Halo 3 engine upgrade analysis (for ODST)

[Iron Tiger]

Personally, I never understood all the flak about human faces in Halo. Sure they look like ass, but who cares? You spend a lot more time looking at and shooting aliens, which do look pretty good. Halo is about the combat first and foremost, and I'm glad Bungie spent more time on gameplay than worrying about human face modeling.

Now animation is another story. I agree they could use some help there.

They spend enough time and money on getting high profile voice actors, they could at least pair those voices with decent-looking models. Every Halo game from Bungie has had subpar face modelling. Can't blame the hardware for it when most every other game is doing better with the same limitations. Halo 3's faces look bad compared to even last gen stuff.

 

[NRP]

Nevermind

 

[JardeL]

No, no, this is completely misquoted. The problem was never an OS overhead - this has always been and always will be 32Mb.
What Bungie are saying is that development kits don't have any additional memory, so during development some of the memory is reserved for the game to be able to go through proper testing (asserts, memory tracking for leaks etc). If you disable such test features you can use the spare memory but you can no longer easily test the game which is obviously very dangerous.
On my last project the amount of memory we had to sacrifice was around 20Mb. With enough effort it is possible to find good safe use for it, but I suspect not many developers have the time to deal with this so this memory ends up being unused in the shipping game.
This is a no-issue now since MS announced they will have devkits with extra memory (1Gb instead of 512Mb) which should solve all these issues.

What about PS3 devkits ?.. How much memory those "test features" reserve ?..

 

[Boxercide]

What about PS3 devkits ?.. How much memory those "test features" reserve ?..

How is a PS3 devkit relevant in a Halo discussion?

 

[deepbrown]

May I ask...how many threads are open to Xbox 360 developers. I heard only 3, with 2 turned off. Is this true?

 

[liolio]

OT
I don't know how many RAM is reserved for profiling tools on ps3 devkits but it's not relevant if my memory serves me well ps3 devkit had 1GB of xdram from scratch.
On the 360 I think that one thread of the core 2 is devoted to the OS, thus they have access to 5 thread on 3 cores. It's less for indies/XNA community games I remember reading that they can only acces two cores.
/OT

 

[AlNom]

XNA has different restrictions. The rest should be under NDA IIRC. It's not as though the OS would take up a significant chunk of CPU time anyway.

 

[Malibu]

On the 360 I think that one thread of the core 2 is devoted to the OS, thus they have access to 5 thread on 3 cores. It's less for indies/XNA community games I remember reading that they can only acces two cores.

It was said back 05 that:

http://www.xbox360news.com/Blogs/News/hqs/blr_483.aspx

5% computation usage of both core 1 and core 2 are reserved by the system.

 

[assen]

OT
I don't know how many RAM is reserved for profiling tools on ps3 devkits but it's not relevant if my memory serves me well ps3 devkit had 1GB of xdram from scratch.
On the 360 I think that one thread of the core 2 is devoted to the OS, thus they have access to 5 thread on 3 cores. It's less for indies/XNA community games I remember reading that they can only acces two cores.
/OT

No. A tiny slice of the CPU time on one thread on one core is devote to the OS. For all intents and purposes, you have the entire CPU for you. The CPU over the course of a frame will spend much more on cache misses and other obvious and non-obvious penalties (it's VERY good at that), than on the slice allocated for the OS.

To put it another way, your own incompetence will take from the CPU much more than the OS-reserved time.

 

[assen]

Bungie, who have shown the wisdom to extend and improve the same codebase through the years (one of the talks by Mat Noguchi mentioned parts of it can be traced to Pathways into Darkness, the game before Marathon), are not likely to do "a new engine", as in clean-slate rewrite.

Also, re: the constant talk of "anisotropic as major feature they need". Anisotropic filtering is not an "engine feature". It's something that will cost them very little resources in terms of development, but probably quite a lot in terms of runtime performance; if they peg anisotropic as a topmost priority, as some of you seem to ask, they'll need to sacrifice other things.

In some ways, having good anisotropic in Reach would mean they ran out of ideas what to do with all that free frame time, and decided to let the texture fetchers feast on bandwidth.

 

[Acert93]

Bungie, who have shown the wisdom to extend and improve the same codebase through the years (one of the talks by Mat Noguchi mentioned parts of it can be traced to Pathways into Darkness, the game before Marathon), are not likely to do "a new engine", as in clean-slate rewrite.

Also, re: the constant talk of "anisotropic as major feature they need". Anisotropic filtering is not an "engine feature". It's something that will cost them very little resources in terms of development, but probably quite a lot in terms of runtime performance; if they peg anisotropic as a topmost priority, as some of you seem to ask, they'll need to sacrifice other things.

In some ways, having good anisotropic in Reach would mean they ran out of ideas what to do with all that free frame time, and decided to let the texture fetchers feast on bandwidth.

Their assets often look like CRAP in-game because you cannot even see the textures properly--they are a blurry mess. If they need to cut out some resolution or other effect to clean up their image than so be it because it is a MAJOR eye-sore.

The game has large, open areas of flat textures ground as well as huge "artifacts" that have large, flat, and smooth surfaces that lose all sorts of details because of the lack of AF.

Their art design clashes dramatically with the IQ decisions they made and it makes the "clean" look they are looking for look like a mistake at times as you go from some beautful scenes to blurry/washed out ones.

They really have two options: show off all their flat landscapes and architectures with improved texturing or address their poor technical/art decision and change their art and effects to minimize the aversion they show toward texture filtering.

FM2 is a prime example of a game with really poor texture filtering that turned detaied textures into a complete mess. FM3 has addressed this issue while at the same time improving texture detail--so it is possible to have your cake and your icecream. The problem is probably what you hinted at: Bungie recycles a lot (I will say what Al won't: I swear some of the human models in Halo 3 have a Halo 2 poly budget!!) and are focused on game features and experience (which they excell at) but that isn't necessarily a free pass for such poor mistakes.

Another area Halo 3 had some issues is it really seemed like the engine (from AI to graphic load and design) was initially aiming for large and open "sandbox" gameplay where the player had a lot more control over their path. ODST shows the engin is quite apt at such. For whatever reason, Bungie indicating this was the goal at one point, went with a more confined/linear approach. It could be gameplay related/pacing, but the presences of these sandboxes indicates it was possible. I think one of the general issues is enemy diversity as well as group size. While Halo 3 has a lot more enemies with complex AI than most shooters, it does seem the engine was not refined enough to really "open up" much larger and more frequent sandboxes hence the more linear gameplay paths. Of course more freedom costs more $, time, and compromises control.

 

[NRP]

It's something that will cost them very little resources in terms of development, but probably quite a lot in terms of runtime performance; if they peg anisotropic as a topmost priority, as some of you seem to ask, they'll need to sacrifice other things.

So is (good quality) AF really that expensive? I'd really like to hear more discussion about this.

 

[zed]

depends on various things but prolly about 15-20% hit

as some of you seem to ask, they'll need to sacrifice other things.

well first thing to go would be there HDR method
get rid of that + I woulnt be surprised if they could do 720p @ 2xAA + 8xAF

I thought Ild already written this but cant see it, heres my take
someone high up at bungie thought of this HDR method + didnt want to ditch it, even though its proven bad.
we've seen countless examples of others keeping with there own methods even when proven to be worse or wrong (eg reading a book on W bush at the moment, the invasion of iraq is a good example) ego's sometimes dont lead to the best decision

 

[assen]

If they lose the HDR solution, they lose the single best thing about their visuals, IMHO.

I basically agree with Joshua Luna, but I'll phrase it differently: their engine doesn't handle high-poly scenes. Why is that, I can't guess.
Adding AF would make things only slightly better, details wont' magically add details everywhere.

AF is costly because the texture sampler has to fetch from the texture several times; it might be free if you're heavily bound elsewhere, or it can be very expensive if your bottleneck is texture samplers and/or bandwidth in this part of the frame.

 

[Laa-Yosh]

Yeah, HDR has to stay.

As for less sandbox, I think the engine is quite capable of it, so it was more about a lack of game design resources and that they did not dare to reinvent the wheel in such an important release. And we have to admit that their approach worked, seeing how Halo3 is still selling better week by week than you-know-what.

ODST and Reach are however great opportunities. As for Reach I'm thinking a 4-player coop oriented game is a no-brainer. Everyone can do 2-player coop now, so Bungie has to up them
Even in single player you'd get 3 NPCs on your side... and there'd have to be a lot of enemies to challenge four Spartans. So the engine would require more work on scale and friendly AI.

 

[NRP]

I PM'd urk, the community guy at Bungie, about any possible visual upgrades to ODST. Here is what he said:

Halo 3: ODST is built on top of the existing Halo 3 tech. The team's focus was story, content, and gameplay.

So there you have it. Those wanting brand new rendering tech will have to wait for Reach.

 

[ShadowWolf64]

I PM'd urk, the community guy at Bungie, about any possible visual upgrades to ODST. Here is what he said:



So there you have it. Those wanting brand new rendering tech will have to wait for Reach.

ODST's lighting will probably look better then Halo 3's since Bungie had time to refine the Halo 3 tech for ODST.I remember Bungie and MS hired Corrine Yu, which is supposedly a programming genius in helping Bungie make an game engine from the ground up.

Also, do any of you think that Halo Reach will incorporate multi-core rendering?Im asking, because at this years GDC, Bungie was talking about the Zen of Multicore rendering and perhaps they are using that for Halo Reach's development.

 

[Rangers]

ODST's lighting will probably look better then Halo 3's since Bungie had time to refine the Halo 3 tech for ODST.I remember Bungie and MS hired Corrine Yu, which is supposedly a programming genius in helping Bungie make an game engine from the ground up.

Also, do any of you think that Halo Reach will incorporate multi-core rendering?Im asking, because at this years GDC, Bungie was talking about the Zen of Multicore rendering and perhaps they are using that for Halo Reach's development.

Corrine Yu is on the separate MS (not Bungie) Halo team though I think.

Possibly working on Halo 4 presumably on the next gen box I suppose. Basically where MS separates Halo from Bungie as is eventually necessary when Bungie left.

There's been some clarification of the incredibly cloudy halo dev team picture with the reveals of ODST and more recently Reach, but much still remains cloudy, such as the status of the Peter Jackson Halo project.

 

[Laa-Yosh]

Corrine Yu is on the separate MS (not Bungie) Halo team though I think.

AFAIK her husband is Kenneth Scott, ex-id Software art director and super talented monster creator... also on the Halo team?

 

[ShadowWolf64]

Corrine Yu is on the separate MS (not Bungie) Halo team though I think.

Possibly working on Halo 4 presumably on the next gen box I suppose. Basically where MS separates Halo from Bungie as is eventually necessary when Bungie left.

There's been some clarification of the incredibly cloudy halo dev team picture with the reveals of ODST and more recently Reach, but much still remains cloudy, such as the status of the Peter Jackson Halo project.

I think her main project is the Peter Jackson Halo game but she could be also assisting Bungie in the development of an all new engine for Halo Reach.In my opinion, I believe Halo Reach will end up being a graphically amazing game when its released.I think Bungie themselves is a bit disappointed with how Halo 3 ended up looking in terms of graphical fidelity.Because they know the game was graphically inconsistent and im certain Bungie is not going to let Reach suffer from that problem like Halo 3 did.

Don't get me wrong, I love the lighting in Halo 3 but Bungie needs to work on a different method that yieleds the same impressive results or better then they're current HDR method allows for.