DO GRAPHICS CARDS matter anymore? The short answer is yes, but not for long. The problem is progress, and GPU makers are going to work themselves out of a job.

Progress can be a bitch. And GPU vendors are facing down two really daunting trends, monitors and eyeballs. Both have limits: monitors are somewhat flexible, eyeballs are absolute. Progress matters until you hit a wall, then you have a hard sell, a very hard sell. Extra undercoating for your new car sir?

The problem is pretty simple, and for the sake of lessening arguments, let us define a high-end card as $500 with bleeding edge performance and nothing out there faster. Mid-range parts are defined as $250, and about half as fast as the high-end parts.

Just to lessen the arguments some more, let us define the GPU development cycle as a year long cycle with the first part being a new architecture and the second, six months later, being an update or refresh. Each new generation doubles the performance of the last, more or less.

Any change in the raw numbers, say the refresh cycle going from 12 to 18 months, will change the times and numbers, but not the conclusions. The end result is inevitable,

The fundamental limit is that your eye can only resolve so much information. We have long passed the point where a GPU can crunch more bit depth than a human can parse, but have rarely used that because the displays to do so are expensive and the payback for it is small.

If a modern GPU works in 32-bit depth most of the time, with some capable of 64, 128-bit, aka extreme overkill is a factor of four GPU power away. To put that in doubling periods, it is two generations away to get from 32b with current frame rates to 128b with the same frame rates. Based on the numbers above, that would be a year if GPU makers wanted to go there and end that argument once and for all.

How about resolution? Well, we are already there. A two generation old ATI X1300 low end card can drive two 30" monitors, and the upcoming HD34xx parts can do the same with all the DRM infections you might care to inflict on yourself. They will even support Display Port and the rest. All this for well under $100. Pushing rez is a dead issue, unless you are buying Nvidia cards, it simply is not a problem anymore.

So, what is left? Is there a grand challenge? Sure, it is frame rates, polygon pushing and shaders, also known as the main things that GPUs are meant to do now. This is enough to keep us busy for a while, right? Not really, we are fast approaching adequacy now with mid-range cards on a mid-range system. That means high end everything is within shooting range.

Think of it like this, if an ATI HD3870 can play most games adequately at 1920*1200, AKA 1080p, right now, that is fine. Not many people have 20+ inch monitors that can display those rezes, according to the steam survey the here, only 2.28% have 1920 capable monitors. The 'other' category that includes 30" 2560 * 1600 beasts as well as a few oddballs like the 1366 * 768 of the laptop I am on now has a mere 1.36%.

Lets assume that each level of AA and AF each takes double the GPU power of the predecessor. 2x takes twice the power of the normal, and 16x takes twice as much as 8x. If a HD3870 can run a modern game at 1920 * 1200 with 0x AA and 0x AF, then a 1920 * 1200 16x/8x will take 2^4x and 2^3x as much power.

2560 * 1600 has about 1.75 as many pixels as 1920 * 1200, lets round it to 2x. Just to give it the most pessimistic outlook possible, lets say a future game will need 4x the horsepower for miscellaneous things as current GPUs. Between the rez and the 'other' category, that is 2^3 times more power needed.

In total, we have the need for 2^10 more power, or about 1024x the power of today's HD3870, a crazy insurmountable number right? Even if this would get us to the holy grail of GPUdom, perfect frame rates with everything turned on, it can't happen. 1000x is more than we can expect, right?

In powers of two, that is 10 doubling periods, the first one will happen in January. The next nine will be spaced about 6 months apart meaning this will be 4.5 years down the line for the worst case scenario possible. At that point, GPUs will be good enough for the highest end monitors out there today.

But wait you say, what if there is an uber-rez monitor in those intervening years? Well, if it is not out now, it won't be reasonably priced then, and probably relegated to the top 5% of the market or less. If not, a 5120 * 3200 monitor would only add another year to the curve. The flip side of that is that a human eye would be hard pressed to see those many lines, and anything more would probably go to waste.

With the upcoming R680 cards in January, the curve will probably be ahead of where I say it is, and those cards will most likely run a 2560 * 1600 game with low AA and AF settings standard. It is also unlikely that any massive jump in rez for mainstream monitors will come out of the blue.

So, the problem facing GPUs is that in a few years, they will catch up to monitors, and a few years after that, they will be good enough for eyes. Right now, a $250 card is good enough for, at least according to Valve, the vast majority of monitors out there. Glass is the choke point, and it is not scaling very fast.

GPUs on the other hand show no sign of slowing down. Between ATI and NV, Intel in about 18 months as well, you have a cutthroat market where anyone slowing down looses. This means that for all involved, it is full throttle until they go flying off a cliff.

Then it is over. GPUs will no longer matter. Time for a new challenge. Right now, they are hanging on by a fingernail hitting the gas. Remember what happened to the thriving sound card industry? µ