Game Design and Development for iPhone OS, Part 2

Session 402 WWDC 2010

The iPhone SDK provides a phenomenal platform for mobile game development. Dive deeper into the technologies available to iPhone game developers and learn best practices. Understand real-world solutions for enhancing gameplay and creating high-quality content through rapid design iteration. See how the best games harness the rich capabilities of iPhone OS, and learn how to create a compelling and entertaining experience of your own. This two-part is session essential for developers wishing to create rich gameplay experiences for iPhone, iPad, and iPod touch.

Graeme: Welcome to the second session on Game Design and Development for the iPhone OS.

In, well, I really hope you attended Part One.

Did you all attend Part One?

[ Applause ]

Good game designers.

This session is all about the technical aspects of how we made Quest and we're going to go a lot into optimizing for the platforms.

As I said in the first session there is more than one platform now and optimizing for each of them requires a little thought.

We're going to talk about that.

We're going to talk about our lighting algorithm and how we lit the dungeon and how we actually got to 30 fps and higher.

Our game actually runs at a high frame rate.

We're going to take a, check in with our art department.

Pete will talk about the artwork at Quest and how the art pipeline even made a tremendous difference in getting to the higher frame rates that we see in the game.

Finally we're going to talk about the one dozen things that we think will make a tremendous difference to you in your game development and hopefully you'll get to apply some of that tonight to your games so you can talk to us about it in our labs tomorrow.

So to quickly reiterate Quest was written in two months by three programmers and one artist who really were passionate about making a highly performing application to show off on the iPad, the iPod Touch and the iPhone.

Let's have another quick demo, take a look technically at some stuff.

I'd like to invite Jeff to the stage for a little talk through that.

[ Applause ]

Jeff: So this demo will be more of a technical programmer look at it, sort of look behind the curtain.

So, what you see here is our dungeon level exported via Open Collada from our 3D art package.

You see in this frame as you, it varies as he runs around but it's around 45 to 50,000 polygons a frame and we should be doing around 40 frames a second so go ahead and bring up the console and change to the perf display so we can see what's going on, 50 and 45 so you know you can do the math and we're pushing lots of polygons.

These devices are super awesome.

This is an iPad running 1024x768 and pushing tons and tons of polygons.

It was so much fun making this game on the iPad and on the iPhone 4.

These devices are just awesome to play with.

So we can see he's going through the light and we'll talk about how we do the per pixel lighting here.

We'll show you the shader that did that but as you've seen us kill the demon over and over again, there's been some stuff that's been bothering me so let's just go kill the demon real quick and kind of describe what the problem is.

So as he's fighting you can kind of see that the hero is way up in the demon's kitchen a little bit like we need to [ Laughter ]

We need to back him out.

I mean the demon is just engulfing him right?

So one way we can do that is bring up the console and we can change the runtime, the character collision radius of the hero.

By default it is 100 world units so if we back that out a little bit, let's say 150 and we fight again, we can kind of see the instant change.

So it's better.

Nice crit.

But still, still let's back it out.

Let's try 200.

So as you can see we're changing variables in real time without doing another build.

We're on device and we're making this loop as tight as we can.

We can change as fast as we can.

So let's see how that looks.

Alright that looks a little bit better.

That looks I would send that out the door.

Alright. OK one, a couple more things I want to show you before we leave this demo is let's show you the facade of Quest.

So, let's turn on rotation.

Let's kind of rotate around the dungeon a little bit.

Toggle rotation.

This is the part where you don't want to run over and grab it from him and start typing it yourself.

So as we rotate around you can kind of see we don't have any polygons where you can't see from our default camera view.

We don't have anything down the stairs.

We don't have anything behind the hall of lights for example.

We cut out everything that the camera won't see as a final post process to help our frame rate.

We don't need to render triangles that you're never seeing.

You know, it's a game.

It's a Hollywood set.

We don't, we're not making the most realistic dungeon.

I've never seen a real dungeon so it's fake and now we're showing how we cheat and as a final step let's turn off the diffuse map because we're going to show off lighting right away.

So this is our lightmap for the dungeon.

You can see as he walks this lightmap is shading our guy so as he walks through the red, you see it gets red and as he walks out of the red he doesn't and so we'll go into how we made this lightmap next.

[ Applause ]

So, OK. First quiz of the day.

How many people here see this and see 10 lights in the scene?

10? Alright, how about choice of B, how about 20?

Couple? Alright.

How about multiple choice C, which is always correct, 0?

How many people see 0 lights in the scene?

Alright. Yes and then the final D answer to throw you all off, 50.

Anybody see 50?

Yes. Alright.

Everyone's correct.

So what we do in Quest is we lighten up our dungeon.

We start off with our diffuse map which is just our detail and our base color and then we go into our art package and we put in lots and lots of lights.

Static lights.

In this scene we have over 50 static lights so if you raised your hand on D, you were correct.

But we also have no dynamic lights so if you chose 0 you are also correct.

Our first pass at this was your standard take your diffuse, multiply it times the lightmap and you get this result and it was muddy.

Our artist Pete looked at it and he said yeah that's great but it doesn't look very good so he and he said what I want is the hard light shader from Photoshop.

I want the shadows to be darker and I want the highlights of the frame to be brighter and so, you know, I'm a programmer.

I know math.

I'm like yeah and I have shaders.

Let's do it in a shader, you know.

Let me go look up the hardlight math and put it all in the shader because shaders solve all problems.

So we did that.

Looks great, 12 frames a second.

It was terrible.

It was very, very slow.

So not only do we have.

We have two problems now.

We have one, we have an unhappy artist because it's slow and we've also taken all the power out of our, out of his hand.

He doesn't have any control over the final frame any more.

We do all the fancy math in the shader.

So, we needed to go back to the drawing board and kind of think of a different way to do this.

So, one way to do this was to pre-compute what the hardlight does ahead of time.

What we did is we pre-computed the brightening part of the hardlight shader in the diffuse map ahead of time and then we do our final multiply with the lightmap and our dungeon is now fast again, running very, very fast using a very simple shader.

So we're now fast and we also have the power back in the artist's hands.

He now controls both sides of the equation.

He can brighten up the lightmap, the diffuse map in Photoshop, he can level it, and he can saturate it.

It's up to him and then he also generates the lightmap for us and we just render it.

So here's the hardlight textbook version of the fragment shader that was super slow and I'll kind of walk you through it and it's going to take a long time because it's a very, very slow shader.

First we sampled the diffuse map and scale it.

Then we sampled the lightmap and then we start messing with the diffuse color -1 so we get it from the range of 0 to 1 to -1 to 1.

Next we have to start prepping the left and right sides of the equation for our mix and then we figure out our step and then we do our mix and we finally step on the FragColor and you know the vertex shader is probably still rendering them even though I've taken this long to explain it and as a final don't do this ever, you can see that in all these vec4's there are no precision qualifiers.

There's no lowp or mediump or highp anywhere in the shader so we've given the shader compiler nothing to go on to help optimize it for us.

So, what you see is what you get, very, very slow here.

Here's the final fragment shader.

It's your standard lightmapping shader, sample and diffuse multiply it times the lightmap, stuff it in the fragment color and you're off and running.

So now that we have the dungeon lit how do we load it in, in a fast way.

The first version we tried was let's do the first very naive way to do it, load it all in at once.

So what you see here is our top down or 3/4 view of the dungeon and this was maybe 130,000 polygons at once.

This final version I think ended up to be about 85,000 polygons.

So we did that.

We did the first thing which was a very quick thing to do.

We wanted to iterate as fast as we could so we loaded it all up in memory and that caused a big problem.

It took 5 or 6 seconds to get to that Play button so we've already broken that rule of getting to game play as fast as we can so we needed to come up with a different way.

The way we finally ended up doing it is we split it into logical sections.

The first thing we load is just what you see at the main menu.

Then as you walk down the hallway we start pulling those objects back in.

Now this also helped us be a good citizen in Open jail because there is a limit on how many triangles that you can index per 1 drawl call.

So breaking this up into logical sections got us around that and then finally, look at your game design and think about ways you can optimize your level loading.

For example, in Quest, we didn't end up doing this but we could have if we had designed one of those gates to shut after you walk through.

You reach a checkpoint and the gate shuts behind you.

You know you consider dropping off the first two sections.

You don't need that any more.

You might need the memory for particles and all sorts of fun stuff.

So, drop those old sections if you can but the main takeaway is, think about your game design and what about your game design can allow you to take these shortcuts you know.

You're making a game.

This is the best and most fun thing you can do as a programmer is find those shortcuts and exploit them.

Finally, this is the part of the geometry that you don't see in a game because given our camera point of view and you don't rotate in our game play, you'd never see these triangles so we never export them.

They're never in our binary.

It makes it small over.

It's faster to load.

Another important thing is we keep this around on the art side because we use these polygons to cast shadows and generate the lightmap.

So don't delete them entirely.

Keep them on the art side of the design side because you might need to use them for some use later.

So, now that we've got the dungeon lit and we're loading it let's throw the character in.

Our first pass at this was not good.

We threw down the guy.

He's got a diffuse texture, no shadow.

He sticks out like a sore thumb, doesn't look like he fits at all but we also have another bigger problem which is well we chose to statically light our dungeon with no dynamic lights.

How are we going to light this guy so that he fits in our world in a fast way given that we've already chosen to take static lighting into account?

One way you can do it is to take the minimap and drape it over him as an environment map so as he's walking around you know where he's in world space on the minimap because you already done that for the game, so you sample the texture and then you think well, we're just going to distend our lightmapping algorithm as he walks around.

This worked great and when we first thought of it we were like you know it's not accurate but it's going to look pretty good.

It will keep our frame rate pretty high and it's going to be fun.

Right. It'll be fun enough.

So this worked great for about a second.

Right. You can see as we walked over that beautiful grate that has light underneath it shining up, he's now being shadowed by something he's standing on top of.

Yeah. I laughed.

It was a good laugh but I also felt pretty stupid but I knew I was on the right path because it looked cool but something was; it just needed an extra little step.

So as I said being a game programmer we can take some shortcuts here.

We're going to fake it.

We're going to not use the actual 1024x1024 minimap.

We're going to make a new one and this time when we generate this lightmap we're not going to take any information that's below his knees.

We're just going to take any light that's above him because that's what we really want.

We really just want the lights from above him to shine and we don't, our camera view we'll never see underneath him.

So this is the final lightmap that we came up with and this is the final render.

Now when he was walking down the hall of lights and he was all red and he walked into the darkness and it looked cool but he was, he got kind of chameleoned into the background.

We lost him in the frame.

Now when he was moving it's easier to see but if you take a screenshot and you send out a mail, people look at it and they go well, yeah, he's shadowed but I can't see him.

One way to make him pop off the screen is to add one dynamic rim light.

Now a rim light is one that we put on the camera and we wanted the effect of if the vertex normal is exactly perpendicular to our camera, we want that to be fully bright and as the vertex normal gets closer and closer to the camera we want that to fall off to 0.

You can see that on the demon's shoulder.

The back of his shoulder is our fully bright value and then as it gets closer and closer to the camera, it falls off.

So here's how we did that in the vertex shader.

First we set up a varying.

We need a place to store our result value and then we get the normal from the camera to the vertex, dot it with the vertex normal and then we smoothstep it.

Smoothstepping is basically a nice little quadratic blend that we wanted to do between values between .5 and 1 and that final rim value you stuff into that varying that we created at the very top.

Now as an exercise for you guys, this is a great place to stop and think can I give my artist a knob here to turn because artists love knobs to turn.

They love it.

Give them as many knobs and things to play with as possible.

So one way you can give an artist some love here is to add an attribute called rim light color.

Let them set the color in your 3D world so that when they're walking down the red hall of lights maybe he wants the light to be tinted a little orange, blue, or whatever.

It's up to him right he's the artist.

Or another way you might want to use this is in a very dramatic way.

If we roll a critten and the hero hits the demon for that frame and for the next few frames we can pump that red to be fully red I mean pump that light to be fully red and you know he'll flash red for a second and it'll be you know dramatic.

So there's ways that you can play with this.

This is just the base case.

So we send that rim light over to the fragment shader and you can see that this fragment shader looks a lot like our dungeon fragment shader.

Sample the diffuse multiply it times a lightmap and simply add in the rim light and then stuff it in the fragment color and you're off.

Let's talk about how we animated our guy.

There are two types of animations we considered.

We could have done skeletal animation or mesh animation.

Mesh animation is one of the older techniques where at every key frame in the animation you record the position of every vertex.

Given that our guy when we started was 3200 polygons that information gets big fast, on disk and in RAM.

So we chose skeletal animation.

Skeletal animation is simply what you see here.

This stack of bones which is our matrix transformations and then you also have a mapping of every vertex to each bone or maybe a vertex matched to multiple bones.

For example if you have an elbow, this vertex on your elbow might map to the lower arm and the upper arm equally 50%.

So you have this mapping and then when you animate the bones because in your animation for every key frame you're only storing the bone transformations you then can draw your guy and it looks correct.

So we chose skeletal.

Now there's two ways that you can do this in a performant way.

The first way you should choose is you should look at the GPU first because the GPU is very, very fast.

There are some tradeoffs to using the GPU though.

Let's say that your game is pegging the GPU fully, you have no space left and but you still want to do some skinning so but then you look at your CPU and you're like oh look I have some space there.

Consider moving it back to the CPU.

You have two there you get to play with.

It's a balancing act.

Another reason you might want to do it on the CPU is if you need at any time to know the vertex position in the middle of the animation for, like in the swing.

If you even know that vertex position exactly you need to do that on the CPU because once you send the GPU down to render your stuff it's, you can't get that back in any sort of performant way.

So it's a mix and match.

It's a game.

It's a game within a game.

So as I said before our first Sergeant Shock was one that we made just to get in the game as fast as possible.

It had 3200ish polygons, 60 bones, tongue bone, eyelid bone.

I mean it was a lot right; 4 weights per vertex.

It was just done to get him in the game as fast as possible.

Now from our camera view there's really no difference between a 60-bone guy with 4 weights per vertex and a 2 weights per vertex, 20-bone guy.

There was no visual quality difference so making that optimization was a great win for us.

Do less, your game will run faster.

It works for Quest but it might not work for your game because you're doing a first-person shooter.

One solution does not fit all here.

It's up to you to test, play-test, redo the art assets again and again and again until you find one that works for you.

Alright. Let's get to the console.

A typical game loop that I've been involved in and guilty of is a designer will come to me, you know or Graeme will come to me and say let's make the hero 100 HitPoints.

I'll be just typing away in my, making my best feature ever, I'm like yeah, yeah I'll do it.

Then I go to lunch and then I'll go to the bathroom and I'll go to the meeting, I'll forget about it you know.

I'll finally make him a build.

A couple hours later I finally give it to him.

He plays it.

He's like ah you know what 100 is not fun.

Let's make it 200.

By that time I've gone to bed and he's still late at the office and he'll send the email and the loop just continues.

The loop is very, very slow.

The best way that we've found to manufacture fun is to make that loop as tight as possible.

Stay on device in game like we showed in the demo.

So here's the fastest way to do it.

Designer sets it to be 100 in game, he play-tests right there on the device, changes it to 200 and does it again and again and again and again.

One way you can do this is to data drive your engine.

Expose as many knobs and variables as you can to your designer, to your artist, to your other programmers.

Now we've chosen to use the game console as the throwback.

There's many ways that you can expose variables to your designers.

This is just one example.

Here's how we did it in code.

So this is one type of way you can do it.

You can say at the start of time you can say addConfigVariable heroHitPoints with a default value of 100 and a callback.

So when they bring up the console, type in heroHitPoints = 100, you get this callback.

You get the value.

You can do whatever you want with it.

You can log it.

You can set the heroHitPoints to 100.

It's up to you.

Another pattern is to use a console command.

Console command is one it's just like a variable but it doesn't really have a value.

So for example, like the toggle diffuse that you saw.

ToggleBBox here.

AddConsoleCommand toggleBBox withCallback and then when they toggle the BBox you get the callback and you can do whatever you want.

You can toggle your BBox rendering or you can reload your assets.

You can do whatever you want.

But the point is that you are in game and not reloading the game.

Here's one I like to use a lot.

Let's say that you get inspired one night and you start working all night long and you've got this awesome feature that you're going to do, 6:00 a.m. you're done.

Checking all your stuff, send out the build email, all proud.

Check out my new feature.

This is awesome.

I'm going to bed.

Come in at 2:00.

Find out that you broke it.

You didn't check for an all.

You didn't, something's wrong and now you're the goat.

Anybody? Yeah.

I've done it.

How about if you send that email off at 6:00 and then you get on a transcontinental flight to Maui.

Yeah. Here's how you save yourself from doing this.

You add a ConfigVariable.

Let's call it JeffsCrazyFeature and I'll turn it off by default.

I'll send out the email and I'll say if you want to see my feature, turn on this ConfigVariable and at runtime they can turn on features, turn off features, turn on debug features and you can get on that plane to Maui and feel confident that you haven't broken the build and if you did break the build they can just turn it off and their work is not blocked.

So, you can get a sample of this console code at the attendee website.

It's locked.

It's a sample of the console hooked into the default OpenGL sample that comes with Xcode.

I encourage you to use it.

Figure out new ways to use it.

Extend it.

Use it in your games to data drive your experience, to give your designers every knob possible to let them design and every artist to let them just make great art, you know.

Let them do what they do best.

Thank you for your time.

I'd like to bring up Pete to talk about game art.

[ Applause ]

Pete: Hi everybody.

My name is Pete.

I'm an artist on Quest.

So today I'm going to talk about art, the art style that we chose for Quest, how we built the art assets for Quest and the steps that we took to optimize the artwork to make it run really fast, way over 30 frames a second.

So to get started let's talk about style.

We really wanted to pick a style in Quest that showed off really well on all the Apple platforms whether it was the smaller screen on the iPhone, the larger screen on the iPad we wanted to see the characters.

We wanted to know what they were doing.

We wanted to relate to the world and we really wanted that style to be cohesive.

We wanted it to feel exactly like you're seeing the exact same game whether you were looking at the UI, the characters, the dungeon and because we only had two months to create this, we picked a style that was achievable in that amount of time.

We didn't have a lot of time to do high polynomial mapping.

We wanted it to look good but we really needed to make sure that we were doing the right thing.

So when you're designing your game, style is also very important.

You want to pick a style that's really going to speak to your audience.

If you're designing a history-based game starring Abraham Lincoln you probably don't want to render him looking like an animated character with a bazooka.

You're going to alienate a large portion well maybe that's the greatest idea ever [ Laughter ]

But the people who are coming to your game are going to have an expectation of the style and they want to relate to it and love it.

So let's start getting into it a little bit about how we created Quest.

The. How many people out there know what a texture atlas is?

How many people are currently using texture atlases in their applications?

So this is a great way to really optimize your game and get a lot of really good performance.

You might be surprised to know that this one texture, this 1024 texture accounts for 80 to 90% of our dungeon.

We're loading one texture.

So if out there if you don't know what a texture atlas is basically instead of taking all of this, these small little textures and setting them to the GPU to get rendered and having all these draw call saying render me, render me, render me, render me, we're combining all of these little textures into one larger texture and you're sending one time to get rendered, one draw call.

It's a lot more efficient.

Now as a programmer you're looking at this and you're saying that's great.

You know we get a lot of really good efficiency and it's a great way to work.

As an artist if there are any artists out there, you might think this is insane.

You might have never worked like this before.

You're asking yourself why do I want to work this way and I kind of had that opinion at first but I actually learned to love it because one thing it does is it keeps your artwork unified.

The style of the dungeon is very unified.

When I'm working on the dungeon I bring up this one texture.

I know where all the pieces are going and I can start looking at the colors and looking at the values of everything and unifying it so it all looks like one cohesive piece.

You can see in the upper-left hand corner that's our tileable texture for our floor.

To the right of that is our tileable texture for our wall.

You can see we have pieces for the banner, more organic pieces on the right-hand side for our sculptural pieces like the Gargoyle head or the sculpture of the knight holding a sword.

Bottom left-hand corner we have pieces that we're constantly reusing on doorways, on bridges, so when we start to make a new piece we can take all of these things and know that we have a little bit of the stairs.

We have a little bit of wall.

We can combine them all into one object and load that game and it's very fast and very efficient.

I'm a little bit obsessive about building on the grid.

It's a very efficient way to do both 2D artwork and 3D artwork.

We use the grid for our interface, for a lot of our models, for how we put our texture libraries together, our texture atlases together so you can see this is all on a power of 2 grid or the majority of it is and one of the reasons why we do this is to optimize our texture as much as we can.

We don't want a lot of wasted space in our texture.

Every time you waste space in a texture you're losing an opportunity to make your game cooler.

So we've tried to use almost every single pixel we possibly can in this texture because it's going to be loaded into our game.

We also build on the grid to get a consistent texel density in our dungeon.

So when you see a doorway next to a wall that has the exact same texel density so the details are going to match up exactly.

So you can see on the upper left, that is a 2D texture.

It's on a 256x256 grid.

It's from our texture atlas.

We're applying that to a polygon that's in our 3D art package that's also on a 256x256 grid and that's being rendered in the world at approximately 256x256 pixels on screen.

This is very efficient.

This is a great way to work because you're not putting a super large texture in an area where you're going to waste that resolution.

So if we were to map our floor with a 2048x2048 texture and you're only going to see that on screen at 256x256, then that's a lot of memory that's wasted.

Now given that this is a 3D environment, a 3D world you're not going to always be rendering stuff on screen at that exact texel density.

When things go farther away from the camera you're, they're going to get more, they're going to get tighter when they get closer to the camera they're going to get a little bit blurrier.

So this is just a good rule of thumb to use so you're not putting an exit sign on a doorway that 2048 and you're basically wasting a lot of your memory.

So let's start building a dungeon.

This was one of the first textures and the first pieces that we built when we were building the dungeon.

It's on a 256x256 grid.

It's 256x256 and it's one of our floor textures.

We took it and we started tiling it.

Because it's geometry we can flip it, we can rotate it and we can orient it in such a way that we're getting rid of a lot of that tiling pattern.

We're going to add some baseboard trim.

This is going to be used to kind of soften the transition between the floor and the wall.

It's also on the grid, a little bit crazy about the grid and we're going to add some walls.

We, because these pieces are tileable, we can make as big of a room as we want, as small a room as we want and later on I'll show you even more of the pieces we use to add a lot of variation.

This is just an outline of how the pieces fit together.

Because again we're also, we have all of our vertices on the grid because these pieces fit together in such a way like puzzle pieces we don't see any gaps.

We don't see any holes in our geometry.

There's no weird, rendering artifacts.

It just all comes together.

Now the grid is also just a suggestion.

Because we're building a very you know square very man-made structure like a dungeon it makes sense to build on a grid.

Other things like sculptures, stairs, the broken pieces on the stairs; they're not on the grid.

They don't need to be on the grid.

We wanted to kind of give more of an organic feel to some of the areas so those are completely off.

Although the main structure of the stairs, the height and width of the stairs is on the grid and that's for animation purposes.

If we had an animator who wanted to animate a character walking up the stairs instead of doing a lot of kinematics we can basically tell him the width, tell him or her the width of the stairs and they can animate to that scale.

If we were to make an entire game, if we had that constant scale of stairs, we could use it on everything, even though the look will change, the animation will always sync up.

You won't see any clipping.

It'll look pretty good.

So here is just a shot of a few of the pieces that we used to build the dungeon.

You can see some grates on the floor, some doors, doorways, parts of our sculpture.

Initially when we were building these pieces we were thinking especially with Jeff and Graeme that we would take these and do a random map dungeon where we'd apply a set of rules through code and hit a button and it would give us our own unique dungeon for that time when we started the game.

We didn't have enough time to implement that but it might be an idea that we use in the future.

Art efficiency.

Like Jeff said earlier the initial character that we did, we did pretty fast and it was way above 3,000 polys.

It had way too many bones, actually had individual fingers that could animate.

He had a tongue.

He had some teeth, eyeballs and it was way too much.

It was overkill.

On our devices it looked beautiful but given the camera view of our game you're not going to see any of it.

So we took another pass on him and we just got rid of it all.

We took polygons out of the face.

We got rid of his fingers.

We tried to give the player the maximum amount of quality in art with the minimum amount of resources.

Efficiency is always important even if you had an unlimited amount of RAM and video memory and whatever, you still have to take the time to create this.

So, working with resolutions that are a little bit lower, it's actually will save you a lot of time as well.

Here's a shot of a hand where and it's kind of hard to see but you can see where we just completely cut out all the fingers.

Our hero grabs a sword.

He doesn't do any pointing barely any waving.

He needed a mitt and that's all the player's going to notice is that mitt of that character.

So, got rid of all those polys and made it a lot more efficient.

Next gen graphics are awesome.

I love them.

They make games look amazing but they also come with a huge tradeoff.

If you want to cut your texture memory in half, then you might be able to add more maps and that might make your game look awesome but you have to understand these tradeoffs before you start jumping in to using things like norm maps or ambient occlusion maps or realtime mesh displacement, dynamic lighting, dynamic shadows.

Some of the most successful games out there right now are using very little technology.

Now, it's not to say that there's not a lot of really awesome games that use a lot of technology too but you really need to pick what's best for your game and what your audience is going to notice is if it's very subtle like a subtle specular map and it's causing you a lot of frame right issues, then maybe that feature is just not that important for your game.

So in closing I'd just like to say that a lot of the stuff that we've talked about is not new.

The gaming industry has been using it for many, many years.

It's tried and true methods and at the same time we're not saying that this is how you have to make games.

These are just some ideas that might help you optimize, might help you make better games.

If you were making a Gumby surfing game building on the grid might not make any sense at all but hopefully that we've given you a little bit of information so that you can continue to make awesome games and we can continue playing them.

So, thanks for your time.

[ Applause ]

Graeme: Thanks Pete.

I think you did awesome, huh?

So the one dozen things that we learnt whilst we were making Quest that I think can really apply into your game that I think you can take home tonight and hopefully come and talk to us in the labs tomorrow about.

Number one is optimizing assets.

As Pete just showed, optimizing your assets not only increases your frame rate, it decreases your load time and it makes a tremendous difference to the final frame rate and your right production schedule.

If you're making polygon, you know, 3,200 polygon seismic shocks, it's going to take a little while longer to keep going with all the characters if that's the resolution that you choose.

Choose a careful resolution.

Stick to it.

Optimization also extends out to things like texture atlases and being able to use a single texture for most of your level and as Pete pointed out texture atlas that is full, is a happy texture atlas.

Optimization also applies to shaders.

Our first shader was awesome.

We could have stuck with it.

We really could have cut down number of polygons and finally got that thing up to 30 Hertz and we'd have ended up with BoxQuest but we didn't stop there.

We optimized.

We came up with a way to produce lightmaps that were actually could work with a simple multiply.

The challenge is never in really getting the right results.

The right result is always there.

The challenge to game developers is always to think about how can I get to kind of the right result quickly.

So by ramping up our diffuse map and making it much, much brighter in the very beginning we were able to do a lot of the pre-computing ahead of time and that made a difference in our shader, bringing it down from the 8 lines of code to 3 lines of code that it finally was and increased our frame rate dramatically.

Also, giving the control back to the artist who can then go and go crazy at the higher value diffused texture.

Something else is it's a game.

Now if you have been to Hollywood or seen a Hollywood set you know that Hollywood sets are just these facades.

They're just there to be put in front of the camera and what a lot of us do when we make our world is we make these fantastic 3D worlds that you could really walk around and look in any direction and there's stuff there.

There's trees behind me.

There's all sorts of things going on and you have to remember that just a facade means just a facade.

Build only what the camera sees and in Quest we built only what the camera sees.

If there's a door, there are no polygons on the other side of the door.

If there's a wall that you can't see the ground, that wall is not even there.

It doesn't matter that it's not even there because your game players will never even see it.

It does matter and make a tremendous difference to your bottom line and your frame rate.

So optimizing, always think about your level as a Hollywood set.

It's completely and utterly fake.

Memory is something that's on your device.

I want to have the most wonderful looking game in front of me.

I want my menu to be great.

I want a large Quest logo on the screen.

I want to use an interface that looks beautiful.

I also want to stream my level in so that my level has the most polygons I can get onto it in a particular part of the level.

It doesn't matter that it's going to be gone in the next thing.

I'm going to get rid of it and eject to out of memory.

As soon as this first user interface is gone I'm getting rid of it.

I'm freeing it.

I want that memory back so I can have particle systems.

I want that memory back so I can have two more bad guys.

Purge assets which are not in use, reload them when you need them again.

It lets you be a much more, much better memory citizen and get a lot more into your game.

If you were trying to load your entire game at once, then use UIs and all, you know that wonderful settings screen the user only ever sees once and you're not doing your game or your players any service.

This also applies to texture size.

As Pete said the number of times that we see textures which are 1024x1024 hidden in the end of the level that just shows a little exit sign, I've actually seen that and it's taking up all this memory and it's taking up all this bandwidth as it's rendered because the renderer is going through it because he didn't, bitmaps and bitmaps are made to even using more memory because it's 1024x1024 is ridiculous.

Look at how your game is being rendered and look at that texel density.

Texel density is tremendously important.

You want your texture map to be represented in your 3D world as your artist has authored it.

One way that you can really optimize both memory and loading time is to use compressed textures.

Compressed textures are 1/16th the size of the uncompressed texture and as you can see here on our wall texture, the uncompressed versus the 4-bits per pixel versus the 2-bits per pixel there's almost no difference and something else that's key here with our texture is we are going to take this and we are going to multiply it by some random lightmap value that's between 0 and 1.

We're going to abolish it so just in the name of making something that looks good in the game so this will not even be seen in the game world even once we're finally there.

Texture compression is a great way to load assets faster and to be able to get into your game quicker and saving a whole bunch of memory.

We talked about gameloops and how important it is to do the gameloop in the right order.

Load up that GPU.

Then handle user input.

Then let the aliens update and invade the world because you prepositioned your little base to fire back just in time.

Multiple devices.

You might think that screen size and resolution mean exactly the same thing but in fact it means two different things.

Screen size on our devices is different.

I'm going to hold the iPad in my hand very differently than I'm going to hold the phone or the iPod Touch.

I might hold one in one hand and one in two hands.

You have to be thinking about how what that means into your game design for you to actually implement the same game on different devices.

It's going to mean different interfaces.

It's going to mean that different user interfaces.

It's going to mean different controls.

The resolution difference is also something you're going to have to deal with.

A low-resolution game running on the iPod Touch needs to be up rest and look fantastic on the iPhone 4 and I have to tell you the full resolution game on the iPhone 4 is like nothing you've ever seen.

When you hold that in your hands, oh my God, that thing's a kicker.

Absolutely the most beautiful thing I have ever worked on in my life.

Then the iPad, an amazing device but I hold it differently and I play games on that for longer than I play because I'm sitting on my couch and my wife's watching House.

I don't like House because there's blood and I'm going to play a game for a whole hour.

I'm not the only one that does that.

I know that.

So games play very different on devices too.

Think about the capabilities.

Now I already gave you my little rant on playtesting so I'm not going to reiterate my rant on.

I am. I'm going to reiterate my rant.

Watch other people play your game.

Exceptionally important.

Take their input and apply it to your game.

Always remember that you are not there when someone else buys your application.

They're going to buy your application.

They're going to hit the button and they're going to wonder what to do so you might as well get that feedback ahead of time and watch someone play your game.

Do that constantly.

One of the things that game developers strive for is reality and oh my God we're so big on that.

We must have real physics in our games with 3D rigid bodies and inverse kinematics and all these fantastic things that we can put into games now with, you know because it goes into the back of the box right?

It goes in the iTunes store description.

Real 3D physics.

Real lighting.

I want the perfect full photon pacer that does accurate lighting on my environment and makes a tremendous difference to what how my character looks and does the whole diffuse shadow behind me that spreads out with the right thing on stage with you know all kinds of cool shadows.

Games are not reality.

That's all fabulous.

I'm here to tell you that's all fabulous but if your game runs at 4 frames a second because you have 3D physics running with absolutely fantastic lighting with super realistic shadows, guess what.

Your game's not fun.

Games are like Hollywood and like Hollywood movies they need bright lights shone onto them to look dramatic and look cool.

You need to be thinking about physics that operate within your game environment.

If we were to apply that to Quest, one of the things you may have noticed in Quest, I'm walking on a flat plane ground the whole time.

So therefore a 2-dimentional physics system would actually probably do me fine.

I really don't need a 3D physics system.

2D physics would actually do fine and even what I would try because I like frame rate, is I'd try to get my artist to animate something to fake physics completely so there is no physics in my game but it sure as heck looks like I have it.

Lighting I think as we showed, until you saw how the shader worked I think a lot of you thought oh boy, they have real perfect pixel in their games.

That's absolutely fantastic.

You couldn't tell the difference at our angle and our gameplay what our lighting algorithm actually was but it looked real and that's all that matters.

It's an illusion.

Real lighting in some games makes a tremendous difference if I was a first-person you know version of Quest I would be thinking wow.

Maybe I'll get real lighting algorithm here with real shadows to make a real dramatic difference.

You know what?

My polygon count is very different there.

My. The way our memory in the world and the aim I'm going for emotionally, exceptionally different.

So always remember games are not reality.

What you're faking needs to be fake.

Rendering the world.

We see an awful lot of people take 10,000 draw calls per frame and wonder why they're app goes at 4 frames a second.

They have you know have tens of thousands of draw calls they do to cause into GL and think that their app should run at 60.

Well I'm here to tell you you're orders of magnitude off.

You should be thinking in terms of 10 draw calls per frame.

How many of you have 10 draw calls per frame in your game?

How many of you have more than 100?

The more than 100 crowd got some work to do.

Ten draw calls a frame is how you're going to bring your game frame rate up to where it needs to be.

You need to be looking at the tricks we have told you today with texture atlases.

You need to be looking at how you can load your level and use GL draw arrays as a huge boost to actually get all of your as many polygons on screen as possible using all the features that everyone's going to be telling you about this week in the GL lessons.

We did a lot right in Quest I think given that we had two months.

One of the things that really proved out was our data driven pipeline.

We had to change our variables on the fly.

Getting that power to the artist, giving that power to the game designer to be able to load things on the fly, drop assets in, data driving our game pipeline made a tremendous difference towards us getting this game finished on time.

Collada was a fantastic format to export our 3D assets in and load them on the devices.

Collada is easy to parse, easy to manipulate and it made writing scene graph based 3D system that we used for Quest really easy.

We cheated in the right places.

We cheated with the lights.

We cheated with the lightmaps.

We cheated in the right places to achieve the right effect.

Finally we used Apple technologies.

We trusted the UIKit and AVFoundation, OpenGL 2.0, game kits; that all those fantastic frameworks would work as advertised and guess what?

They do. Of course it was not all roses.

There were many late nights on this project.

One of the things that we initially did was we promised Pete, our artist, we're going to give you fantastic tools.

Well, the number of tools we came up with was exactly zero and we probably should have spent a little bit more time on tools.

We had wrote a fantastic particle system, a parametric particle system that you just had to pass time in and it would render out these fire things, splines and everything and I thought my programmer art looked good but Pete tells me it was really bad but and we wrote this world space particle system where foibles could pass in the sky and just reflect in the ground but we didn't provide Pete with a tool in order to edit particles.

We should have spent more time upfront writing tools and thinking about our asset pipeline.

Our asset pipeline is still not perfect.

If we had written those tools maybe we would have started to write something that loaded the particles directly into the game and dropped them via the HTTP line that we did build.

We started to build the HTTP connection but we didn't actually go ahead and complete that.

We probably scoped a little bit on the impossible side but this is one of the things I actually liked about doing.

The challenge was exciting.

Make something that looks super fantastic, sexy in two months and we did it.

So I'm going to say that we did that right but it was frightening let me tell you.

When you drive home at night and saying well we're going to make Quest in two months.

Finally, a lot of people ask us are we going to make this into a real game?

And the answer to that is no.

We're not.

Quest was only made to be shown here this morning to all of you and to be used for the sessions here throughout the week and we'll be repeating on Friday so I guess that counts.

We're already onto the next thing.

We're onto our next two months and I've got to tell you that's even better.

We read a few books and magazines.

We want to share a few of those with you.

Game Developer Magazine.

How many of you read that?

It has these fantastic post-mortems in every single month of games coming out on all the platforms and we modeled our talk today a little bit on those post-mortem series.

I love that it tells the story of what went right, what went wrong, the approach they took to managing their project, the approach they took to getting the thing out in the store, game design elements that work, game design elements that didn't.

That is a fantastic series and worth getting that magazine monthly just for that alone.

If you're a programmer you should have the Graphics Gems series right there on your shelf.

If you want fast rays intersection you know that's in Graphics Gems 1.

If you want you know fast access align boundary box to boundary box, Graphics Gems 2.

You kind of learn where those tricks are.

Those fast algorithms, you want to get those books, have them on your shelves.

Same thing applies to the Games Programming Gems which kind of has much larger systems in it.

If you want to think about terrain systems.

If you want to think about you know BSP rendering environments.

A lot of that is in there and a lot of those stories in there are also very good about the games they play too.

We recommend a whole bunch of sessions for you to go to so, especially the Game Center sessions, learning about that framework.

It is going to become extremely applicable later this year when millions of people have access to sign up accounts and make Game Center their homes for their games.

A lot of the OpenGL sessions this week are absolutely fantastic, absolutely incredible stuff is happening out of If you go to those you'll see something incredible.

Core Animation was used for everything in our game to drive IUI, to drive our characters, to drive anything animated.

It's a fantastic framework.

The AVFoundation, being able to use that for augmented reality applications as well as sound.

I mean just the access you get from AVFoundation in iPhone OS4 absolutely incredible.

But the most important thing I can tell you is to have fun.

If you are a game developer and you are not having fun, you are in the wrong industry.

Let me tell you.

If you're not coming to work and thinking this is the best thing ever, what the heck are you doing making a game because your game isn't going to end up as fun and you're not going to make any money.

Make something that you think is fun and that you are passionate about.

You have to be thinking every single day oh my God I'm working on the coolest thing ever!

Look! I want you to show me that stuff in the last of our and I want to see how excited you all are about making games on our devices.

You need to be the only way you're going to get the extra little bit of magic into your games is by having that passion and by really applying that passion into the games.

That extra little 5% you had because you had fun making the game and you had fun playtesting it and you had fun showing it off makes a difference.

It's the difference between an application that falls flat and an application that sparkles.

So if you're in the game industry make sure you have fun making games and absolutely finally, if you have any questions please send an email to our Graphics and Game Technologies Evangelist Allan Schaffer.

Thank you.

It's been great talking to you today.

[ Applause ]

Apple, Inc. AAPL
1 Infinite Loop Cupertino CA 95014 US