Hardware (CPU / RAM) for academic computing

SlicedBread · October 30, 2011, 06:19

I am trying to configure hardware specs for a research computer, possibly a cluster and would like to know what would be best the task at hand. This computer will be part of a lab and will be a server for at least the next 4 years. It would be nice to have it expandable as well.

Primary use : Ansys 13.0 CFD - CFX, Fluent, etc.
Mesh sizes : 10M+ for now, maybe 30M-40M in the future.
Preferred OS : Windows

I was looking at a single tower Dell T610 with dual Xeon X5675's and 96GB ram. However I was also wondering about a cluster of systems with an i7-2600K (most likely would have to be custom built) .

I have read posts here and the ANSYS parallel guide that it is not good practice to use all cores in a multicore machine because of memory bandwidth limitations. If possible can someone clarify that a little? Would there be little increase in using 3+3 cores in the Xeon vs. 6+6 or is just that it wont scale linearly?

abdul099 · November 5, 2011, 15:58

It depends on your systems (and the software, of course) how it scales with the number of cores. Usually it's not linear, but when using too much cores, it can slow down the solution process.
On my machine it's fine with 4+4 cores (but Star-CCM+), efficiency is less than using only 2 or 4 cores, but still good.

The i7 are usually faster than the Xeon machines, but can't handle that much memory. Also when connecting several machines, it's not just the memory bandwith but also network latency and bandwidth.

For 30 - 40M cells, you will not need 96GB memory. 64GB should be enough, maybe it also works with 48GB (but that's not sure).

October 30, 2011, 06:19	Hardware (CPU / RAM) for academic computing	#1
SlicedBread New Member Jeff Join Date: Feb 2011 Posts: 4 Rep Power: 15	I am trying to configure hardware specs for a research computer, possibly a cluster and would like to know what would be best the task at hand. This computer will be part of a lab and will be a server for at least the next 4 years. It would be nice to have it expandable as well. Primary use : Ansys 13.0 CFD - CFX, Fluent, etc. Mesh sizes : 10M+ for now, maybe 30M-40M in the future. Preferred OS : Windows I was looking at a single tower Dell T610 with dual Xeon X5675's and 96GB ram. However I was also wondering about a cluster of systems with an i7-2600K (most likely would have to be custom built) . I have read posts here and the ANSYS parallel guide that it is not good practice to use all cores in a multicore machine because of memory bandwidth limitations. If possible can someone clarify that a little? Would there be little increase in using 3+3 cores in the Xeon vs. 6+6 or is just that it wont scale linearly?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Superlinear speedup in OpenFOAM 13	msrinath80	OpenFOAM Running, Solving & CFD	18	March 3, 2015 05:36
OpenFOAM 13 Intel quadcore parallel results	msrinath80	OpenFOAM Running, Solving & CFD	13	February 5, 2008 05:26
Ram, cache and cpu upgrade help	zonexo	Main CFD Forum	14	January 24, 2007 09:05
Increasing RAM decreases CPU time!!!	Melih GULEREN	FLUENT	2	April 5, 2004 06:21
more RAM or faster CPU??	Fabrizio Grieco	Siemens	11	January 23, 2001 07:35

November 5, 2011, 15:58		#2
abdul099 Senior Member Join Date: Oct 2009 Location: Germany Posts: 636 Rep Power: 21	It depends on your systems (and the software, of course) how it scales with the number of cores. Usually it's not linear, but when using too much cores, it can slow down the solution process. On my machine it's fine with 4+4 cores (but Star-CCM+), efficiency is less than using only 2 or 4 cores, but still good. The i7 are usually faster than the Xeon machines, but can't handle that much memory. Also when connecting several machines, it's not just the memory bandwith but also network latency and bandwidth. For 30 - 40M cells, you will not need 96GB memory. 64GB should be enough, maybe it also works with 48GB (but that's not sure).