Provide Processor Information
Updated: July 1, 2009
Applies To: Windows HPC Server 2008
To calculate the theoretical efficiency of your HPC cluster, Lizard needs to determine the number of floating-point operations per clock cycle that each core in the processor of the head node of your HPC cluster is capable of performing, and uses that information as a reference for the rest of the compute nodes in your cluster. If Lizard cannot automatically determine this information about the processor, you will be required to provide it.
|The number of floating-point operations per clock cycle of a core in a processor should not be confused with the theoretical or calculated peak floating-point operations per second (FLOPS) of the processor as a whole. For example, a processor might be able to perform at 24 billion floating-point operations per second, or 24 GFLOPS, because each of its two cores can perform 4 floating-point operations per clock cycle, and its clock speed is 3 GHz. Basically: 4 floating-point operations per clock cycle x 2 cores x 3 billion times per second = 24 GFLOPS.|
The theoretical peak performance of the whole HPC cluster is calculated based on the number of floating-point operations per clock cycle of each core, which in turn is used to calculate the efficiency results of the tuning process, as explained in Performance and efficiency results. To ensure that the theoretical peak performance of your HPC cluster is properly calculated and avoid calculating efficiency results that do not accurately represent the efficiency of your HPC, you must provide an accurate number of floating-point operations per clock cycle for the processor that is used as a reference. This information is usually available on the Web site of the manufacturer of the processor, and it is a positive integer, usually equal to 2 or 4.