A question concerning hardware. I am thinking of gettinga Dual Xeon quad core setup with 16 G ram for some serious analysis but I have concerns about version 11 on XP64. Currently 11p5 does not really work on XP32, transient runs don't write results, parallel does not work etc (have not tried the XP64 variant but I expect similar results). I have until the end of Feb 2007 to do my work for a contract I have and cannot count on XP64 being ready in time for my work load, hence the following question.

Can I run a cluster of boxes running XP32 over distributed Parallel mode and be able to {exceed} the memory limitations of XP32. eg if I run a simulation of 16+ million elements over say 4 dual core XP32 boxes, each with 4 Gig ram, will the distributed parallel option effectively work around the 4 Gig memory limit to enable me to solve large problems in XP32?

My personal time line factors aside, am I better off buying an 8 core XP64 box with 16G ram, or 4 dual core XP32 boxes with 4G ram each? What would be the diference in procesing speed etc.

Linux is out of the question as I don't understand it, and cannot be bothered learning another OS just because XP64 is late.

Thanks Justin

"Can I run a cluster of boxes running XP32 over distributed Parallel mode and be able to {exceed} the memory limitations of XP32. eg if I run a simulation of 16+ million elements over say 4 dual core XP32 boxes, each with 4 Gig ram, will the distributed parallel option effectively work around the 4 Gig memory limit to enable me to solve large problems in XP32?"

Yes.

"My personal time line factors aside, am I better off buying an 8 core XP64 box with 16G ram, or 4 dual core XP32 boxes with 4G ram each? What would be the diference in procesing speed etc." A linux cluster (for parrallel performance reasons) would be ideal using the intel quad core Q6600 procs being released in January.

If you insist on windows (silly but your choice, its no harder to use than windows and it handles memory properly unlike windows) then the above two options would likely have negligable speed differences. Hence I would go for the 8 core single box (which could be doubled in future for a 2 box cluster with high speed interconnects etc)

You are being overly greedy on the memory side. CFX is CPU limited. A 5E6 cell problem (ie. 4 GB of ram) on a single quad core chip still takes about 100 seconds / iteration. 512 MB of RAM per core is more than enough. 2 GB per core is a waste ... spend the money on more CPUs.

Can you run distributed parallel, either PVM or MPICH over two machines where one is XP32 running CFX10 for XP32, and the other is on XP64 running CFX10 for XP32?

Will the rsh 32 bit cfx 10 windows variant run on a XP64 box?

Thanks Justin

Sorry I forgot something. Can the account you use for rsh (either PVM or MPICH) have a blank password? Would this cause problems

Thanks Justin

PVM definitely. MPICH probably not.

q1: It might work.

q2: yes.

why don't you ask your support rep for the cfx 11 pre-release which has all the p5 bugs you complained about fixed.

q1: yes. q2: no.