View previous topic :: View next topic |
Author |
Message |
GNUtoo Veteran
Joined: 05 May 2005 Posts: 1919
|
Posted: Fri Jul 15, 2005 9:54 pm Post subject: Clustering for games |
|
|
Is there any clustering progrrams that use open gl and graphic card that are suitable for games?
(and if so is it aplicalble only to open source games as nexuiz or also to others) |
|
Back to top |
|
|
HPRichard Tux's lil' helper
Joined: 02 Feb 2005 Posts: 96 Location: KA, Germany
|
Posted: Sat Jul 16, 2005 10:23 am Post subject: |
|
|
What exactly do you want to do with clustering?
Do you want to use multiple machines for graphics output for being able to use some more screens than you can use with one computer? (A kind of SuperTiling across multiple Machines like ATI with their CrossFire solution)
Do you want to use multiple Computers because the computing burden for the game is too high for a single machine?
Do you want some machines to do precomputations on theie GPUs for the GPU of the output machine not having to do that much, as done here http://gamma.cs.unc.edu/switch/
or what is your goal of "clustering"?
Clustering can only work if you can parallelize the work to be done with not-that-much communication between the different computation tasks. One possibility could be to offload the physics calculations to another machine with an idling physics processor like this one http://www.ageia.com/ or with an idling GPU if someone wrote GPU code for that calculations. For using other/multiple GPUs you may also want lo look at http://www.gpgpu.org/ where you can get an Idea what you can do with a second GPU not occupied with drawing monsters.
So the question remains: What kind of clustering do you think of? |
|
Back to top |
|
|
lifeform n00b
Joined: 05 May 2004 Posts: 25
|
Posted: Sun Aug 14, 2005 11:52 am Post subject: |
|
|
i'm playing some games using cedega.
the refresh rate n response seems slightly sluggish as compared to a pure windows environment.
was thinking of clustering to improve the performance.
what kinda clustering shld i b looking at? |
|
Back to top |
|
|
Jeremy_Z l33t
Joined: 05 Apr 2004 Posts: 671 Location: Shanghai
|
Posted: Sun Aug 14, 2005 12:42 pm Post subject: |
|
|
lifeform wrote: | i'm playing some games using cedega.
the refresh rate n response seems slightly sluggish as compared to a pure windows environment.
was thinking of clustering to improve the performance.
what kinda clustering shld i b looking at? |
You can forget about that.
You cannot do clustering for things not previously meant for that. That means you would have to write it in the code. _________________ "Because two groups of consumers drive the absolute high end of home computing: the gamers and the porn surfers." /.
My gentoo projects, Kelogviewer and a QT4 gui for etc-proposals |
|
Back to top |
|
|
HPRichard Tux's lil' helper
Joined: 02 Feb 2005 Posts: 96 Location: KA, Germany
|
Posted: Sun Aug 14, 2005 8:50 pm Post subject: |
|
|
Clustering for making an existing Windows game run faster is near to impossible. Most Windows games do not make use of more than one CPU (with the exception of Quake 3, which got a slight improvement from a second CPU). So clustering is not benefical as the work cannot be distributed among several machines. The only theoretical benefit could be that you have a faster machine/CPU in your cluster and thet by chance the process of the game is moved over to that particular CPU. At that point you would do better playing on that machine (if it has a halfway decent graphics card) as all the 3D-API calls have to make the way from that CPU to your graphics card, which has lots of overhead and is probably slower than playing on your own machine.
Clustering can only be benefical if you have the possibility to prallelize your problem, for example when using a software renderer, wherer the game can run on one machine, and the software renderer on the other one, but only if the architecture of the rendering API allows this.
The game is completely different if you develop a new game and have the nerves of writing massively parallelized code, that is a case where clustering comes to being useful. |
|
Back to top |
|
|
Redhatter Retired Dev
Joined: 20 Sep 2003 Posts: 548 Location: Brisbane, QLD, Australia
|
Posted: Thu Aug 18, 2005 11:56 am Post subject: |
|
|
To steal a post on another forum...
[playerone] wrote: | People seem to think that a beowulf cluster == awesome super good games and server OMGFBIBBQ!!!1!!11!
Supercomputer Clusters are not designed for everyday desktop/server operation. They do fairly specific things that arent really usefull to johnny average user.
If you read the article you see that it explains about this a bit saying something along the lines of what it does is distribute the work around. If your work isnt designed to be distributed then it does..... bugger all.
Supercomputer Clusters are designed for applications that do a lot of little bits (or big bits) like rendering a frame, computing the effect of THIS particle on THAT one, or similar things. |
I think that pretty much summs up what I have to say on the topic... Quite simply, games are just not designed to run on clusters... nor are most desktop apps. They're built, with the sole intention on running on a single system, with one or more CPUs sharing the same memory space and peripherals. They're not designed, to run, distributed across 20 different CPUs, each with it's own RAM and peripherals attached.
Not to mention, I think you'll find the local bus that hooks CPU to RAM & peripherals, MUCH faster, than any form of computer networking interface. Games don't handle these sorts of delays very well.
Don't get me wrong, they have their place... for large compile jobs, they're a dream to use -- I'm already looking into how I can coax my Indy and Indigo2 Impact to talk distcc (although I've gotta watch it, one's o32, the other, n32, and the I2I requires special patches to work). OpenMOSIX isn't an option for my needs, nor are most other clustering implementations, but they certainly do have their place. However, their place is not for games. _________________ Stuart Longland (a.k.a Redhatter, VK4MSL)
I haven't lost my mind - it's backed up on a tape somewhere...
Gentoo/MIPS Cobalt developer, Mozilla herd member. |
|
Back to top |
|
|
Jeremy_Z l33t
Joined: 05 Apr 2004 Posts: 671 Location: Shanghai
|
Posted: Thu Aug 18, 2005 12:02 pm Post subject: |
|
|
Well, distributed rendering works well for realtime 3D, but it needs every machine to be able to render (not 133 with 10 years old video card), and does require support in the code (and/or hardware support). _________________ "Because two groups of consumers drive the absolute high end of home computing: the gamers and the porn surfers." /.
My gentoo projects, Kelogviewer and a QT4 gui for etc-proposals |
|
Back to top |
|
|
tHe_IkkE n00b
Joined: 05 Mar 2005 Posts: 5 Location: Den Bosch, The Netherlands
|
Posted: Thu Aug 18, 2005 1:27 pm Post subject: |
|
|
The only "real" use for multiple computers and gaming I can think of is running a multiplayer game and then let other computers connect to "simulate" another player so the AI of your bots can be balanced.
You'd have to write your own bot i think.... |
|
Back to top |
|
|
brendaniabbatis n00b
Joined: 25 Nov 2003 Posts: 54 Location: Sacramento, CA, USA
|
Posted: Mon Aug 22, 2005 7:07 pm Post subject: |
|
|
tHe_IkkE wrote: | The only "real" use for multiple computers and gaming I can think of is running a multiplayer game and then let other computers connect to "simulate" another player so the AI of your bots can be balanced.
You'd have to write your own bot i think.... |
That would be cool! The next level would be to run the game entirely with networked bots. You could have all the fun of strategizing and beating your opponents without actually wasting time playing the game yourself! |
|
Back to top |
|
|
Redhatter Retired Dev
Joined: 20 Sep 2003 Posts: 548 Location: Brisbane, QLD, Australia
|
Posted: Tue Aug 23, 2005 8:32 am Post subject: |
|
|
brendaniabbatis wrote: | tHe_IkkE wrote: | The only "real" use for multiple computers and gaming I can think of is running a multiplayer game and then let other computers connect to "simulate" another player so the AI of your bots can be balanced.
You'd have to write your own bot i think.... |
That would be cool! The next level would be to run the game entirely with networked bots. You could have all the fun of strategizing and beating your opponents without actually wasting time playing the game yourself! |
Heh, for once, the luddites on 2400 baud modems have a chance to finally 0wn those 1337 gamers on cable. _________________ Stuart Longland (a.k.a Redhatter, VK4MSL)
I haven't lost my mind - it's backed up on a tape somewhere...
Gentoo/MIPS Cobalt developer, Mozilla herd member. |
|
Back to top |
|
|
nokesc n00b
Joined: 08 Sep 2006 Posts: 1
|
Posted: Fri Sep 08, 2006 8:26 pm Post subject: |
|
|
Check this out ...
http://www.intel.com/technology/itj/2005/volume09issue02/art01_ray_tracing/p01_abstract.htm
Along with multiple core cpu's, clusters could take advantage of games based on ray tracing rather than polygons ... games like wolfenstien and Doom used a similar technique called ray casting and heightmapping. With the much faster speeds of today's processors this could be taken a step further ... imagining a game running @ 640x480 using ray tracing, awesome eye candy! |
|
Back to top |
|
|
renrutal Tux's lil' helper
Joined: 26 Mar 2005 Posts: 135 Location: Brazil
|
Posted: Sun Sep 10, 2006 7:06 am Post subject: |
|
|
nokesc wrote: | Check this out ...
http://www.intel.com/technology/itj/2005/volume09issue02/art01_ray_tracing/p01_abstract.htm
Along with multiple core cpu's, clusters could take advantage of games based on ray tracing rather than polygons ... games like wolfenstien and Doom used a similar technique called ray casting and heightmapping. With the much faster speeds of today's processors this could be taken a step further ... imagining a game running @ 640x480 using ray tracing, awesome eye candy! |
Wake me up when cheap computer systems are able to ray trace a multi-layered multi-million polygon image faster than one frame every 18 hours.
Umm, but considering how much the GPUs have evolved in the last years, running a kernel on them to do that work really fast isn't really a far fetched thought. |
|
Back to top |
|
|
|