tag:blogger.com,1999:blog-4212234230338648875.post5748285396568560206..comments2024-03-27T20:43:05.862-07:00Comments on The Eternal Universe: Modern Supercomputers vs. Simeon Denis PoissonJoseph Smidthttp://www.blogger.com/profile/02583891162785742138noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-4212234230338648875.post-47383113681207266552008-05-17T10:28:00.000-07:002008-05-17T10:28:00.000-07:00"Joe, the problem really lies in the fact that we ..."Joe, the problem really lies in the fact that we are solving the MHD equations"<BR/><BR/>Oh Nick, I solve equations like this on my personal Core-Duo laptop just as a quick test to see if my MPI libraries are working.<BR/><BR/>I'm kidding of course. Yeah, I can see why doing Poisson's equations on an MHD level could be tricky on as few as 16 processors.<BR/><BR/>(Why aren't you using Matlab? I thought as an undergraduate I was told Matlab was the best thing to learn for doing computational physics?) :)Joseph Smidthttps://www.blogger.com/profile/02583891162785742138noreply@blogger.comtag:blogger.com,1999:blog-4212234230338648875.post-31163567545825213242008-05-17T10:09:00.000-07:002008-05-17T10:09:00.000-07:00Joe, the problem really lies in the fact that we a...Joe, the problem really lies in the fact that we are solving the MHD equations which are all coupled and only one of them is a Poisson equation. Sixteen processors is fine for low resolution jobs, but when we want to push resolution up to something like 512^3 or higher, we really need hundreds of processors in order to be able to do all of the computation required at each point for each time step. <BR/><BR/>It would be very handy if we could somehow de-couple the Poisson equation from the rest of the equation, but that requires doing things like assuming that bulk velocity, temperature, and magnetic fields do not influence the pressure balance in the gas, which are terribly bad assumptions to make.<BR/><BR/>So for now, if we use hundreds or thousands of processors to solve the problem, we have to split the Poisson equation up into hundreds or thousands of pieces too.Anonymoushttps://www.blogger.com/profile/17367937708444729356noreply@blogger.comtag:blogger.com,1999:blog-4212234230338648875.post-40068053461187766602008-05-16T17:34:00.000-07:002008-05-16T17:34:00.000-07:00Nick, I agree having massively parallel communicat...Nick, I agree having massively parallel communication is a major issue.<BR/><BR/>However, I am confused. Can't you just solve Poisson's equation using 16 processors connected on the same network? Then you shouldn't have a communication problem right? Or solve it on the 4 processors that are on the same quad-core chip?<BR/><BR/>However, having the whole machine solve it at once would obviously be a nightmare.Joseph Smidthttps://www.blogger.com/profile/02583891162785742138noreply@blogger.com