Interesting Techniques To Achieve Photorealism

[Nameless]

In previous threads we discussed the time it would take for consoles to produce real-time Hollywood quality CG images. I viewed an interesting presentation provided by Kavita Bala from Cornell University regarding the subject and some new programming approaches to tackle this problem.
(Grab a Mountain Dew or Coffee!)
The presentation is about 50 mins so I will list some highlights:
Source:
http://www.acm.uiuc.edu/conference/archive/2005/video/bala.wmv
When will developers be able to generate in real-time an animated film such as 1998’s A Bug’s life by Pixar? The short answer is decades, generating an average frame of A Bug’s Life takes three to four hours on a 330 MHz Ultra SPARC, and requires about a gigabyte of geometric data alone, as of 1998. To reach a minimum real-time rate of 12 frames per second, this means a speed-up of about 150,000 times is needed. Moore’s Law gives an acceleration rate of 10 times every 5 years; this puts real-time A Bug’s life at around the year 2024”. (Computing power needed for real-time rendering)
From the book “Real-Time Rendering” by T. Moeller and E. Haines

Also, memory access does not follow Moore’s law and model complexity is growing exponentially...

Toy Story 2 is twice as complex as A Bug’s life which is 10 times more complex than Toy Story, Ed Catmull, Pixar’s founder and CTO.
Toy Story 2: 2-13 hours render time/frame
Bug’s Life: 3-4 hours render time/frame
Toy Story 2: 10 min-3days render time/frame
Monster’s Inc: Render time more than Toy Story, Toy Story 2 & Bug’s life combined!

To address these concerns Bala suggested interesting techniques:

Global illumination techniques
Texturing techniques

The illumination technique was basically taking the idea of GI (global illumination) and mixing in 'smart' sparse sampling. Lights are grouped/clustered and trees of lights are made and using pre-defined % error limits each pixel only uses the number of lights necessary to get an accurate illumination value within the given 2% error. So instead of tens of thousands of lights calculated per pixel, only dozens are required... The technique was a unified technique that worked for hard and soft shadows, HDR, direct and indirect illumination... and very high quality AA.(all requiring little computational resources) Also, using the technique of creating cluster tree of lights and partitioning this cluster substantially cuts down on the required memory accesses per pixel.(Reduces the amount of system RAM needed)

The texturing technique FBT ("featured based texturing") in simple terms a technique to interpolate detail in textures eliminating the issues with resolution when you zoom in on textured objects. (Common in videogames, texture mapping)

Hopefully we will see additional advancement in the areas of texturing, lighting, animation and frame rates due to programming advancements and not simply hardware advances (Moore’s law). If developers work smarter we could see huge advances in visuals in a few years, but if we rely on Moore's law it may be a few decades...

Feel free to share your thoughts. Peace :angelgrin:

[GTShotoKen]

I think that most devs are already in the mind set to improve their rendering techniques instead of waiting for new hardware.

The ideas already in use, such as normal mapping and parallax mapping, are a god send to gamers. The efficiency of shaders are only going to increase over time and are always getting better.

Fantastic post by the way Nameless!

+rep

[Arnaud_M]

I think that we will have a real time game with similar graphics quality to "A Bug's Life" far, far before year 2024 :-) Why ? Because for the graphics to be similar in quality, you don't necessarily need them to be produced in the same way. Pure CPU power was used for "A Bug's Life", but today's computers are producing very high quality graphics with GPUs, not CPUs. Raytracing quality, or something very similar, could be approximated with advanced pixel shaders, optimized (or even partially precomputed) global illumination model, and some CPU power for the rendering of specifics aspects (such as refraction, as demoed by IBM recently, in realtime with CELL). Look at the "Heavy Rain" demo, we are not that far away from 1998 high end CG. I do not claim that we can approximate with 100% fidelity the look of raytracing and photon mapping (the most common global illumination method) with advanced pixel shaders and some precomputed stuff, but that soon enough we will be able to produce games that will look so close to that, that only the most purists among the purists would complain about the difference. 2024 is far too pessimistic I think. I would say 2012, the next gen of console.

Arnaud

[Nameless]

I think that we will have a real time game with similar graphics quality to "A Bug's Life" far, far before year 2024 :-) Why ? Because for the graphics to be similar in quality, you don't necessarily need them to be produced in the same way. Pure CPU power was used for "A Bug's Life", but today's computers are producing very high quality graphics with GPUs, not CPUs. Raytracing quality, or something very similar, could be approximated with advanced pixel shaders, optimized (or even partially precomputed) global illumination model, and some CPU power for the rendering of specifics aspects (such as refraction, as demoed by IBM recently, in realtime with CELL). Look at the "Heavy Rain" demo, we are not that far away from 1998 high end CG. I do not claim that we can approximate with 100% fidelity the look of raytracing and photon mapping (the most common global illumination method) with advanced pixel shaders and some precomputed stuff, but that soon enough we will be able to produce games that will look so close to that, that only the most purists among the purists would complain about the difference. 2024 is far too pessimistic I think. I would say 2012, the next gen of console.

Arnaud
Arnaud, I'm not sure if you read the entire post, but let me clarify my position.
I started with the 2024 date to show that hardware is not the best approach to achieve high-end CG visuals. If developers implement several programming techniques the results could be achieved in a fraction of the time... Personally lighting is the most important aspect if you want to represent realistic visuals. (Global Illumination in real-time is the Holy Grail):cheers:

[GTShotoKen]

Arnaud, I'm not sure if you read the entire post, but let me clarify my position.
I started with the 2024 date to show that hardware is not the best approach to achieve high-end CG visuals. If developers implement several programming techniques the results could be achieved in a fraction of the time... Personally lighting is the most important aspect if you want to represent realistic visuals. (Global Illumination in real-time is the Holy Grail):cheers:

Will the mythical ray-tracing gpu that is in development at Nvidia solve this problem?

[PhYmon]

Lighting such as Crysis shows.. have u guys saw the comparation between the real life and Crysis scenarios photos?? I have to tell u, those are hot!! and u can barely tell the differences between the real and the game one..

[Nameless]

Will the mythical ray-tracing gpu that is in development at Nvidia solve this problem?
Guru, perhaps it would be a step in the right direction.
I know ray-tracing is an integral part of Global Illumination techniques...

[PhYmon]

Can I ask what the ray tracing do? I mean I dont get the whole concept of it!!

[makeitlookreal]

Thanks for the great link Nameless!

This is what I am talking about. By working smarter and not harder we could get high quality graphics in far less time than waiting for super advanced hardware. Bumb mapping is just one example of how you can do much more with less. I think there will be many, many techniques used by the PS3 to make very realistic looking graphics that are beyond the Uncanny Valley.

[Arnaud_M]

Arnaud, I'm not sure if you read the entire post, but let me clarify my position.
I started with the 2024 date to show that hardware is not the best approach to achieve high-end CG visuals. If developers implement several programming techniques the results could be achieved in a fraction of the time... Personally lighting is the most important aspect if you want to represent realistic visuals. (Global Illumination in real-time is the Holy Grail):cheers:

I confess, I only read your summary of the article ;-) I agree with you that lighting is a very important aspect. It is also probably the most difficult, because the observable lighting in a scene is not a direct results of objects properties, but a result of objects interactions, which are order of magnitude more complex to compute because rays of lights needed to be traced through numerous bounces, and in a lot of spatial directions for reflexion, refraction, color bleeding, etc. I don't think we will see such real-time, physically correct (or at least, as far as photon mapping currently is) global illumination before a long time (and I am not sure if even 2024 is not too *optimistic* for true GI). But, fake GI, yes I agree, much sooner than that.

Arnaud

[Nameless]

Lighting such as Crysis shows.. have u guys saw the comparation between the real life and Crysis scenarios photos?? I have to tell u, those are hot!! and u can barely tell the differences between the real and the game one..
Phymon, The lighting is going to play a huge factor regarding realism.
Remember the pics of GT4 comparing racing track pics to real life pics.
The photos were compelling, but we all know the results in action were not photorealistic.
Crysis is a step in the right direction, but trust me seeing the game in motion will not fool your eyes into perceiving real environments and a lot of that is driven by the lighting techniques used, HDR is just the tip of the iceberg. ;)

[Arnaud_M]

Can I ask what the ray tracing do? I mean I dont get the whole concept of it!!

Ray tracing is, in essence, a very simple algorithm: you "trace back" the rays of light from the camera taking the picture, to the light sources present in your scene. Instead of following the rays of lights from the light to the camera (which would be a complete waste of time, as many rays of lights will never reach the camera !), you trace back from the camera to perform computations only for the rays of light you will see in the final image.

So for each pixel of the screen, you trace a line "into" the picture, you look for the object that is visible in this direction, and then you find the lights that are lighting this pixel of this object, either directly or indirectly (through one or more bounce on nearby objects). And then you do the same for the next pixel. And the next, until the image is computed.

Arnaud

[Nameless]

Can I ask what the ray tracing do? I mean I dont get the whole concept of it!!
In simple terms Ray tracing is a technique for rendering 3D graphics with very complex light interactions.(creating realistic global lighting)
Basically you model reflection and refraction by following the path of light bouncing through an enviornment.
(The result is a realistic simulation of light)

I hope that provides some clarity...

[PhYmon]

Thanks Arnaud and Nameless, but in game how is going to be used? I mean I already saw its potencial of scanning a landscape but what other uses can take advantage of it?

[GTShotoKen]

Thanks Arnaud and Nameless, but in game how is going to be used? I mean I already saw its potencial of scanning a landscape but what other uses can take advantage of it?

Isn't that inherent in the thread topic.."Photorealism".

All this is just to provide a more realistic image. I can't imagine any gameplay aspects from GI unless Boktai gets really freakin complex.

[Nameless]

Thanks Arnaud and Nameless, but in game how is going to be used? I mean I already saw its potencial of scanning a landscape but what other uses can take advantage of it?
In games this technique would provide realistic shadows, reflections & indirect illumination. The result would be far more realistic lighting than today's gaming engines.

Also, ray tracing could be used for physics, acoustics & collision detection.
Think of a radar system that uses light rays to determine the distance to nearby objects. This could potentially prevent clipping in objects and prevent artifacts in rendered objects.

In a nutshell it's one step closer to realism in gaming... ;)

[GTShotoKen]

Also, ray tracing could be used for physics, acoustics & collision detection.
Think of a radar system that uses light rays to determine the distance to nearby objects. This could potentially prevent clipping in objects and prevent artifacts in rendered objects.

In a nutshell it's one step closer to realism in gaming... ;)

Ray-tracing can be used for more than simulating light?

So your saying it could simulate sound waves too or am I missing something?

I understand your logic when it comes to using ray-tracing for physics, but wouldn't that be even more time consuming (a hell-of-a lot more accurate though)?

[cpiasminc]

I was there for the talk myself back in October. When it gets near the end and you hear her answer a question about soft shadows, that was me who asked.

Probably the most valuable piece of work there for gaming, at least, would be the FBT work, but the vector annotation business is a bit unwieldy when art teams have to produce several thousand textures, some of which may benefit from FBT and some of which won't. One of my current little side projects in which I'm referencing the good doctor is the idea of weighting FBT sampling in an image-based approach (which I think is a little more practical for GPU-based implementations) that simply would involve an edge-detection pass... yeah, it's not quite as resolution-independent, but it's a little more general in nature and less labor-intensive.

There are certainly problems in games where the whole "Lightcuts" theory could work nicely, but not quite for the same reasons. In any case, the complexity of the minimization metric is a bit much. There are probably some differential schemes you could use that may be a lot simpler, but a fast and still generic approximation is really, really hard to devise.

[Saibo]

ray tracing is doable on the PS3, but not global illumination. GI will give you photorealism you so desire.

ray tracing(which is not to be confused with GI(indirect illumination):

http://www.cs.unc.edu/~rademach/xroads-RT/RTarticle.html

[makeitlookreal]

By the way, I just watched the video at that link and I think her techniques will be vital to the PS3 and all next-gen games.

[Nameless]

Ray-tracing can be used for more than simulating light?

So your saying it could simulate sound waves too or am I missing something?

I understand your logic when it comes to using ray-tracing for physics, but wouldn't that be even more time consuming (a hell-of-a lot more accurate though)?
I'm not saying it could simulate sound waves, but it could simulate the behavior of an acoustic space...

If you would like I can expand on this concept...

[cpiasminc]

I understand your logic when it comes to using ray-tracing for physics, but wouldn't that be even more time consuming (a hell-of-a lot more accurate though)?
Actually, it's pretty common. The difference is that we're not raytracing a million rays pixel for pixel through an image. For instance, when you check if a bot has a line-of-sight to see you, it's a raycast, but we're only tracing one ray.

It's also pretty common in a physics engine to do volumetric sweeps along a ray or line segment. This is relatively easy with convex hulls, spheres, boxes, and a few other shapes. With some, though, it can really be a pain. A volumetric sweep of a sphere is really not that different from a raycast -- a raycast is simply a sphere sweep for a sphere of zero radius ;). These sorts of volume sweeps are part of what is referred to as "continuous collision detection", which is really an awful term for it. If you're interested, you might want to take a look at some stuff on the "Bullet" physics engine.

To give you an idea, at my last job, we put out one title where we individually simulated every bullet fired as a moving object and did a per-frame sphere sweep collision test for each of them. And that worked just fine at 30 Hz on a PS2.

[Nameless]

Ray tracing has the potential to truely effect game development during the next-gen... Time will tell!

[GTShotoKen]

Thanks for the incite guys.

How well do you guys think the Cell is adept at operations like these?

-Edit- Hey Nameless, your going to be a future game developer. What kind of game would you personally like to develop? [/offtopic]

[Z]

shaders is another key point of game development. as nVidia said, this gen was about textures and polygones. the next generation will be about shaders.

I think the most evident features of next-gen gaming will be:
lighting
shaders
physics

[Nameless]

Thanks for the incite guys.

How well do you guys think the Cell is adept at operations like these?

-Edit- Hey Nameless, your going to be a future game developer. What kind of game would you personally like to develop? [/offtopic]
A suspense thriller, if I had to compare to a current game RE4 (Omen movie) would be a fair comparison. (no monsters, but supernatural/religious aspects)I would like for the game to be more focused around the story line, character interaction, visuals & cinematography.
I would like to raise the bar from a visual perspective and not just for the sake of eye candy, but to tell a compelling story with true emotion...(Kojima is a great inspiration)

[Nameless]

Thanks for the incite guys.

How well do you guys think the Cell is adept at operations like these?

-Edit- Hey Nameless, your going to be a future game developer. What kind of game would you personally like to develop? [/offtopic]

I almost forgot about the first question here's my thoughts:
The Cell/RSX architecture is adept for some ray tracing concepts, but it will take time before we see anything of great significance... We will more than like see Ray Casting implemented initially, because Ray Casting is a simple version of ray tracing.
Ray tracing takes a ray bounces it off of objects going through transparent objects as necessary, and splitting as necessary. Ray casting only takes the very first step of that, taking the color of the first object the ray hits after going through the virtual pixel. Developers will take baby steps and honestly I'm not sure if we will see full blown real-time Ray tracing on the PS3 or 360...

CPI your thoughts?

[CARTIER90]

you have to remember as well people that cinema resolution is 4 * 2 K......4 times 1080 P resolution.........

[GTShotoKen]

A suspense thriller, if I had to compare to a current game RE4 (Omen movie) would be a fair comparison. (no monsters, but supernatural/religious aspects)I would like for the game to be more focused around the story line, character interaction, visuals & cinematography.
I would like to raise the bar from a visual perspective and not just for the sake of eye candy, but to tell a compelling story with true emotion...(Kojima is a great inspiration)

You had me at suspense thriller! :lol:

By the way, have you ever tried entering the mod community to test out gameplay and content possibilities? Modding is a great way to get your ideas out into the community and see how they respond to it.

Well anyways, I appreciate you for answering my questions. I guess we will just have to wait and see how things play out this generation.

Wait a minute...hey cpiasminc, have you experimented with raytracing on any of the next-gen console hardware? Maybe the lighting revolution could start with you?

[cpiasminc]

The Cell/RSX architecture is adept for some ray tracing concepts, but it will take time before we see anything of great significance... We will more than like see Ray Casting implemented initially, because Ray Casting is a simple version of ray tracing.
Ray tracing takes a ray bounces it off of objects going through transparent objects as necessary, and splitting as necessary. Ray casting only takes the very first step of that, taking the color of the first object the ray hits after going through the virtual pixel. Developers will take baby steps and honestly I'm not sure if we will see full blown real-time Ray tracing on the PS3 or 360...

CPI your thoughts?
Yeah, but to get costs down, I think people will subsample further than that. Below the pixel level down to the "object element" (objel?) level. For instance, you take the whole talk of Warhawk's raycasted clouds. I doubt they'd raycast by the pixel, and instead raycast by the particle, and simply used particles of sufficiently low alpha range (and NO color variance) that it works. So instead of casting rays through say, 100,000 pixels (which is certainly possible given you've got a million on screen) for one cloud, you cast rays to about 250-300 particles.

Full blown raytracing in realtime at HD resolutions... IN A GAME... I don't think so. As powerful as CELL is, it would need the support of a memory subsystem hundreds of times faster than XDR on all aspects (latency, random access, streaming, total bandwidth, number of memory ports). And quite frankly, even 200 GFLOPS isn't going to cut it... 1 TFLOPS, maybe, but when we get out there, everybody is going to be thinking global illumination, and GI would easily demands hundreds of TFLOPS.

Wait a minute...hey cpiasminc, have you experimented with raytracing on any of the next-gen console hardware? Maybe the lighting revolution could start with you?
Ha ha... no, not yet -- even though I am a raytracing nut, I have to keep things in perspective of what's feasible. And trying to introduce raytracing into a shared development base which is so huge that it would span multiple Bluray discs is not exactly a quick job. Though I am working on something vaguely related, but that has less to do with doing anything fancy and more to do with making the art pipeline more flexible in regards to lighting and shading (Eat your heart out, UE3)...

[Saibo]

cpi

what about ray tracing for a simple game like Loco Roco, but in 3D, would that be feasible?

[Nameless]

cpi

what about ray tracing for a simple game like Loco Roco, but in 3D, would that be feasible?
You may want to clarify your question.
Are you refering to using Ray tracing for visuals (lighting) or gameplay (i.e. physics & collision detection)?

Based on the visual style of the game, creating complex lighting would not be one of the priorities of the developers. Also, isn't this game in development for the PSP?

The discussion of Ray casting was mentioned regarding photorealistic visuals it would not be applicable in all games and all genres... That's my 2 cents.

[cpiasminc]

The biggest problem with raytracing/raycasting in software in game is really the cost scaling with respect to certain factors.

Certainly one of the things that makes a game like Loco Roco easier on a software renderer of any kind is the monochromatic nature of the objects (no texture lookups lifts a huge memory access burden).

However, in a naive approach, raytracing/raycasting is going to scale linearly with each of the variables affecting complexity -- which is to say -- resolution affects it linearly (each pixel is a new ray). Number of objects is going to affect it linearly (in a naive approach, you'll blindly check every ray against every object). And lighting complexity will affect it linearly (you check shadow rays against every light).

Now in a more optimized approach, you can be logarithmic against objects and lights, but not against resolution. The minimum number of samples you make is the minimum number of samples you make.

In essence, the more you want to get out of a raytracer, the more it's going to cost you. What's funny about this is that every step forward with GPU programmability is simply inching us towards a rendering benchmark which is anyway full of things inherently possible through raytracing.

[GTShotoKen]

Ha ha... no, not yet -- even though I am a raytracing nut, I have to keep things in perspective of what's feasible. And trying to introduce raytracing into a shared development base which is so huge that it would span multiple Bluray discs is not exactly a quick job. Though I am working on something vaguely related, but that has less to do with doing anything fancy and more to do with making the art pipeline more flexible in regards to lighting and shading (Eat your heart out, UE3)...

Props to you cpiasminc. I would like to see someone give epic's engine a run for its money.

I know every major game company in the world is dying to find someone way to lower cost down in the art development pipeline.

I just have to ask cpiasminc...have you you ever touched the PC space when it comes to games? I wouldn't mind playing an FPS with small credits saying, "Developed by Cpiasminc."

I say to myself sometimes,"Wouldn't it be cool to play a game with lots of guns made by cpiasminc?"

[cpiasminc]

Props to you cpiasminc. I would like to see someone give epic's engine a run for its money.

I know every major game company in the world is dying to find someone way to lower cost down in the art development pipeline.
Hmmm... I can't say for sure that what I'm working on will really lower costs, as it does add one potential layer of tasks to the art staff. Though we've kind of partitioned our artists into "technical" and "non-technical" artists, the technical ones actually being the ones who will use my part of the job.

The main point is more about giving them more power and control to do some fancy-schmancy things with material properties. And the big thing that makes it theoretically more powerful than anything UE3 has is the fact that they get control over everything (except shadows -- which they control elsewhere).

I just have to ask cpiasminc...have you you ever touched the PC space when it comes to games? I wouldn't mind playing an FPS with small credits saying, "Developed my Cpiasminc."

I say to myself sometimes,"Wouldn't it be cool to play a game with lots of guns made by cpiasminc?"
Sadly, the best I can tell you is that my time in the PC space was really while working at "The Creature"'s company. Getting time to actually code was hard during the major part of the workday which is why I was willingly putting in long hours back then (most everybody else would be gone and I could actually get something done). Being the guy managing the programmers, doing about half of the IT work and server administration, doing more than half of the effects work, localization, and writing all the music, and doing all the sound effects.... yyyyeah. Not so much fun.

Convenient that you mention guns, since the weapons system in that game was basically my baby all the way through, and I scrapped and rewrote it from scratch about 4 times. And it was general enough that I not only got first person shooter gameplay working, but third-person melee and interchangeable vehicle weapons all working... then the creature got sued by an old lady over a defectively constructed wheelchair ramp -- I'm lost as to how you can mess up a wheelchair ramp, but leave it to that moron to do it.

I don't have to tell you where that went. It hasn't amounted to nothing. My salary nowadays is about industry standard for someone of my experience -- which is sadly enough, nearly triple what I was making back then.

Well if you ever dig through any old Anime-Keep fansubs of Chobits or Samurai Deeper Kyo or Kousetsu Hyaku Monogatari or something similar and see the credits read "Translation : CPIAsminc", that's me.

One of my earlier jobs back pre-gaming involved a standalone Java database which was meant for a local medical clinic. They could run their little server over an intranet and access it from any machine via IE. I effectively never got paid for that one because my employer shot himself. So yeah... that's how fun my life is. Interestingly enough, I haven't even laid eyes on Java since then.

[Phryne Astynome]

I don't have to tell you where that went. It hasn't amounted to nothing. My salary nowadays is about industry standard for someone of my experience -- which is sadly enough, nearly triple what I was making back then.

Well if you ever dig through any old Anime-Keep fansubs of Chobits or Samurai Deeper Kyo or Kousetsu Hyaku Monogatari or something similar and see the credits read "Translation : CPIAsminc", that's me.

One of my earlier jobs back pre-gaming involved a standalone Java database which was meant for a local medical clinic. They could run their little server over an intranet and access it from any machine via IE. I effectively never got paid for that one because my employer shot himself. So yeah... that's how fun my life is. Interestingly enough, I haven't even laid eyes on Java since then.

Good Lord! Your salary was THAT low?! There are UIUC (from the CS dept.) grads working at Lazard and Goldman Sachs making over 110K straight out of college. God damn, that is insane that you worked at such a shithole considering the college you graduated from.

[Saibo]

Good Lord! Your salary was THAT low?! There are UIUC (from the CS dept.) grads working at Lazard and Goldman Sachs making over 110K straight out of college. God damn, that is insane that you worked at such a shithole considering the college you graduated from.


I think it has to do with "experience". the more experience you have the better you get paid, also its somewhat of a catch 22. Its even worse for 3D modelers.

[cpiasminc]

Game programmers are anything but the highest paid programmers in the software industry. I wouldn't go so far as to say they're the lowest paid, as the absolute lowest paid people are in general those working in hellholes, which aren't really exclusive to any sub-field of software development.

But then a hellhole game developer will probably still be making less money than a hellhole scientific computing fellow. Of course, when you get into otherwise successful independent software developers, typically they pay okay... but independent game developers -- they not only pay crap, but the flow of income is often unstable and you can't always count on your next paycheck not bouncing (yep, I've been there).

Even companies that are known to otherwise pay reasonably well, don't pay that highly for their game development teams if they have any. I think about the only exception is Sony, and even that is not *all* the time. I know of a few places that deal with contract programmers and such who make reasonable sounding offers at first sound, but they're still short-term contracts with no benefits and no withholdings, so it works out about the same in reality.

I think it has to do with "experience". the more experience you have the better you get paid, also its somewhat of a catch 22. Its even worse for 3D modelers.
Well, the upward mobility of salary is there, but the catch-22 is that it is rarely possible for you to achieve that while also staying at one studio. You typically only see that mobility by gaining experience and then changing jobs. The catch of course is that it looks bad on your resume if you never stick at a job for very long. At least for programmers. I'm told by several of my former artist roommates that the same isn't always true for them.

If I got into programming for the money, I wouldn't be in game development anyway. I would have done a BS in CS and an MBA and gotten into middle management somewhere and not cared that my job lacked the intrigue of linear algebra or fell short on the elegance of computational geometry. Who needs intellectual stimulation when you've got the frustration of having to deal with hundreds of people?