Our Quantum Chemistry group owns a 98 CPU PC cluster named Orpheus, with a total peak performance speed of over 130GHz. You can read more about the number and specs of the nodes here.

It is also worth to mention that the management of the jobs sent to the cluster is done by a PBS queueing system. You can find more info about the different queues we have set up here.

Aditionally, you can view some pictures of the cluster here.

I have recorded some historycal data of the usage of the cluster, and this data can be viewed following the links listed below. The meaning of the keywords given is:

Average load per user: total CPUs times clock GHz speed, averaged over the time range from the starting data to the actual point. Effectively the integral of the instantaneous use (not given) divided by the time span.

Total use: sum of instantaneous use by all users at a given date. The peak performance (sum of all processors in the cluster) is also plotted.

Use for user: number of CPUs a given user has used in each queue, averaged over time.

Work load by queues: total number of CPUs working on each queue, averaged over time.

Queue use: percent of the total capacity of the queue used, in average, over the whole time period.

Queue performance: queue use, times clock speed for CPUs in that queue, thus faster CPUs get bigger figures, even if they have been exploited just as much as other slower queues. This graph gives a quite accurate idea of which nodes/queues are "doing the big work".