Intel® Moderncode for Parallel Architectures
Support for developing parallel programming applications on Intel® Architecture.

Hyperthreading in dual-dual processor

mattsdad
Beginner
269 Views

I am running an extreemly CPU intensive multi-body simulation which lend itself nicely to parallel processing. We have run several tests with the same simulation using a single-threaded compilation and a multi-threaded compilation. We have run the same simulation with the hyper-threading on and off. Here are the results:

The 12-body simulation on the dual-dual core Xeon machine runs about 70% faster when using the multi-threaded compilation compared to the single-threaded compilarion.

When we turn hyper-threading on, the execution of the multi-threaded compilation slows down about 15%. In this CPU intensive simulation, the 8 processor threads fight for CPU time and only average about 50% CPU time for each hyper-thread. With the hyper-threading off, the same simulation uses all four CPUs at almost 100% throughout the simulation.

So, we are are turning hyper-threading off because most of our processor time is spent running these CPU intensive simulations.

0 Kudos
2 Replies
TimP
Honored Contributor III
269 Views

Memory contention seems a more likely reason for low CPU utilization with increased number of threads. Several factors could be involved, including which OS you use.

0 Kudos
mattsdad
Beginner
269 Views
There is no question or problem here. I was only providing the data so others could use this data point in their evaluations. Maybe encourage checking the speed with and without hyper-threading for their own programs.
0 Kudos
Reply