More Cake is Coming (Portal 2)

[Starglider]

I feel like an old man reminicing about the good old days when you had to buy a new PC every 3 years or you'd be rooted trying to play anything modern. But boy did new games visually wow me frequently.

To be fair, graphics are plateauing as photorealism is approached. The next console generation will be delivering consistent 1080p60 in stereo 3D with multiple usable teraflops of GPU power (yay no more HDR kludges), and that will pretty much be that for graphical realism. Yes it would be nice to go to real time ray tracing and quad HD, but I think serious graphics apathy will set in at that point.

You've no doubt heard this before, but I'm much more excited about better physics, procedural content and general realism. GP-GPU techniques are possible on the current console generation, but they're a pain in the ass to implement and barely worth the effort due to overheads. DX11 finally implements a decent high-performance GP-GPU API and it will be fully polished up and generally exploited in DX12 / eighth generation consoles. I would say I was excited about better game AI as well, but alas I haven't seen anything really promising in that direction since Radiance AI failed to deliver.

[adam_grif]

To be fair, graphics are plateauing as photorealism is approached.

Yeah I know, I remember predicting a few years back that before 2020, the bottleneck on graphics quality would be the amount of money a company is willing to invest into it as opposed to hardware. Turns out it will be far sooner than that.

You've no doubt heard this before, but I'm much more excited about better physics, procedural content and general realism.

Procedural content has never done anything for me. Physics is nice of course.

I would say I was excited about better game AI as well, but alas I haven't seen anything really promising in that direction since Radiance AI failed to deliver.

I remember when F.E.A.R. was the shit. A.I. still holds up today. It's what, four years later? Five? A.I. is easily the most underperforming element in the entire industry, as far as I can tell. The trend seems to be towards pushing multiplayer as an alternative to good AI, which makes me sad since it means I have to cope with unfathomable levels of douchebaggery. Watching a multiplayer game try to be story driven is like watching a one legged man in an asskicking contest. Suspension of disbelief is nigh impossible.

[Starglider]

Procedural content has never done anything for me.

Procedural content boosters usually talk about the ridiculously over-ambitious stuff, like that Infinity MMORPG project Blayne was going on about. Or endless square km of procedural terrain. That isn't terribly realistic or even interesting. What I want to see in the near term is this; take Mass Effect 2. There are about three crates, two terminals and two fuel tanks that are prominently replicated across nearly every mission in the game. This is annoying on both a conscious and subconscious level. The limitation is partly storage space, but mostly artist time; character and weapon models got most of the attention, it wasn't practical to make more than a couple of brushes of each type.

By contrast even in games as old as Oblivion, every tree is unique, due to procedural generation. It turns out that writing a procedural generator for trees is relatively easy, manufactured objects are harder, but eventually someone is going to make libraries for this. In the long term it's the only way to stay on top of rising production costs, and if you want to do it at runtime you need more computing power. Cracking this problem will eliminate unwanted repetition of environment detail, which IMHO will do more for visual quality than implementing super-phong trinormal megapixel inverse UDR subrefraction shading (tm).

I remember when F.E.A.R. was the shit. A.I. still holds up today. It's what, four years later? Five?

Fear's AI isn't even that advanced; it's actually similar to HL2's AI, but with extra voice acting to make the player think the enemies are using complex strategies. In actual fact there is no co-ordination between enemies, but turns out that if you have some 'flank them', 'fall back' etc radio chatter players will imagine it.

EDIT : Now that I think about it, I've got STALKER and FEAR confused. FEAR was the game that actually implemented STRIPS in real-time, albeit in cut-down form, which was genuinely impressive. It is a real shame that that was something of a high water mark in FPS AI.

A.I. is easily the most underperforming element in the entire industry, as far as I can tell. The trend seems to be towards pushing multiplayer as an alternative to good AI, which makes me sad since it means I have to cope with unfathomable levels of douchebaggery.

Yeah, don't get me started on that.

[wautd]

Yay, finally to be re-united with my beloved companion cube.

[adam_grif]

There are about three crates, two terminals and two fuel tanks that are prominently replicated across nearly every mission in the game. This is annoying on both a conscious and subconscious level. The limitation is partly storage space, but mostly artist time; character and weapon models got most of the attention, it wasn't practical to make more than a couple of brushes of each type.

I often get annoyed in real life because I see a whole bunch of cars that use the same models over and over, and are blatantly just pallet swaps of each other. For shame reality, for shame.

The tree example is a good use of it, because trees tend to be very unique and a bit chaotic in their form. But something mass produced like the crates isn't as necessary.

Fear's AI isn't even that advanced; it's actually similar to HL2's AI, but with extra voice acting to make the player think the enemies are using complex strategies. In actual fact there is no co-ordination between enemies, but turns out that if you have some 'flank them', 'fall back' etc radio chatter players will imagine it.

But... monolith wouldn't lie to me would they?

[Starglider]

The tree example is a good use of it, because trees tend to be very unique and a bit chaotic in their form. But something mass produced like the crates isn't as necessary.

It's acceptable when the whole game is confined to one location, e.g. Doom 3. When you're supposed to be traveling the galaxy visiting alien civilisations on two dozen different planets, not so much.

Another use for this is in tailoring level geometry to maximise player enjoyment. Left for Dead 2 does this quite successfully, but it's limited to plugging together hand-designed sections in different orders, which is a) expensive to design and b) wouldn't really work in an open-world game. Future systems will do the same thing in a more fluid and flexible manner.

[adam_grif]

True enough, but we only really see human colonies and some merc bases using those generic assets. It didn't bother me so much.

[Zixinus]

Here's a question: is there anything particular in modern engines that couldn't allow gameplay that could be done with an engine from 5 years ago? Or 10 years ago even?

Procedural generation allows far greater variation in gameplay. But what I would like, is a physics system where I can actually destroy stuff and not feel like being in a static room.

[General Zod]

Here's a question: is there anything particular in modern engines that couldn't allow gameplay that could be done with an engine from 5 years ago? Or 10 years ago even?

Procedural generation allows far greater variation in gameplay. But what I would like, is a physics system where I can actually destroy stuff and not feel like being in a static room.

Destructible environments are something relatively new that couldn't be done on older engines at all or very well if they could, and probably the most obvious example.

[White Haven]

Portal's own...portals...are a good example as well. Not so much allowing you to place teleporters in realtime, but what it does to line-of-sight and line-of-effect in the game's environments.

[Zixinus]

Portal's own...portals...are a good example as well. Not so much allowing you to place teleporters in realtime, but what it does to line-of-sight and line-of-effect in the game's environments.

Yes, but that was done with the original HL engine as well. Remember that Portal's team was hired because of their mod that also featured Portals (if a more primitive form of them). It was called Nautical Drop or something.

Destructible environments are something relatively new that couldn't be done on older engines at all or very well if they could, and probably the most obvious example.

While I think that's probably a good example, I have to ask: how many games truly feature destructible environment rather than just some environment with destructible bits?

[General Zod]

While I think that's probably a good example, I have to ask: how many games truly feature destructible environment rather than just some environment with destructible bits?

Prototype, TFU and Red Faction come to mind, but that probably depends on exactly what you mean by "truly destructible".

[Joviwan]

Yes, but that was done with the original HL engine as well. Remember that Portal's team was hired because of their mod that also featured Portals (if a more primitive form of them). It was called Nautical Drop or something.

The hell are you talking about? HL-1 never had portals that you could see through, shoot through, throw things through or hop back and forth between, let alone ones that you could dynamically place yourself. Unless you're trying to say that Nautical Drop was made on the HL-1 engine (and I'm pretty sure Nautical Drop was proprietary), I have no idea what you're talking about.

[Ace Pace]

HL-1 also never had buy menus, or pretty much anything a mod has. An engine is just a system for loading and displaying resources, such as maps, textures, models, and running scripts such as AI behavior.

To whit, Narbacular Drop (the proper name!) was based off a proprietary engine developed by the student team. Since no one bothered to just check this, I went to their student website and read this document.

Their sole connection to Half Life was using the WorldCraft map editor.

[Zixinus]

Yes, but that was done with the original HL engine as well. Remember that Portal's team was hired because of their mod that also featured Portals (if a more primitive form of them). It was called Nautical Drop or something.

The hell are you talking about? HL-1 never had portals that you could see through, shoot through, throw things through or hop back and forth between, let alone ones that you could dynamically place yourself. Unless you're trying to say that Nautical Drop was made on the HL-1 engine (and I'm pretty sure Nautical Drop was proprietary), I have no idea what you're talking about.

I am talking about the HL-1 engine, not the game itself.

Although, looking it up Wiki, Nabacular Drop was developed on a different engine. However, my point was that it was developed in 2005 (ie, a 5-year engine). You could develop a Portal-like scheme on a 5-year old engine (hell, Prey wanted to be that, however the portal thing was downplayed a lot). As for its capability:



Same developers as Portal.

[Starglider]

Nabacular Drop doesn't use a conventional 3D engine, it's implemented on a voxel map (similar to Wolfenstein 3D) translated into polygons for rendering. A good decision for a student game project, but not acceptable for any major game.

Here's a question: is there anything particular in modern engines that couldn't allow gameplay that could be done with an engine from 5 years ago?

It depends on the engine design; how well modularised, how scalable, how good is the documentation etc.

Or 10 years ago even?

There are a few key technologies that would be impractical to retrofit to really old engines;programmable GPUs, pervasive physics (particularly the consequences for the collision detection) and bones vs morph target animation.

But what I would like, is a physics system where I can actually destroy stuff and not feel like being in a static room.

Like most things, this (probably) isn't a fundamental engine issue, it's a question of priorities. HL2 like most FPSes has smashable crates, tables etc but most of the geometry is invulnerable. A few games (e.g. the Frostbite ones) have destroyable wall segments; harder because it means the occlusion, lighting and pathfinding have to be fully dynamic. Mercenaries 2 and Red Faction: Guerrilla have fully destroyable buildings, but the tech is kinda flakey and involves significant trade-offs (at least, if you want it to run on a current-gen console). Getting this adopted in all games is a combination of the technology progressing so that it can be done without sacrificing other bits of the game, and the marketing progression from 'standout feature that helps sell the game' to 'checkbox feature that we have to have to be taken seriously'.

I'm not familiar with Source internals so I don't know if it has core data structure issues that preclude fully dynamic levels. It's possible, Portal didn't have any mobile AI enemies, so no dynamic pathing, and didn't need dynamic occlusion due to the portals being implemented by render to texture (not practical for more than a handful of appertures). As mentioned L4D2's dynamic geometry is based on a few prebuilt blocks. I doubt it though, it's a pretty obvious forward compatibility issue.

[Stark]

Hen people praise FEARs AI expectations are low. It LOOKED good because they were confined, easily pathed static areas heavily scripted. Regardless of how 'good' it was it tactically sucked.

And Starz, Epic are sensible enough to version their engine constantly. I'm pretty sure they're currently selling UE3.75.

[Joviwan]

HL-1 also never had buy menus, or pretty much anything a mod has. An engine is just a system for loading and displaying resources, such as maps, textures, models, and running scripts such as AI behavior.

To whit, Narbacular Drop (the proper name!) was based off a proprietary engine developed by the student team. Since no one bothered to just check this, I went to their student website and read this document.

Their sole connection to Half Life was using the WorldCraft map editor.

Thanks a mundo, for looking it up and the name. I knew it wasn't 'nautical,' but as I'm typing this I'm also imaging three workstations and juggling hard drives, so looking it up was secondary to ranting and raving. or something.

And "buy menus" is hardly equivelent to the aperture technology in Portal. One is a GUI interface similar to something the game already head (a HUD, and weapon selection,) the other is a way to dynamically paint doors that take you across vast gaps of non-linear geometry while looking around at where the otherside takes you, being able to throw physics objects through, and having AI both see you and shoot you. I'm not convinced you can do that in the HL1 engine, and that was my point. I'm open to being corrected, of course.

[Dominus Atheos]

Portal's own...portals...are a good example as well. Not so much allowing you to place teleporters in realtime, but what it does to line-of-sight and line-of-effect in the game's environments.

You can get that effect on much simpler engines, like ASCIIpOrtal:

[Uraniun235]

Also, I've been playing games on the Source Engine since 2004. Even when Ep 2 came out it was starting to get a little bit long in the tooth. Yes, it's seen upgrades, but 6 years is a loooong time for an engine. They've added a few effects and optimized a thing here and there. It's really depressing that their tech has stagnated to the point that they are only building things that run on the 360 and (occasionally, if the stars are aligned correctly) PS3. This isn't restricted to valve, of course; I can't recall a PC game since Crysis that has pushed the platform. There would be a few, but they're a dying breed.

The real question is, did any of the art resources get updated between HL2 and HL2:E2? Because if they're still using the exact same models and textures for the grabbity gun, it's not going to appear to improve much. Unless there were certain engine features you were hoping for that you missed in the game?

[Vympel]

Yeah yeah, gameplay over graphics, blah blah blah. I don't give a shit, as far as I'm concerned the Source engine simply isn't good enough nowadays - games like Left 4 Dead and what not are possible because the Source engine is so ancient that any computer can run Left 4 Dead's content and not be taxed in the least, and thats great- for Left 4 Dead.

However, remember when Half Life 2 came out and it looked amazing? Yeah, I want that again. I don't want to wait years and years for a few hours of content produced on an ancient engine. I could get another Crysis in that time- oh wait, I am.

Note I never restricted my comments to graphics either. Don't get me wrong, I like that game advancement has started to slow - how else to explain my 10/2006 PC still runs all games perfectly at max settings in 3/2010? But that doesn't mean it has to stop, and Half Life 2 has stopped since it came out. There's been no noticeable improvement whatsoever.

[JointStrikeFighter]

Borderlands has procedural generation of guns.

[Starglider]

The real question is, did any of the art resources get updated between HL2 and HL2:E2?

Yes, the polycount and texture resolution was bumped, and all the shaders were rewritten for HDR. Actually the techniques used to make Lost Coast / Orange Box HDR run on the Xbox and low-end systems (at good frame rate) were quite ground-breaking at the time, though of course it's old hat now.

However, remember when Half Life 2 came out and it looked amazing? Yeah, I want that again. I don't want to wait years and years for a few hours of content produced on an ancient engine. I could get another Crysis in that time- oh wait, I am.

Are you still going to buy the game if it doesn't have Crysis 2 level eye-candy? If yes, why should Valve spend money making the game look awesome for the 2% of the market who have high-end gaming PCs?

You'll get thoroughly revamped graphics in mainstream games when there's a new console generation. It just isn't cost effective to support too big an asset quality gap between the console and PC platforms. In the mean time, if you want Crysis 2, just buy Crysis 2.

[Vympel]

Are you still going to buy the game if it doesn't have Crysis 2 level eye-candy? If yes, why should Valve spend money making the game look awesome for the 2% of the market who have high-end gaming PCs?

Depends on price. I'll pay $10 for it, max. Otherwise its not worth my cash. Its DLC level content, IMO.

[Starglider]

Depends on price. I'll pay $10 for it, max. Otherwise its not worth my cash. Its DLC level content, IMO.

If it's the same length as episode 2 then yes, it'd be on a par with say Broken Steel, but I get the impression that it will be significantly longer.