Real-Time, Movie-Quality CGI For Games

An anonymous reader writes "An Intel-owned development team can now render CGI-quality graphics in real time. 'Their video clips show artists pulling together 3D elements like a jigsaw puzzle (see for example this video starting at about 3:38), making movie-level CG look as easy as following a recipe.' They hope that the simplicity of 'Project Offset' could ultimately give them the edge in the race to produce real-time graphics engines for games."

Comments

Wow

[binarylarry]

They've discovered the hidden secrets to rendering Academy Award winning films such as "Gears of War" and "Crysis."

Congrats Intel dev team!

Re:Wow

[Yvan256]

Those aren't award winning films.

They're award winning slideshows.

Re:Wow

[Anonymous Coward]

And then there's South Park, which appears to have been created with PowerPoint.

Re:Wow

[stephanruby]

This is lame. The guy doesn't even claim the video was made in real-time. He claims that the editing of the game can be done in real-time. That distinction is important, because most of the time I see someone demoing a 3-D editing tool on Youtube, they've accelerated the demo by a huge factor -- just to make the video look cool (and it does look cool that way, but it's also misleading). By the way, here is the same demo "teaser" referenced through youtube, there is actually no need to have to wait for the 3 minutes and 38 seconds on that other video for the boring guy to stop droning on, it's essentially the same teaser (with the same building and the same shading) -- it's just been spliced into the interview in small pieces (as if to imply that the teaser was made at the same speed the interview was videotaped at).

What this really means is ...

[WrongSizeGlass]

... now they can pump out crappy movies that have quality CG faster than ever before?

Re:What this really means is ...

[Anonymous Coward]

Hating on the current quality of movies/games/music automatically gets you karma points even if you haven't the least bit of idea of what you're talking about....

Re:What this really means is ...

[Korin43]

Wouldn't this mean that this is just another step in the direction of letting anyone make movies (without needing a billion dollars with of computers and another billion dollars worth of actors)?

Re:What this really means is ...

[biryokumaru]

I hate it when people hate on people hating on something they hate just to get karma points just to get karma points.

It's almost as bad as people hating on people hating on people hating on something they hate just to get karma points just to get karma points just to get karma points.

Grammar works like nesting things, right?

Re:What this really means is ...

[biryokumaru]

I read a really great short story once about a future where all films are made completely on computers, with AI actors. Then one guy starts filming movies with a real girl in them, just with computerized scenery, and doesn't tell anyone. It blows people away just how "real" his films feel compared to normal movies.

Anyone else read that? It was pretty good.

Re:What this really means is ...

[Pseudonym]

Anyone can already make movies without a billion dollars worth of computers and a billion dollars worth of actors. The difficulty is finding a million dollars worth of animators and fifty thousand dollars worth of screenwriters.

Great...

[Beelzebud]

Now maybe they can get to work on shipping on-board graphics cards that can actually play games released within the past couple of years...

As long as Moore's law holds

[Anonymous Coward]

How can there be any doubt that realtime rendering will approach the quality of today's offline rendering when computing power grows exponentially?

Re:As long as Moore's law holds

[binarylarry]

Unfortunately, the faster the processors get, fancier rendering features become possible in the offline space as well.

Realtime rendering will never be on par with offline rendering of the same vintage.

Re:As long as Moore's law holds

[AmberBlackCat]

I think those are Mike & Ike's...

Re:As long as Moore's law holds

[Pseudonym]

Absolutely true, but there is an apex that both achieve to reach which is photo realistic rendering.

No, because "photo realism" is not a goal that visual effects aspires to. If you can take a photo of something, then it's almost always cheaper and better to do that, even though it usually requires many thousands of dollars on crew, make up, sets and lighting.

CGI is used for things that you can't take a photo of, such as a Na'vi or a talking ant. If the space ship can travel faster than light, or the penguin can dance, then "realistic" is not a goal.

(Disclaimer: I used to work in visual effects.)

Re:As long as Moore's law holds

[dredwerker]

I think those are Mike & Ike's...

There's a meme that is gonna stick if only I had mod points :)

Re:As long as Moore's law holds

[ardor]

However, there is a point where CGI is "good enough" for most purposes. Yes, the maximum scene complexity may grow, but even there you may reach a "good enough" point, where you can easily fake the bits that cannot be done. Example: an outdoor scene with a forest in the distance. If the scene is rather static, with little action, the forest in the background may be just a picture. If more movement is involved, but the forest is always far away, impostors can be used. These tricks are cheap to implement and very effective.

You don't need hyper-realistic spectral rendering for typical CGI movies. You could even get away with the CryEngine 2 or 3 for several low/mid budget flicks, provided you do some work on the animation. In my opinion, *animation* is the one factor that is still significantly better in offline rendering.

CGI-quality graphics

[Anonymous Coward]

now there we have an accurate statement: "Computer Generated Imagery" quality graphics

Re:CGI-quality graphics

[pushing-robot]

Obviously he's a member of the Tautology Club that has him as a member.

"Movie-Quality"

[nitehawk214]

"Movie-Quality" is basically a worthless statement. Which movie? Avatar, Final Fantasy, Toy Story, Tron? The quality of digitally produced movies, and the quality of game graphics power are constantly moving targets.

Re:"Movie-Quality"

[Beelzebud]

It could even be a Sci-Fi Channel movie. I have games with better graphics.

Re:"Movie-Quality"

[drinkypoo]

This is basically what I was going to say. The latest crop of "funny fuzzy animal" movies have graphics about as good as the best video games — the secret to making games look as good as movies is apparently to make movies look shitty. I just can't sit through a movie that doesn't look as good as playing a game. I also can't sit through a movie with a worse plot than nethack, but that's a separate issue. Unfortunately, the aforementioned movies suffer from both of these failings.

Re:"Movie-Quality"

[MBCook]

Can anyone tell me how close we are to being able to render Toy Story in real time? Say 1080p?

I know the state of the art keeps moving, Avatar is far better looking than the original Toy Story, but with the limited visual "feature set" used in Toy Story, are we very far from being able to do something close looking in real time?

Can we do it raster, now that we have so many GPU based effects?

Re:"Movie-Quality"

[Anonymous Coward]

Not sure, but I can tell you that we're nowhere near rendering state of the art movie CGI in real time. Vertex and pixel shaders have enabled a class of effects that were previously impossible in real time, but those are all direct lighting effects or crude approximations of indirect lighting. Shadows are not really smooth, they're just blurred. Realistic smooth shadows depend on the size of the light source and are computationally prohibitive on current hardware under real time constraints. Movie-quality CGI includes a class of light interactions which is currently impossible in real time, for example caustics: A caustic is light which is reflected or refracted onto a surface which reflects diffusely. Light being refracted by the surface of a swimming pool is an effect which can be faked but not simulated in real time. Render farms use an algorithm called Photon Mapping to simulate this and other complicated light interactions. This algorithm is conceptually related to Raytracing but even more computationally intensive. It does not map well to the hardware which is currently used in the real time rendering pipeline.

Re:"Movie-Quality"

[afidel]

Toy Story isn't particularly difficult to render, even at the time you could render scenes with better quality in a matter of minutes so with a decade and a half of doubling every 18 months I'm pretty sure it could be done by your average gaming GPU in realtime. The biggest problem was sufficient memory for texture and model details but with 2GB of ram available on consumer level video card's I don't think that's such a big deal these days.

Re:"Movie-Quality"

[Jonner]

If Pixar had been able to render scenes with better quality in a matter of minutes, they wouldn't have needed over 100 machines in their render farm. In fact, each frame took "from two to 13 hours."

Re:"Movie-Quality"

[afidel]

Well, I was talking about a year after the movie came out, obviously the stuff *before* the movie came out that was used on the multi-year project would have been less powerful. Figure 120 minutes, three doublings of cpu power so divide by eight and you get 15 minutes. Increase ram and you can use better textures or more complex poly's.

Re:"Movie-Quality"

[afidel]

Just found some numbers, the SPARC CPU in the SS 20's used for Toy Story were capable of 15 MFLOPS peak, an Alpha 21164 433 which came out about 6 months after the movie could do over 500 MFLOPS peak or about 30 times more. Even the PPro 200 could do 150 MFLOPS.

Re:"Movie-Quality"

[Zerth]

According to this, the original Toy Story needed about 7 TFLOPS to render in real time, although I've seen higher estimates.

87 dual-processor and 30 quad-processor 100-MHz SPARCstation 20s took 46 days to do ~75 minutes, so you need to be 883.2 times as fast to render in realtime. Anyone overclock a quadcore processor to 8 GHz? I suppose setup with 4 quadcore cpus @ 2GHz isn't out of reach.

But then again, the machines might have been IO bound instead of CPU bound, needing to send 7.7 gigabytes per second.

Re:"Movie-Quality"

[Pseudonym]

Blinn's Law states that the amount of time it takes to compute a frame of film remains constant over time, because audience expectation rises at the same speed as computer power.

I think it was Tom Duff who commented that one eyeball in a modern Pixar film requires roughly the same amount of work as a frame of Toy Story.

Re:"Movie-Quality"

[beelsebob]

And just to put this in perspective, current GPUs manage somewhere in the region of 2TFlops, so assuming we can encode Pixar's raytracing/radiocity algorithm into OpenCL that will actually run on one of these cards and not drop to software, then the hard-to-render frames would still take 1.17 seconds to spit out. We need about another 2 orders of magnitude improvement before we're there. That will only take a few years from now though, so we're close, but no cigar.

Re:"Movie-Quality"

[beelsebob]

At 5 TFlops you're still talking 0.5 seconds to render a single frame from toy story, even assuming we can encode their rendering algorithm efficiently onto a graphics card in such a way that it reaches peak performance (unlikely).

As a former (contract) developer on Project Offset

[whiplashx]

4 or 5 years ago, it was basically comparable to Unreal 3. The motion blur was probably the best feature I saw. Fine graphics, but nothing really mind blowing. Having said that, I have not seen what they've done since Intel bought them, but I'm guessing its basically support for Intel's research projects.

As a developer of modern console and PC games, My Professional Opinion is that there's nothing new to see here.

Re:Who will write the software for the bird?

[PotatoFarmer]

At what point will the hardware capabilities exceed the software we can write?

Never. More hardware means programmers can get away with writing less efficient code.

Re:Priorities first!

[binarylarry]

Oh, so *Doom 3* played like a survival horror game.

I see.

define movie quality

[poly_pusher]

As stated by other posters, "film quality" is misleading. Primarily it refers to resolution and remember many cameras record at up to 4k, so the ability to render in real time at ultra-high res is definitely sought after.

Currently, the big push in 3d rendering is towards physically based raytrace or pathtrace rendering.
http://en.wikipedia.org/wiki/Path_tracing
http://en.wikipedia.org/wiki/Ray_tracing_(graphics)
Physically based rendering produces a much more accurate representation of how light interacts with and between surfaces. It has always taken a long time to render using physically based techniques due to the huge amount of calculations necessary to produce a grain free image. This has changed somewhat recently with multi-core systems and with GPGPU languages such as CUDA and OpenCL we are about to experience a big and sudden increase in performance regarding these rendering technologies.

While this game looks great, the engine is by no means going to be capable of rendering scenes containing hundreds of millions of polygons, ultra-high res textures, physically accurate lighting and shaders, and high render resolution. We are still pretty far away from real-time physically-based rendering, which is the direction film is currently headed. So that would have to be what "Movie-Quality CGI" is defined as and this game does not live up to that definition.

Re:define movie quality

[im_thatoneguy]

It's also misleading because films can cheat. You can't see something from every angle and cameras don't always have to move through a space so a lot of what you see are flat cards carefully hand painted and positioned in 3D space.

In the end what really holds back video games is their memory. A small scene can consume in excess of 8GB of memory. That's fine on the CPU where you have a lot of RAM and you can swap back and forth from the HDD. With a GPU you have to load everything into memory which is extremely limiting.

Renderman which is one of the most popular renderers in feature film production is really a rasterizer with a raytracer slapped on top.

As long as games can't go through a post-process hand tuned by a team of artists for weeks they'll look inferior to something in which every frame is hand crafted. It's much harder to create a photoreal game than a photoreal movie.

Re:define movie quality

[Xyde]

>As long as games can't go through a post-process hand tuned by a team of artists for weeks

Well I don't know anything about movie production, but I highly doubt they do this. Are you really saying they take their pristine movie output and begin to photoshop it and make adjustments at the frame level? Do you know how laborious that is when you could just, oh, i don't know, adjust the model you already have and rerender those frames? D

Pictures?

[incubbus13]

Okay, so this is slightly off-topic, but something I've always wondered about.

I can take a 12megapixel picture. And reduce it down to a 12k gif. Or 120k or whatever the compression results are.

At that point, it's just a .gif. (or .jpg or whatever). The computer doesn't know it's any different than a .gif I created in MSPaint, right?

So if I open GameMaker 7, and use that photo as one of the frames in my character's animation. By repetition, I could create a character moving and walking frame by frame.

Right? What's wrong with this?

I understand that on-the-fly rendering is nice. And that the goal is to get a computer to generate a 'real' picture. But. The difference between a 'great' game and an okay one is the graphics. I could (if I could draw) take a pencil and do one of those black and white sketches that almost looks like a photo, and scan it in and use it too.

What are the technical hurdles or barriers that prevent someone from just doing this?

K.

Re:Pictures?

[am 2k]

Actually, that's how the characters in the older Myst games worked (except that they used this great new technology called "video camera" to get moving pictures into them).

This was fine in those games, because the viewpoint was always fixed. That's a restriction you don't want to have in current games.

Re:Or...

[QuaveringGrape]

Or maybe just start supporting OpenGL hardware acceleration? Any day now, Intel...

Boof

[Windwraith]

So this means we are going to see games with movie budgets and no gameplay at all...we already do, but the balance will detriment gameplay even further by reasoning of manpower.

Re:As a former (contract) developer on Project Off

[Jackie_Chan_Fan]

I used to work with Sam Mcgrath and I consider him an old friend. I was fortunate enough to be there from the very start of his new engine and see it develop back when there was no company or anything...

He blew me away years ago with the very basics of its shader editing and render quality. I havent seen newer versions of it in years but... Sam was kicking ass from the start of it.. trust me.

Sam is an incredibly talented coder, perhaps one of the best and most hard working out there. Sammy, best of luck to you if you see this. And Jon, if you're reading.. and I know you are... Modern Warfare 2 rocked ;P Great job. I'm fucking hooked.

CGI

[Lorens]

CGI is awful, they could at least have tried for EGA

FINALLY!!!one

[AlgorithMan]

FINALLY we can have CGI-quality in computer games!
It was such a pain, when computers couldn't achieve the quality of COMPUTER GENERATED IMAGES

"Movie quality" is a relative term

[George_Ou]

Anyone can claim "movie quality" if we're going by Star Wars Episode 4 (original version) standards. The problem is that movies have obviously gotten a lot better though still not completely realistic. The fact is that real-time rendering will always be vastly inferior to slow rendering because you can throw at least 100 times the hardware and 100 times the time for movie making than any gaming computer. Furthermore, you only need 24 fps for movie making while you need a minimum of 60 fps with an average of 120 fps for a good gaming experience.