Another thing to consider is memory bandwidth. Many calculations these days are memory bandwidth bound, not disk/CPU bound. If you get a single machine with lots of cores and lots of RAM this is more of an issue. The current generation of Intel CPUs (the Xeon 5500's) are better with their triple channel memory, but you will still run into this eventually. A few years ago we did some benchmarks of very simple parallel calculations (embarassingly parallel) with IPython. We found that with as few as 8 cores, our calculations stopped speeding up. We tracked it down to the memory bottleneck.
The other issue is that if you stick with dual quad core systems you can use the cheaper Xeon 5500 CPUs. If you go beyond that, you have to use the more expensive 7000 series CPUs, which increases your cost per core. Here is a great site for messing with configurations: www.thinkmate.com Obviously, as others have noted, having more hosts with fewer cores on each increases the sys-admin burden quite a bit :( I am in the process of purchasing a cluster myself in this same price range. My plan is to stick to dual socket systems with 4-6 cores in each socket (not sure if I will go Intel or AMD). It looks like I will be able to get as many as 72 cores with 3-6 GB RAM per core.....we'll see if I can handle the sys-admin on this... One final thing is that Intel is going to be coming out with a new 6 core CPU in the next few months. But, you could wait for faster hardware forever ;-) Sounds fun though! Cheers, Brian On Fri, Jan 22, 2010 at 11:52 AM, William Stein <wst...@gmail.com> wrote: > Hi, > > We are considering purchasing a new computer for the sage.math > cluster, which will act partly as a Sage notebook server. The > budget is about $20-30K (!). If you're a hardware lover, and have > been looking at what one could get for that much money (tons of RAM? > cores?) these days, let me know. We could just got 1 or 2 computers > like sage.math.washington.edu (i.e., 24 2.6Ghz cores, 128GB RAM, > etc.)... but maybe there is something new or about to come out that we > should be aware of? > > -- William > > -- > William Stein > Associate Professor of Mathematics > University of Washington > http://wstein.org > > -- > To post to this group, send an email to sage-devel@googlegroups.com > To unsubscribe from this group, send an email to > sage-devel+unsubscr...@googlegroups.com<sage-devel%2bunsubscr...@googlegroups.com> > For more options, visit this group at > http://groups.google.com/group/sage-devel > URL: http://www.sagemath.org > -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
