| 5:26:16 PM PST - Thu, Dec 13th 2018 |
|
Hi,
I am in the process of defining the hardware in individual nodes for a cluster that will run NWChem. We will run these processes across 24 to 40 nodes depending on the configuration of the nodes. The total amount of cluster memory would be the similar for both configurations, the only difference would be the node count. Both would also be connected via 100G InfiniBand. The two types would be:
• 2 sockets, 12 cores per socket, base clock 3.2GHz, turbo clock 4.2GHz (40 nodes, 960 cores)
• 2 sockets, 20 cores per socket, base clock 2.4GHz, turbo clock 3.7GHz (24 nodes, 960 cores)
Would the increased single threaded performance of the faster CPUs be offset by the increase in the amount of nodes? Which one of these setups would have better performance for large simulations?
Thanks!
|