Email info at x-plane dot com for more info!

Also, have you considered porting your program to IRIX? SGI's Onyx2/3xxx systems seem like just the type of hardware X-Plane would run on. Plus it's a great excuse to own an SGI.

—

The Criterion tech team are quick to dismiss the notion that the current generation, or indeed any generation of hardware, has been pushed to its absolute limits.

"It's nothing new, you hear it all the time… we've maxed out, we're the best," says Fry dismissively. "That's proof that you're not the best," Parr chimes in. "It means you're out of ideas."

"X-Plane Lite" is a great idea. Austin obviously has no problem calling XP a game* and selling it to iPhone users. XP-Lite would retain (even expand) one market, while "X-Plane Pro" would satisfy the users who have been commenting on your last few posts.

And there's no reason that the old ruse about being a small company should stop you. Simply take existing XP and remove any user-adjustable settings that would cripple performance on a low-end machine, and don't allow the addition of scenery packages or add-on aircraft. Pretend you're creating version for a giant iPhone.

Voila. Instant new product and market. And then you can get back to doing what you presumably like the most–making XP qualitatively better.

*Yes, I did catch the "complex numeric simulation" jest.

1. X-Plane is not a game – it is a highly complex numeric simulation! 😉

2. The flight simulator to game comparison is IMO a dubious one…see also the tons of threads where FSX users complained (idiotically) that FSX doesn't look like CRYSIS. Take a modern shooter and fly up to 40,000 feet and it's really not going to look as good as it does up close.

I'm not saying that X-Plane's graphic limits are only due to having to support a truly flexible viewpoint – but I am saying that that is enough of a difference in engine design to make it an Apples to Oranges comparison.

(The fact that X-Plane is a franchise, now on version 9, vs. games that might only have one version is another incomparable difference.)

"since most people out there are eyecandy whores im sure you would improve sales a lot more by supporting cutting edge gpu/cpu features compared to the few guys that cant buy a more modern grafikcard."

I'm not sure we can meaningfully discuss this. We would have to compare a feature we did code that wasn't eye candy vs. a feature we intentionally deferred that would have been eye candy, and then connect sales directly to those particular features (not easy to do). The cost-benefit analysis of features is rarely so clean-cut.

Two comments. First on the part we agree about (the low end falling behind): it's not the netbooks that concern me. For NetBooks we need something like X-Plane Lite.

My concern is when we get new Dell customers who have bought a brand new machine and have a Radeon X300. (This was happening a lot a year or two ago.)

The problem is that at the very, very low end of the tech curve, the companies stop cutting fat and start cutting muscle. To save that last $20 in OEM hardware cost, you lose all of your dedicated VRAM. Ouch!

If we told the user for $20 you could get 4x performance, would they take it? Heck yes! But…often the computers that have these constraints have problems that aren't easily swapped out like a graphics card (e.g. the graphics are mobo, not in a real PCIe slot). Call it the "stealth netbook-ization" of desktop hardware.

Back in 2000 or 2001 when I bought a Dell I got, for not much more than $500: the second fastest graphics slot and a decent P-IV (for the time). The graphics card was a bit on the low-end, but the gap wasn't as wide back then between gamer and commodity GPUs. All in all the machine actually flew X-Plane at the time decently, and a GPU replacement was an option.

I don't think this level of performance still exists in lower-budget hardware.

Now absolute vs. relative performance…that's just the problem: X-Plane is art asset and data driven in its graphics, so a lot of the times we don't have wide-range relative scalability. And that means that we are stuck in an "absolute budget" kind of world.

Recognizing that triangle count and any other single metric is a gross simplification of GPU or system performance…consider that we have two different "flavors" of feature:

– Forests. We could throw even more hardware at forests and probably make them loo even better. Since real forests are so detailed, and since forests are generated from rough data by the sim, we can have scalability. For forests, the relative performance is really what matters…we can always set "tree hugger" to eat 8x the resources of "sparse".

– Airplanes. Problem is: the airplane meshes are premade by the author. Even if we had a way to make an airplane use 8x the performance in certain modes (we do not really have this right now, unless you consider the virtual cockpit an optional feature) we'd still have to go back and remake the art asset to keep up with hardware.

When our lead artist works on a new airplane, he asks "what's my polygon budget". This number is absolute – it's just a number.

In fact, neither of these problems are great examples, because the numbers involved in the GPU are so huge. In truth the problem with the GPU is that the performance changes so much that the very question of _how_ we render changes…so it's not just a question of "is there a way to make the rendering of feature X 8x as expensive, and then keep doing that over and over?" It's almost a question of: do I render feature X using technique A or B which are wholly different.

The quantitative gap in hardware leads to a qualitative gap, which I don't think can be described as a multiplier of performance…it represents a fixed engineering cost.

Let's start with the easy part. If you assume Moore's law continues to hold, then the number of transistor's we can pack on a chip indeed will double every 18-24 months. So let's boldly over-simplify and say that performance / capability similarly doubles every couple of years.

That does NOT mean that your gap over a fixed number of years really widens. Yes, the difference as an ABSOLUTE number (whatever you measure, triangles per second, GFLOPS, I don't care) goes up – but that's meaningless. It's still 8 times between old and new. And as much as you care about absolute number of triangles for example, 8 times is 8 times is 8 times.

But the other observation that you made is much more interesting, I think. The fact that while the ratio of improvement on the high end stays about the same, the ratio of improvement on the low end doesn't keep track – instead the low end chips get smaller and smaller (in absolute size) and size is directly proportional to price in the chip business – so things get cheaper!

And that means that while in 1993 the original Pentium was introduced in 60 and 66MHz (that's 10% spread) and the high end (P66) and the low end (486sx25) were only about a factor of 3 apart, the performance difference between an Atom Z500 and a Core i7 975 Extreme Edition is quite a bit wider. One in-order core, no hyper threading, 512KB cache at 800MHz vs 4 out-of-order cores with hyper threading, 9MB cache at 3.33GHz. Oh, and the Beckton chips (Xeon EX, I guess) are about to be released which go to 8 cores and even more cache…

So yes, the range has widened and it will continue to widen. At some point you, the developers, will have to draw a line and have to set realistic minimum requirements. Anyone who thinks that a Netbook is the right platform for X-Plane is simply confused!

Yes, for economic reason you want to reach a large target market – but you don't want to sacrifice the capabilities on the high end in return – because those are the people who continue to drive what's possible, who will continue to pay for new versions of the software (and oh, most likely your high end, high value customers will be focused on the highest end hardware as well).