.5-5.6 IS: $1300

Again, I don't think the 100-400 has much, if any, higher image quality than the 70-300. It is twice the weight and well over twice the price, with a smaller zoom range.

Sigma 50-150mm f/2.8: $650
Canon EF 70-200mm f/2.8L: $1100

Canon doesn't make an APS-C short tele zoom (IMO, it's the one big gap in their APS-C lineup), but the nearest equivalent is half the price and weight of the EF lens, but still pretty damn sharp.

Personally, if I had spent upwards of $1000 on a new 12+-megapixel camera, I wouldn't consider anything less than these lenses to take advantage of it (and the latest cameras, both APS-C and full-frame, reveal shortcomings in all of these). As you can see, the lens cost quickly dwarfs the body cost, for both APS-C and full-frame. More generally, the body makes up a small fraction of the total system cost of your DSLR, and a full-frame body puts greater demands on every component in that system, from lenses to filters to tripod to batteries to camera bag to card storage to computer. I have personally spent upwards of $5K on my fairly-complete camera system, and have estimated a full-frame upgrade would run $8-10K, involving upgrades to all but $600 worth of prime lenses. And my equipment is not especially tied to APS-C lenses: I only have one. So if I decided to build a budget full-frame system, I could do so at a far lower incremental cost than someone just starting out. But that would leave me with a less functional, lower-image-quality system than my current one, so what would be the point in that?

In other words: do your homework before buying full-frame, or you will either end up with an inferior camera system, or out the cost of a small car.

But to get back to the original question:

What is your opinion on the future of APS-C vs. full-frame cameras?

To put it simply: if camera buyers behave rationally (but they almost never do) I think APS-C DSLRs are a dead end, but not because of full-frame. Or, rather, not because of just full-frame, but because it will get squeezed by full-frame on the high end, and Micro Four-Thirds on the low end.

I've been saying for some time now that DSLRs are dinosaurs. Or rather Frankenstein's Monsters. The mirror, dedicated autofocus, and big viewfinder are all legacy technology dating from half a century before the digital era. The reason modern DSLRs still need these components is that their replacements (electronic viewfinders and near-instant contrast-detect autofocus) are not yet good enough. Well, folks, the Panasonic G1 just fixed all that.

Not only is a Micro Four-Thirds camera potentially significantly lighter and more compact than any DSLR, it should be more intuitive to the vast majority of casual photographers. With fewer components, it should be cheaper and more reliable as well. Eventually, it should even be faster than any DSLR restrained by mechanical mirrors and shutters (Moore's Law does indeed lead to electronics getting faster).

Now I've been as skeptical about Four-Thirds as most people, but that has been on the basis of it being an entirely new system, backed by lightweight, half-hearted players, rather than on anything particularly wrong with the technology. The problem was just that with the inherent disadvantage of sporting the smallest DSLR sensor, it needed to provide compensating benefits of smaller size, weight, and cost. The cost savings never materialized, and the size and weight of the smallest Olympus DSLRs was never so much smaller than other DSLRs to make any difference in everyday use. Micro Four-Thirds changes that by removing the bulkiest components of the system, the viewfinder and the mirror. This is what many have been waiting years for: the digital equivalent of a rangefinder. (Not the actual digital Leica M8, which is a trilobite compared to the DSLR dinosaurs.)

Now none of this is specific to the Micro Four-Thirds standard in particular - I'm just using that term as a shorthand for a large-sensor, EVF, contrast-detect-only camera. (We need a name for these: "rangefinder" is factually incorrect; "point-and-shoot" misses the point entirely; and "all-digital" also comes up short since the lens is still analog - I'll just refer to the category as Micro Four-Thirds for now.) Panasonic still has an Olympus-Mons-sized hill to climb to break the Canikon stranglehold on the upper end of the digital camera market, and I'd say the odds are still against it. But this is a true paradigm shift, and the torch will eventually be taken up by other manufacturers once people see the wisdom in throwing off the shackles of 20th Century technology.

And that's where it might get confusing. Because if Sony decides to amortize its full-frame development costs by putting out a Micro-Four-Thirds-style full-frame camera, the APS-C vs. Full-frame debate could just shift to non-SLR cameras. And just when we put to rest the latest Sony-Panasonic HD Video war, we could be glimpsing the start of the next campaign.

But I think the odds are that battle will not be the major theme of the high-end camera marketplace of the next decade for several reasons:

• Canon and Nikon still have a near duopoly that shows no signs of abating, and they will fight tooth and nail to keep their current SLR mounts dominant.
• The Four-Thirds Consortium has not shown an ability to produce a fully-fleshed-out lens system in a timely fashion, regardless of body style, and I don't see any reason that would change.
• Full-frame Micro-Four-Thirds-style cameras don't make much sense from a marketing standpoint. The point of Micro Four-Thirds is to allow an easy-to-use, truly compact camera, and full frame defeats both those purposes. Sony's previous attempt at an APS-C-sized non-SLR camera, the R1, was larger than most APS-C DSLRs, and sold about as well as a Lehman Brothers debt offering.

Which is not to say Sony won't offer an APS-C competitor to Micro Four-Thirds. I see that as most likely. At which point consumers should ask themselves: what is the advantage of an SLR over non-SLR cameras? And then: how much will my SLR lenses be worth if most people move to non-SLR cameras?

So in summary:

• The value proposition of full-frame vs. APS-C is no better now than it was three years ago when the 5D was introduced. The gaps in body and lens prices are as large as ever. If anything, the proliferation of quality APS-C bodies and lenses have tilted the balance in favor of APS-C. (Nikon architectural/landscape photography may be an exception here, due to a singularly-stellar, ultrawide-angle, full-frame lens: the Nikkor 14-24mm f/2.8.)
• Value means very little in the digital camera marketplace. Just like big-screen LCD TVs didn't have to come down to the price of standard CRTs to engender mass adoption, neither do full-frame DSLRs need to come down to APS-C (or full-frame film SLR) prices to see the balance shift to consumers buying full-frame. The difference is consumers don't need to carry their hundred-pound Sony Bravia all day, but they do need to carry their two-pound Sony Alpha.
• It's all moot anyway, since Micro Four-Thirds is the beginning of the end of the SLR, at least for consumers. Any "investment" in SLR lenses is in danger.

But that too is moot if you replace your camera equipment every five years or less, as the vast majority of DSLR owners seem to do. If you're buying your camera today based on what the camera marketplace will look like in five years, I believe you're not seeing the forest for the trees. Non-collectors' lenses (i.e., lenses non-millionaires can afford) are poor investments: they lose quite a bit of value over time. They certainly lose value when a better replacement comes along (which happens regularly in Canon land), or if you drop them or they break through wear and tear. (Ask yourself if you actually engage in photography on a regular basis or mainly just want the latest gadget? I know most people want to think they are serious photographers, but almost everyone I know outside of Flickr don't take more than a couple hundred frames a year.)

Personally, I will be thinking twice about buying any SLR lenses from now on, APS-C or full-frame. But I have the luxury of putting off any lens-buying decisions because I already have a nearly full array of lenses from 10mm to 300mm. Finally I can honestly say I'm not seriously tempted by any of the latest DSLR equipment. Impressive pieces of technology no doubt. But what shots can they take that my current camera can't? Precious few I have any interest in taking.