bit-tech.net | Multi-core in the Source Engine

Tim S 2nd November 2006, 13:18

Quote:

Originally Posted by Marquee
Good dam if my single core computer can't pwn in Source no more, its going to cause me and allot of other gamers to switch to dual core. There is nothing wrong with that but trust me the old gamers wont switch with out a fight.

It would be a big mistake if Valve didn't allow you to run their games on single core processors. I think that there is no doubt that you'll still be able to play source-based games on single core processors, but there will be benefits for users with dual or quad-core processors.

Tim S 2nd November 2006, 13:19

Quote:

Originally Posted by DeX
The rain looks pretty crap to me. Maybe it would look better for real. It's not as if they're demonstrating something new. Particle effects have been around since the first 3D games.

Particle effects have been, but realistic particle effects haven't been seen in public yet. ;)

Joeymac 2nd November 2006, 13:32

Not a sausage on DX10. Nothing.
Doesn't the 8800 card come out next week or something?

Is it all so NDAed up to the point it can't even be mentioned?

Lazlow 2nd November 2006, 14:28

Quote:

Originally Posted by Joeymac
Not a sausage on DX10.

This is DX9.0c at its best. We wont see DX10 pushing itself this much straight away. OK it may look as good as, if not a lot better, but compare this to when DX9.0 was first released - there's a big difference.

Da Dego 2nd November 2006, 14:56

Quote:

Originally Posted by Joeymac
Not a sausage on DX10. Nothing.
Doesn't the 8800 card come out next week or something?

Is it all so NDAed up to the point it can't even be mentioned?

Valve's source is simply a DX9-based engine. There would be little benefit in redoing the whole thing to run DX10 at this moment in time, and even if they were doing that, it wouldn't be related to the discussion here. :D

Firehed 2nd November 2006, 15:43

That, plus DX10 isn't coming till Vista, in February (effectively... Jan30 technically), plus you need to buy the new system to get it going. People already have dual-core systems - Valve were just smart in making sure that the update can scale beyond "just" two cores.

Looks enticing... I really am going to need to bring a proper gaming machine back to school after Xmas break.

Lazarus Dark 2nd November 2006, 16:33

I think some software and game companies need to get off their but like these guys. We are finally getting to a point where hardware performance will only be held back by inefficient programming. cuddos to the source programmers. bring on the 64 cores!

aggies11 2nd November 2006, 16:51

Absolutely brilliant article. Best explanation of multi-core/threading and gaming I 've seen to date. In particular:

Quote:

"Threading is about making things work in parallel - in terms of Kentsfield, making four cores work together at the same time. However, games are by their very nature serial - each system is dependent on the output of the previous one."

That hits the problem right on the head.

Valve is tackling the problem, not by making current games faster, but by changing the way they do things, to make future games do more. Adding in more parallel capable stuff so that games become less serial.

It also goes to show you how absolutely massive the investment in time/effort (and thus $$$) developing with parallelization in mind is. Look at everything they've discussed. This simply isn't "lets to the Source Engine, but use threads". You seriously have to sit down, and think about your software design, where it can benefit, and be extra smart about it, because it's tricky.

Remember, it's never been about getting %100 Cpu util on 4 cores. That can be done by any monkey in any number of ways. It's about getting %100 *meaningfull* utilization, in ways that improve the game, and improve the experience for the gamer.

I'm very curious to see how it works out for Valve, and what the results are. If a seriously tech oriented group like themselves, after spending so much effort/focus, can't get meaningfull results, then the future doesn't bode well for multicore. However if they can pull it of, their success, will actually force the rest of the industry to follow in their footsteps.

Aggies

Rocket733 2nd November 2006, 18:50

Quote:

Originally Posted by aggies11
Absolutely brilliant article. Best explanation of multi-core/threading and gaming I 've seen to date. In particular:

That hits the problem right on the head.

Valve is tackling the problem, not by making current games faster, but by changing the way they do things, to make future games do more. Adding in more parallel capable stuff so that games become less serial.

It also goes to show you how absolutely massive the investment in time/effort (and thus $$$) developing with parallelization in mind is. Look at everything they've discussed. This simply isn't "lets to the Source Engine, but use threads". You seriously have to sit down, and think about your software design, where it can benefit, and be extra smart about it, because it's tricky.

Remember, it's never been about getting %100 Cpu util on 4 cores. That can be done by any monkey in any number of ways. It's about getting %100 *meaningfull* utilization, in ways that improve the game, and improve the experience for the gamer.

I'm very curious to see how it works out for Valve, and what the results are. If a seriously tech oriented group like themselves, after spending so much effort/focus, can't get meaningfull results, then the future doesn't bode well for multicore. However if they can pull it of, their success, will actually force the rest of the industry to follow in their footsteps.

Aggies

Purely from a business standpoint you don't start on a project this big "hoping" its going to work out. Their initial testing must have shown that meaningful results are obtainable and realistic else they would not have progressed this far. At this point in time multicore has been advertised too heavily and implemented too far for it to be anything but successful, with all transitions there will be bumps but the problems will be solved when there are billions of dollars at stake.

CyberSol 2nd November 2006, 19:00

What a great write up! One of bit-techs best.
Makes me very excited! The rain looks better the real life rain.

what exactly is 1.2x%?

FIBRE+ 2nd November 2006, 19:58

Brilliant article and an interesting read too ;)

Hopefully this will scare Valves competition into pushing things forward a bit :)

.... Either that or the rest of them just wont bother making there own engine and buy theres from Valve :|

DougEdey 2nd November 2006, 21:10

Those videos were either really poorly encoded, or the door frames had the worst case of the jaggies ever!

aggies11 3rd November 2006, 04:20

Re: Rocket733

You haven't met valve :P Seriously though, I'd call their Development efforts artfull, amazing, meticulous, expertly crafted, everything and anything but "efficient". When you listen to the commentary of something like Episode 1, you see how much stuff they scraped simply because they weren't 100% satisfied. Completely developed and scripted events. Much like how a director cuts a film down in editing "this scene must go". Except in film it's rather trivial to shoot a scene, where in game development alot more work usually goes into it.

So I could easily see valve doing "research" into an area, spending a considerable amount of time and effort, and then abandoning it because it didn't work out :)

Will multicore be the case? Their initial results seem to say "no", it looks promising. But advancement involves research, which always contains a certain degree of risk. Business or not.

I have faith in the Valve guys though, if anyone can pull it off, it'l be them :)

Aggies

malccy 3rd November 2006, 12:31

Great article - thanks bit-tech.

Wish they had an exact date though for the update - oh well, at least it's coming.

Vergil_117 5th November 2006, 04:53

There is a reason why there wont be a DX10 render any time soon, that being VALVe will only support a tech that they seen is being widly support via their steam surverys and such.

BFGunrunner 6th November 2006, 01:59

DX 10 in source will be waste of time. If they were to do DX 10 in source they might aswell go to source 2 or what ever there going to name there new enigine.

Vergil_117 7th November 2006, 23:40

http://upload.wikimedia.org/wikipedia/en/2/2f/Source_multicore_critter-avoid-danger.jpg

I just found this image on wiki, is there a video to go with this like the other two tech demos?

bufferzz 8th November 2006, 14:29

Quote:

Originally Posted by rupbert
I can only imagine what HL3 will look like.

Well take a look your self.

Half-Life 2: EP1 + EP2 + EP3, together = Half-Life 3 ;)

Bladestorm 9th November 2006, 01:12

Very interesting article :)

On DX10, they already have seperate codepaths in the source engine for DX8.1 and DX9, so I should think they will add a DX10 one when they feel the gains outweigh the investment it will take.

Cthippo 9th November 2006, 07:02

Quote:

Originally Posted by bufferzz
Well take a look your self.

Half-Life 2: EP1 + EP2 + EP3, together = Half-Life 3 ;)

No, HL3 will be on a new engine and I'm sure they're already working on one with more eye candy which will also take into account all the incremental changes such as the multicore support. The episodic content business model has it's limitations and one of thoise will be when they need to sell a big new, expensive to develop game engine.

Vergil_117 9th November 2006, 09:29

HL3 would be using Source V2 I bet.

bufferzz 23rd November 2006, 16:41

Quote:

Originally Posted by Cthippo
No, HL3 will be on a new engine and I'm sure they're already working on one with more eye candy ....

Yes they update the engine all the time and release it as they get done...

It seems Half-Life 3 is in fact going to be Half-Life 2: Episode's One through Three. While there is no direct answer from Gabe, he refers to the Episodes as Half-Life 3 in a select few areas.
http://www.halflife2.net/forums/showthread.php?t=107415

EP1+2+3 = HL3 !!! ;)

bufferzz 23rd November 2006, 16:42

Quote:

Originally Posted by Vergil_117
HL3 would be using Source V2 I bet.

Well it updates all the time...

ikra 30th January 2007, 10:54

no no... half life 3 will be for directx 11 i bet.. which will be...2012 i suppose? By then the source engine and dx10 would have gone through revisions... and im sure somewhere along the road they might revamp source for dx10.. or not at all.. they can still make spectacular effects using dx9 can't they?

TomH 1st February 2007, 18:35

Quote:

Originally Posted by ikra
no no... half life 3 will be for directx 11 i bet.. which will be...2012 i suppose? By then the source engine and dx10 would have gone through revisions... and im sure somewhere along the road they might revamp source for dx10.. or not at all.. they can still make spectacular effects using dx9 can't they?

Maybe DX10.1 for HL3 -- I'd hope that they release HL3 within the next 5 years personally (and not five years after DX11 is released.. If it's released in 2012, which was only your guess :p).

Remember that you need something to work with which is at least finalised in specificaiton, and with hardware to test it on.

Of course, it's been already stated that HL3 is referred to by Gabe, as HL2: Episodes 1-3. So if there is going to be another full Half Life release (maybe with the new engine, but episodic content is lurverly for Valve), it would probably be more appropriate to refer to it as Half-Life: Generation 3? :p

Though I think they'd have to make a big (read: foundation for following episodic content) release, in order to have the media make a big hoo-har about the 'new' game and the 'new' engine it's built on. Source 2.0.

So still Source-based, but you can't fault them: it is a pretty neat engine.

Log in

Multi-core in the Source Engine

Comments 26 to 51 of 57