[roued]

Hi,
We are just installed a new Linux cluster for running Ansys CFX.
Cluster is setup using the latest Intel Xeon E5-2680 @2.7 Ghz and infiniband network.

However when comparing to our older workstations (Intel Xeon X5570@ 2.93 Ghz) we are not seeing the speed-up that we have hope for.

The cases we are using for benchmark are fairly large (more than 50 mill cells) and are about wind environment in urban areas. But have also seen problem with other smaller flow cases. We are comparing the runtime for with:

Cluster:
1 blade using 16 core (2 x E5-2680), using infiniband
2 blade using 8 cores on each and using infiniband

Workstation:
2 x workstation (2 x X5570 in each workstation) connected through Gbit network.

The Intel Xeon E5-2680 contains 16 core and have a faster memory bus compared to the X5570. However the X5570 does a have a small lead regarding clock frequency.

Does anyon have experiance similar problem with the preformance of the
Xeon E5-2680 in a cluster environment ?

And does anyone have a possible solution to the preformance problem ?

Thanks in advance,

Roued

[sainath.s@tdps.co.in]

Because it is 2.93GHz against 2.7Ghz.

[ghorrocks]

Clock frequency means very little by itself for CPU speed. You really need to look at reliable benchmark speeds to have any rational basis to compare CPUs, my favourite benchmark is the CPU2006fp benchmark from spec.org.

Can you describe what performance is poor and how you measured it? Some context would be handy it you want us to make comments about what is wrong.

[roued]

Hi,

Problem solved. Firmward updated on the Mallanox infiniband, and
a missing riser card was installed.

Preformance now show super-linear speed-up on all Ansys CFX
benchmark cases (IndyCar, Internal, Airfoil etc).

Best Regards

Roued