Message boards : Number crunching : Multi-thread models...
Message board moderation
Author | Message |
---|---|
Send message Joined: 5 Aug 04 Posts: 186 Credit: 1,612,182 RAC: 0 |
Heh, how do I set big blue text on this messageboard..? ;-) Pleeeeease come back phpBB, all has long since been forgiven. :lol: +++ Now that simultaneous mutlthread capable machines are becoming the norm', eg. Intel Hyperthreading P4s, and the future of maintaining 'Moore's law' for desktop machines appears to be multi-core CPUs - how about looking at how easy it would be to run a single instance of a model with multiple threads..? This modelling software was originally designed to run on a multicore Cray, so it would actually involve changing things backward some way, wouldn't it..? The CP-boinc software needs to evolve in the same direction as standard desktop machines methinks, especially if you desire to increase the computation load........ We will eventually want 64 bit software too but M$ needs to catch up first I guess. ;-) <a href="http://www.nmvs.dsl.pipex.com/"><img src="http://boinc.mundayweb.com/cpdn/stats.php?userID=6&team=off&trans=off"></a> <a href="http://www.nmvs.dsl.pipex.com/">Distributed Mania</a> |
Send message Joined: 5 Aug 04 Posts: 390 Credit: 2,475,242 RAC: 0 |
I agree that it would be much better to have one model processing faster (with less memory and HD space requirements) than having two, four of even more. I can also imagine that sulphur cycle/couple model keeps CP team bussy enough to put this issue a bit behind. Once SC and CM are running, i would be happy to play with multithreaded 64-bit CP version - having model done on fast Presshot in less than 10 days sounds reasonable. In the meanwhile, M$ is promissed [again] to put Win XP 64-bit edition on market. I hope that free upgrade from Win XP Pro to 64-bit edition still apply... |
Send message Joined: 7 Aug 04 Posts: 2187 Credit: 64,822,615 RAC: 5,275 |
> > Now that simultaneous mutlthread capable machines are becoming the norm', eg. > Intel Hyperthreading P4s, and the future of maintaining 'Moore's law' for > desktop machines appears to be multi-core CPUs - how about looking at how easy > it would be to run a single instance of a model with multiple threads..? This > modelling software was originally designed to run on a multicore Cray, so it > would actually involve changing things backward some way, wouldn't it..? The > CP-boinc software needs to eveolve in the same direction as standard desktop > machines methinks, especially if you desire to increase the computation > load........ > I think this is absolutely correct. Without CPU speeds ramping up like they have in the past, performance gains to run lots of models in the future on more computationally intense experiments will have to come through parallelization and/or 64 bit and additional optimizations for AMD CPUs. Otherwise, the longer work units will take forever, taxing the patience of a lot of home modellers, and maybe even resulting in a smaller percentage of completed WUs as random problems/incompatibilities/instabilities creep in during the longer model runs. I'm sure the investigators of this experiment have been thinking of these issues as well. It will be interesting to hear of their thoughts on the subject. > > We will eventually want 64 bit software too but M$ needs to catch up first I > guess. ;-) > Last I heard was late March or April for the official release of Windows 64bit. Of course Microsoft has never slipped a planned release date ;-) I believe that it will be a free/cheap upgrade for those 64bit capable systems running WinXP Pro (not Home). Can't remember where I read that though. |
©2024 cpdn.org