Don't really agree with this. What's Lair doing with armies that other systems can't handle? Kameo should a couple thousand clones with cloned animations, which is all Lair appears to be.
SPU usage in games
- Thread starter betan
- Start date
Probably they are talking about efforts to make teams of armies act independently.
Many games showed hordes of enemies on screen before anyways, like Dynasty Warriors.
I am not sure how well LAIR is doing it. Perhaps its the amount of enemies calculated in both huge distance and foreground while doing different things in teams and all rendered simultaneously
I think that the game that showcases the best the SPU's ability to calculate huge amounts of enemies independently is HS.
Many games showed hordes of enemies on screen before anyways, like Dynasty Warriors.
I am not sure how well LAIR is doing it. Perhaps its the amount of enemies calculated in both huge distance and foreground while doing different things in teams and all rendered simultaneously
I think that the game that showcases the best the SPU's ability to calculate huge amounts of enemies independently is HS.
archangelmorph
Veteran
Yeah but optimisation allows you to look at this kind of scenario and say "ok well we have all this idle time available at period intervals perframe.. so what can we do with it?"
Maybe setup a s seperate set of software threads which spend some CPU time processing some other sub-system (e.g. cloud simulation for example) and then get the SPu to check if there's other work that's available when it's idling..
The size of each interval and the frequency of them per frame can determine to some degree the amount of extra work that can be done but so balancing and reshuffling can help to make room for whatever you need sometimes..
In short, A full blown production level game engine would work as much as possible to squeeze as much as possible, the work onto each clock cycle.. The only event this doesn't happen is when all the sub-systems (you planned, designed and accounted for..) are in place and your engine is running as planned.. At this point your not going to go back and say "ok so what else can we cram into this thing to use up more cycles?" because it does everything you intended and in reality you probably don't have the time/budget available to do it (especially if the added feature would remain unused since the development schedule doesn't allow enough art time to support it in the game)..
http://ps3.ign.com/articles/794/794165p2.html
Above is in regard to StardustHD
Killzone 2 quote:
http://www.killzoneunit.com/kz/?p=647
Last edited by a moderator:
Sure, in fact the Lots of Ducks demo was used as an example of squeezing more out of fewer SPUs via a management system (might have been spurs) rather than spreading across more SPUs and idling them for longer.
But if you've no plans to use you're freed up SPU power, you might as well leave things as is.
I don't know about the latter bits..working on different screen areas and general load balancing, but Alan Heirich was/is doing work for SCEA on load balancing pixel shading across Cell and RSX in a deferred renderer. I'm sure you remember that though..can't remember if the paper was published or not.
CELL could help RSX in rendering gbuffers removing primitives that don't contribute to the final image, moreover CELL could share with RSX the burden of rendering a frame via full screen passes (remember that is a deferred renderer)What bennefits and gains in performance would Cell bring to different parts of the rendering engine?Gains in pixel?, vertex?, shadows? or All? And how easy would that split rendering be to implement?
This threads like a reference libary of all SPE usage, sweet
SPE's are vertex beasts...
LOL! I don't remember that! I'm really getting old...
Anyway..if CELL could also take care of full screen effects then one might overlarp CELL and RSX rendering times with CELL doing full screen effects while RSX is generating next frame g buffers,shadow maps and occlusion terms, at cost of having one sync point between the 2 processors.
If we also have transparent stuff to render then we probably need to sync processors a couple of times per frame.
F5's comment is (has always been) controversial. I think they are refering to the entire army of little men, wooden boats, flying dragons and gigantic monsters (on both sides) all at the same time; and their approach to the problem. The larger monsters should be complex to process (and I'm curious about their geometry work). As for AI, the little men seem repetitive, but I don't know how well the bosses behave.
Last edited by a moderator:
totally agree
but I guess it can make sense for some gamesHey, NNN has up to 1-2k enemies on screen, with induvidual AI. Shitty game, but still, i agree with you, Lair hasn't shown anything that would say only can be done with the cell
What about the sheer number of on screen objects and there AI and the culling its doing aswel? Surely that takes some serious hourse power?
Here's a link to the abstract if you're interest..
http://www.aurorasoft.net/symposiums/ccs06/presentations.php
Re. Lair and games with 'clones', they're difficult to compare without knowing how much processing is going on per 'clone', so to speak. Although I thought I read Lair has armies of up to 10,000 soldiers...but again, that's difficult to really compare against. I'd hazard to say anything about "only possible on Cell" at this point...it'll happen over time, no doubt, but I think we'll have to see more of what this generation has to offer before we can draw a line with a good degree of confidence.
He says sometimes AI can take up to a SPU implying most of the time it does not. So yes it looks like AI part of Lair is certainly doable in other systems. Many individual SPU tasks should be doable in other systems (possibly even easier).
Coincidentally, Insomniac was asked this question in an interview GameDaily has just put up. They sort of wonder about the merit of sticking a single number on 'usage'.
http://biz.gamedaily.com/industry/feature/?id=16729&page=1
http://biz.gamedaily.com/industry/feature/?id=16729&page=1
I can understand Shifty and Ostepop's points. The issue is with F5's statements. They are not detailed/exact enough to articulate the real meat. I am hoping the real game will also include "Behind the Scene" and "Developer Interviews" to elaborate on the finer details.
These guys presented 2 sessions in SIGGRAPH. So it would be great if they can help mortals like me understand their work better.
With the latest SpiderWasp screenshot they released, I am ever more anxious to see the entire game in action. I agree with the notion that Lair is impressive because it can handle so many details all at the same time at good framerate even though individually they are doable on existing systems.
Similar threads
- Locked