Tim Sweeney Talks Unreal Engine 3

An anonymous reader writes "Following the recent unveiling of Epic's Unreal Engine 3, Beyond3D has interviewed Tim Sweeney of Epic about the next-gen videogame engine. The discussion is mainly about the 3D requirements, but they also touch on other technologies that are used or required: 'Off-the-shelf 32-bit Windows can only tractably access 2GB of user RAM per process. UT2003, which shipped in 2002, installed more than 2GB of data for the game, though at that time it was never all loaded into memory at once. It doesn't exactly take a leap of faith to see scenarios in 2005-2006 where a single game level or visible scene will require >2GB RAM at full detail.'"

Comments

Memory and Windows

[ADRA]

I thought that I read the memory addressing in 64bit windows was also set to 32bit addresses. Would this means that current 64bit windows binaries also limit this? It'd be pretty cheasy to have multiple processes with IPC's to fully load the games, or anything into memory.

Re:Memory and Windows

[NanoGator]

"It'd be pretty cheasy to have multiple processes with IPC's to fully load the games, or anything into memory."

Not for us Dual Processor owners. ;)

Re:Memory and Windows

[Foolhardy]

Ummm, no. All pointers in 64 bit programs are 64 bit. The current amount of that address space devoted to user-process memory is 512TB. See this; it's about win64 on Itanium, but I'm sure AMD64 is the same.
Perhaps you are thinking of PAE on 32-bit systems?
Windows is fully capable of providing real 64-bit addressing. It even causes driver problems; you can't use 32 bit drivers in 64 bit Windows.
Current versions of OSX, OTOH, can't. They use memory windowing similar to PAE.

Re:Technology goes forward...

[Grand]

I think the introduction to realistic physics is a huge leap forward in gameplay.

Well....

[El_Ge_Ex]

It doesn't exactly take a leap of faith to see scenarios in 2005-2006 where a single game level or visible scene will require >2GB RAM at full detail.

That might help explain DNF

(ducks...)

RAM Inefficiently Used

[shadowcabbit]

UT2003, which shipped in 2002, installed more than 2GB of data for the game, though at that time it was never all loaded into memory at once.

This makes sense. I was able to run UT2K3 without a problem, but after installing UT2K4 I've been playing less solely because the game is a bit jerkier, takes forever to load initially, and is less reliable (I get "hardware failures"). I have a suspicion that this is very much related to RAM usage. I'd love to see an accurate depiction of how detail settings affect RAM usage-- ie on such and such a detail level, you use X amount of RAM. How about a patch for the UI to optionally show this? I know it would be useful for about, oh, a thousand users tops, but knowing how much leeway I have in my detail settings would be a damn nifty thing to have.

Re:RAM Inefficiently Used

[Vaevictis666]

How about using the console?

From Planet Unreal,

MEMSTAT - Displays Windows memory usage
STAT ALL - Shows all stats
STAT AUDIO - Shows audio stats
STAT FPS - Displays your frames per second
STAT GAME - Displays game stats
STAT HARDWARE - Shows hardware stats
STAT NET - Shows network game play stats
STAT NONE - Turns off all stats
STAT RENDER - Displays rendering statistics

You probably want to use memstat. While in the game, hit the backquote key (often called the tilde key, ~) to bring down the console. Type in the command, hit enter.

I tend to avoid stat all because it just crowds the screen, but stat fps is useful for determining the effects of display settings as well (for performance)

Re:RAM Inefficiently Used

[netfool]

If you have some time to waste and extra RAM, try installing UT on a Ram Disk.

I did this with Quake 3 a couple years ago and it worked great. I didn't exactly see a huge gameplay performance increase, but the levels loaded instantly.

Tim Sweeney and... Unreal ZZT?

[MiceHead]

Tim Sweeney will, in my mind, forever be linked to the wonderful ZZT. This interview, given some time before Unreal 2 is an interesting contrast to the one posted above. In particular, he talks about ease-of-creation:

Hercules: You moved onto to other, bigger projects long ago. It must be good to know that the first thing you ever created is still used/played a lot. Does ZZT still cross your mind, sometimes?

Tim Sweeney: Yes, one of the interesting things to do is contrast ZZT and Unreal, and look at how incredibly far we've come in graphics quality in that time. But also to see how little the industry has progressed -- or maybe even gone backwards in some respects... So, how will game development be 10 years from now? If levels take six months to build, and compiles take 5 hours each, and it costs $20 million to develop a game, then developing games won't be fun or even possible anymore.

I'm a fan of creation tools that are accessible to anyone who can play the game. (Casual players who may not be technically inclined.) As a developer, I'm hoping that we will be among the first to offer something that lets even the most casual user plink around. As a player, I'm hoping that Sweeney has retained this philosophy, and that future Epic offerings let us build -- at least a little bit -- with the same ease that ZZT did.

Re:Technology goes forward...

[Anonymous Coward]

Why? Is it because newer games simply look more realistic? Or maybe because the threshold for suspension of disbelief somehow goes up the older and grizzlier you get, so your aged brain needs to see that particles are moving in the right direction and speed?

Or maybe realistic physics somehow makes games more FUN?

Me, I think people are impressed by realistic physics (especially with regards to PC games, and within that group, especially FPS) because that's what developers tell them to look forward to. It's something that a bunch of mathematicians in a back room can easily improve upon from the previous generation (along with graphics and AI), not something that necessarily makes games more fun to play.

So I can somewhat agree with the original poster. Gameplay hasn't improved, or even gone backwards; it's stagnated. It's just the display technology and marketing that's improved.

Bah, I can't resist.

[LordPixie]

It doesn't exactly take a leap of faith to see scenarios in 2005-2006 where a single game level or visible scene will require >2GB RAM at full detail.

<Insert obligatory Microsoft Longhorn joke here>

Kidding aside, this really isn't surprising in the slightest. Games and memory have all been exhibiting 'Moore's Law'-like performance. Heck, most high end rigs nowadays could easily install and play, in memory, an entire game less than a few years old.


--LordPixie

Re:Bah, I can't resist.

[88NoSoup4U88]

One of the interesting things from Counter-strike : Condition Zero, is that you can play the game straight from CD ; it installs and plays in memory, and does that without a hitch.

Taken the HL engine is highly outdated, it was still very funny for me to see.

Re:Bah, I can't resist.

[ObsessiveMathsFreak]

Didn't they say longhorn will require over 2GB of RAM just to start up?

Lets face it. By 2008 PC games will need 10GB of RAM, quad itanium 5s, DVD Blu-Ray drives, 3D Monitors, sidewinder joysticks, headset and a credit card for each and every play. The gameplay will be just as shallow as it is now, but the graphics will be 'unreal', heh, and we will all flock to buy them.

While all this is going on the 14 year old Super Mario World is still stellar, Quake is still being played, and Micro Machines 2 is STILL the pinacle of racing game fun.

Wake me up when gameplay gets past 1999.

Who's gonna make that?

[KDR_11k]

That's all nice and well, but who actually makes the content that fills up those 2GB? You'd need a pretty large team and several months or years to make that much stuff, if you need it per room it wouldn't surprise me if future games were as long as movies or had a level design like Halo or Metroid 1 (that is, you have room 1 ten times then room 2 ten times then a few room 1's and maybe a room 3 with a really big monster for a little variety).
This is going to hurt gaming. We're already seeing shorter games and copied&pasted rooms simply because the effort to make those rooms is too high.
I have a feeling that despite having lower sales, making a 2d game with a tiny team in a few months might actually have larger profit margins than top-end development.
Also, as always, higher costs mean more need for the games to actually sell means publishers won't allow as many risky games to be made since taking a risk on one could blast ther entire company.

Re:Who's gonna make that?

[NanoGator]

"That's all nice and well, but who actually makes the content that fills up those 2GB? You'd need a pretty large team and several months or years to make that much stuff..."

2D and 3D artists make the content that fills that space. The thing to remember is that it isn't necessarily a linear relationship between how much arist time is needed and how much RAM is being taken up. Using 2x the texture size, for example, doesn't take twice as long to generate. A lot of time spent on making 3D art is in shrinking things down to meet the requirements.

Check out this image I made here. (Note: That's not a game model.) *All* of the textures were originally generated at 3072^2 resolution. They were too high for my tiny gigabyte of RAM, so I had to knock them down to 2048^2. If I had started at 2048, it wouldn't have been much faster to generate them. The source imagery was big enough in either resolution, so short of the extra processing time it'd have taken, it would have been pretty much the same.

The real time spent will be in making something more ambitious. Twice as long? I doubt it. Maybe one day when the game machine has specs that exceed the artist abilities, but we are generations away from that. The tools we have today are pretty darned cool, and they're only going to get better as each generation goes by.

In short, these companies already have the talent *today* to put 2 gigs worth of content on the screen.

Re:Who's gonna make that?

[alphaseven]

This is going to hurt gaming. We're already seeing shorter games and copied&pasted rooms simply because the effort to make those rooms is too high.

I don't think it'll be quite that bad. Nowadays you already have game designers making large levels that have to be subdivided into smaller levels with load screens in between, and things like textures and models start off as highly detailed but are reduced down significantly for the final game.

Re:Who's gonna make that?

[Torgo's Pizza]

Give an artist a number and he'll easily double it with content. I've never seen an artist have problems filling up space. The sort of problem will be solved. There's always a company out there that's will to make a content creation tool to help out. If you're seeing copied and pasted rooms, that's more due to a poor developer than due to a space limitation.

You do raise some better points at the end of your post. A good 2D team could have a larger profit margin, if they're lucky to make any money at all. Doom 14 with a 1% margin will make more money overall than a no-name 2D game with a 10% margin just based on volume of sales. Publishers know this and are willing to go with the amount of money rather than the profit margin.

Higher production costs is a big deal and is having an effect on the industry. The recent slate of studio closings is proof of this in my opinion. Smaller publishers, much like the smaller movie studios, are going to be the sources of innovation in the gaming industry.

Re:Who's gonna make that?

[Ford Prefect]

If you're seeing copied and pasted rooms, that's more due to a poor developer than due to a space limitation.

I'm a single-player mapper for Half-Life in my spare time. Stuff I've done has been fairly well received. And I can tell people this - map design for single-player games is difficult. I can spend a week perfecting something that'll last the player ten seconds. A simple room can take days to build, and this in on the original Half-Life where a suitably textured cuboid can be just about anything.

In a modern engine, the workload is increased enormously. You need 3D modellers to create the static meshes (which replace those textured cuboids of the past), texture artists to do the map-specific texturing (high-resolution now, and it can't all be photo-sourced), voice actors for the map-specific dialogue. There's scripting (which needs to be tested for all possible routes and combinations of routes - non-linearity's the thing!), there's map geometry, there's programming, there's playtesting, there's tuning...

And it all has to come together into a cohesive whole - and judging by some of the game demos I've played recently, it often isn't. Copied and pasted rooms imply that the game designers are concentrating on the wrong area of the game - graphics might be lovely, but if the map design stinks, the game's in danger.

Re:Who's gonna make that?

[burns210]

"I have a feeling that despite having lower sales, making a 2d game with a tiny team in a few months might actually have larger profit margins than top-end development."

Even more so, if you write the engine to a solid, expandable, and powerful (while easy to build upon, ofcourse) 2d game, then have multipl in-house, independent teams write games off of it, and lease the engine to other companies... You could have a team of 7-10 people, using an engine your company wrote, acting as their own independent game development company within your own corporation.

Instead of a 50 person team for a high-end 3d FPS or something, try 5 teams of 10 people writing storylines, graphics, nuances and gameplay around an single engine and get 5 times as many games(that sell for less money and in fewer numbers, mind you) that are factionaly less cost to produce!

Re:Technology goes forward...

[trompete]

Yeah. UT2004 at least had vehicles, but I've been using those on Battlefield 1942 for almost two years now. Show me new gameplay, and I'll consider buying one of the new games. I'm just sick of the same old FPS crap. Deathmatch is dead.

I bet Epic and ID Software are making these new engines just to license them to other companies.

well, uh...

[metalmario]

will the AI be still as dum as today? scripted and all that boring stuff? all those MiBs go to graphics ?

Re:Technology goes forward...

[mahdi13]

I bet Epic and ID Software are making these new engines just to license them to other companies.

Is that really a bad thing? Some of the best FPS games are from these licensed technologies from id and Unreal

the Quake3 engine gave us Return to Castle Wolfenstein and Jedi Knight 2: Jedi Outcast
The Unreal engine gave us Deus Ex and America's Army

We all know that even if Doom 3 sucks as a game, the engine will licensed and used in an even better game

Re:Technology goes forward...

[trompete]

I totally agree that great games have come from other engines.
My point (sorry it wasn't clear) was that they focus on the engine and then let other people worry about the gameplay. That's why the unreal games are lame.
Speaking of Unreal-based games, how about that Duke Nukem Forever game :D

Re:Technology goes forward...

[recursiv]

Given certain constraints, the more complicated the rules that govern a FPS, the more depth it can have. You could try totally non-sensical rules, such as:
"If you enter water while strafing, your movement speed while you are in the water will be 300% normal"
but that's not usually very much fun because people can't relate to it. More realistic physics allow more fine-tuning of strategy with a minimum of annoyance incurred by adding more rules, since it's an approximation of what people are already familiar with.

Re:Technology goes forward...

[KDR_11k]

Compare the vehicle physics of Unreal 2 XMP and Unreal Tournament 2004 and you'll see that physics DO add to certain games. Of course, not every game benefits from physics, but some do. The problem is, the suits tend to think that realistic physics will help ANY game...

OSS Engines?

[joeljkp]

I've been wondering about this: is there a project to create a competitive OSS game engine? Or any commercial games produced with such a thing?

Is it because it's freakin' hard?

Re:OSS Engines?

[shadwwulf]

Crystal Space fits the bill in my opinion.

It is being used for a couple of commercial level games from what I understand.

Re:OSS Engines?

[Moonshadow]

Another good one is Ogre. It's purely a rendering engine, which lets you choose your collision/sound/networking/whatever else libraries, but there are a few engine frameworks springing up around it. It's fast, very clean, and capable of a lot of current generation effects (well, it has full shader support, so I guess it supports most anything you can code a shader for). If C# is your flavor, Ogre has a cousin called Axiom that is just as functional. Axiom is intended to be a game engine, but is very much in its infancy, so there isn't too much besides (rock solid) rendering in place there yet. Still, though, both are very clean and excellently designed, and are both well worth a look.

Hey Tim!

[superultra]

-cough-96k-cough-

Re:Technology goes forward...

[Grand]

For me, it is not because it 'looks' more realistic. Games with 'realistic physics' as of now are much more immersive. Arcade games are fun to play but I tend to only play them for very short periods of time. Things like realistic damage make the game as you say look realistic, but it brings much more satisfaction in gameplay for me. The newer games coming out like HL2 with 'true' physics just brings gameplay to a new level. Take for example FPS's, all of them are basically the same in that as a whole they are pretty linear on things you are able to do. Look at the HL2 video that came out, using the environment adds so much more to gameplay.

Intense

[zaphodchak]

Holy crap. Actually, i have to clean out my case. My MX440 SE (PCI) just wet itself. I've heard talk that U3 will want 1 gig of VRAM for full detail, which needless to say, doesn't really exist (for mortal consumers) yet.

Re:Technology goes forward...

[Blakey Rat]

Yeah. UT2004 at least had vehicles, but I've been using those on Battlefield 1942 for almost two years now. Show me new gameplay, and I'll consider buying one of the new games. I'm just sick of the same old FPS crap. Deathmatch is dead.

So, uh... what you're saying is that you never played Unreal Tourney 2004? Because that what it amounts to.

Deathmatch, although in the game, is one of the less popular modes. Much more popular is Onslaught and Assault modes, neither of which is like Deathmatch at all and both of which are much more fun (IMO.) People play on teams now, not solo.

UT2004 also includes Capture the Flag, Deathmatch (naturally), Team Deathmatch, Double Domination (which is fun but not enough people play it), Bombing Run, Mutant, Last Man Standing and Invasion.

If you can't find anything in any of those game modes that interests you, then you're sick of gaming.

Re:Technology goes forward...

[AuMatar]

Realistic physics makes sense in some games, where photorealism is a goal. But realistic physics is a loss for most games.

Think of the great games of the past. Donkey Kong- realistic physics? Nope. Good game? Yup. Mario- realistic physics? Nope. Good game? Yup. Zelda- same. Even other genres of games: Street Fighter 2, NBA Jams, etc. None featured realistic physics, but all were great games.

Realistic physics is a crutch "feature" that developers claim do to the current realism push. Its the same thing that makes them claim 1st person and 3D are better because "its more realistic", when truthfully most games would be better without them. The truth is that it makes all games indistinguishable- you can play any FPS of the last 5 years and have a hard time telling them apart. Its why the industry is in a rut right now- gameplay has been ignored at the expense of "realism" and graphics.

Woah!

[oman_]

2 Gig of data loaded into ram to run a game?!?

I think while game developers a whining about rediculous resource limitations the creative developers will be doing sensible things like creating algorithmic game assets using iterative fractals or some other more advanced techniques. In the end you'll have products that are smaller, faster, and cheaper to produce.

I guess the lack of creativity isn't surprising considering that Epic is still making the same old FPS games.

Re:Technology goes forward...

[inkless1]

http://unrealmods.org

Check out the mods for UT2004 - that's where real gameplay innovation seems to occur.

Re:Technology goes forward...

[rufo]

So? A lot of people think the same of Quake 3, yet it's spawned countless excellent games (Jedi Knight 2, Star Trek: Elite Force, Heavy Metal FAKK2, Soldier of Fortune 2, American McGee's Alice, Return to Castle Wolfenstein, Call of Duty, Medal of Honor). Unreal Tournament 2003/4 are generally considered excellent, yet some people think they're lame, and they've spawned countless excellent games (UT2003/4, America's Army, Splinter Cell/SC:PT, Rainbow Six 3, Lineage 2, Unreal 2, DNF (yeah, yeah), XII, Deus Ex: Invisible War, Thief: Deadly Shadows). Some of the games on both lists may not float your boat, but chances are at least one of them is your bag of tea (unless you happen to like coffee/RTS/RPG games).

It's no secret that id and Epic make these engines to license them. They try to make good games out of them too (whether or not you care for the gameplay of Quake 3 or UT2003/4), but the games are also largely showcases for the engine, and an easy source of revenue.

(Just for the record, I played Quake 3 to death, and will be picking up UT2004 shortly now that I have a computer capable of playing it).

I've only got 512 megs of ram now...

[Anonymous Coward]

And they think I'm going to have 2 gigs of ram in just two years? Not at these prices! Hell, I'm still running a Radeon 7500! I've been waiting to upgrade until Half Life 2 and Doom 3 come out. No sense in buying a new card before then.

They're nuts. I develop games. I don't see any need to use that much ram for textures. Look at what games like Metal Gear on the PS2 look like, and they've only got a fraction of the ram that PC game developers have.

Before they switch to high res texture,s maybe they ought to start hiring artists who can make the most of what they have. Because IMO, Unreal's art looks like crap, and they're wasting 90% of thr polygons on the characters rather than the environment, because it's easier to throw more polygons at chracters than to build more detailed levels.

Re:Technology goes forward...

[Foole]

None featured realistic physics, but all were great games.

Don't forget Tetris!