[aks]

HI,
i am looking to purchase a new workstation (i have interest in buying DELL T7600 with 16 cores Intel® XEON_8 (E5-2687W) with 128 GB RAM. I want to run ANSYS CFX with 5 Million size mesh on it for a rigid body solver case.
As present i have DELL Optiplex 990 and i just want to know that how many times faster DELL T7600 with 16 cores and 128 GB RAM will be as compared to DELL Optiplex 990 (3.4 GHz processor) with 4 cores and 8GB RAM.

[bindesboll]

Why would you need all that memory?
I am running 9 mio. nodes cases and using max. 22 GB. So even if you have to model particles and radiation 32 GB (8 x 4 GB) would be more than sufficient for 5 mio. nodes. And then you dont need the T7600, but the T5600.

If you are running octo-core processor, the clock speed will not be critical as it is memorybandwith limited. The E5-2687W is expensive, so consider E5-2670, you wont see any difference on the performance.

You could consider the Ivy-Bridge-E generation CPUs that is released in september. The clock speed for octo-core is increased from 2.6 GHz to 3.3 GHz, and the RAM frequency from 1600 MHz to 1833 MHz (E5-2667 v2).

[aks]

Hi Kim,
Thanks for the reply. I am running a transient case and on my present system with 4 cores and 32GB RAM it takes around 30sec. for one iteration and its going to be around 40 iterations per time step and time step is roughly 0.0002 s, and total time is around 8 seconds.
So, for running the complete transient case it will take me around 2.28 years.???? So i am hoping that by cutting down on mesh size by some extent and by increasing the time step and reducing the no. of iterations per time step on some better workstation i hope to complete the case in around 2 mnths time???
Dell T7600 is within the budget of the university so cost is not that much of a problem but if can suggest some good system with 32 cores or something like that it would be great...

[bindesboll]

That sound like a lot of iterations per time step!
I have succesfully used this transient setup in CFX:
Time step length: Automatic
Max iterations: 7
Min iterations: 3
Convergence criterion: 10e-4
Then the solver will automatically adjust your time step size (seconds) to a level where convergence is obtained within 3-7 iterations.
If 40 iterations are required, you have probably too large time steps or something else wrong in your setup.

More memory will not speed up your simulation.

I am just about to order a 32-core system as replacement for my present 4-core Xeon X5687. The new system will most likely be something close to:

2 pcs:
Dual CPU motherboard
2 x Xeon E5-2667 v2 3,3 GHz, 8-cores
32 GB: 8 x 4 GB RAM 1866 MHz
Harddrive for Windows: 250 GB, 7200 rpm SATA-600
40 Gbps Infiniband network (cluster interconnect, no switch: directly coupled)
Gigabit Ethernet netværkskort (systemnetværk)

One of the machines additionally:
3 x 500 GB 10.000 rpm harddrive RAID 5 for ANSYS software and CFD files storage.

The new CPUs will be released during this September.

[aks]

I think the set up of case is okay , but may be because it involves rigid body motion using 6DOF rigid body solver and remeshing etc. i think it is taking more iterations per time step due to force convergence on rigid body moving inside the fluid. (Momentum and mass are getting converged in around 15 iterations! )
I have set the convergence criteria as:
1e-06 for mass and momentum,
1e-04 for force convergence on rigid body

i will try what you have suggested using automatic time step. Thanks for the suggestion..
The configuration of the system also seems great. Is it possible to run CFX is parallel on 2 different machines? If its the case then i think i also should go for 2 machines rather than just 1.
But the processor you talked about is still not available , hopefully it will be available in september..

[bindesboll]

ANSYS can run on several hundreds af parallel compute nodes. However, I am not into how it is configured.