Is Nehalem the right balance?

March 25th, 2009 11:41 am
Posted by Dr. Thomas Schoenemeyer
Tags: , , , , , , , , , , ,

An effective way to judge the balance of a computer design is by the level of efficiency when running actual applications – often measured as the percentage of the peak performance achieved on an actual application. Vector supercomputer designs usually outperform other systems with this ratio due to the additional capabilities in memory bandwidth, I/O bandwidth, and communication connections.

During the last two years computers peak performance increased to levels far above, however at the same rate the efficiency of real applications decreased on these platforms. It is very likely that we will not gain more than a few percent of efficiency.

Memory bandwidth and low latency of the communication network are the key elements for excellent scalability and efficiency.

By the help of the Quick Path Interconnect, the ratio between CPU performance and memory bandwidth increases to 0.75, which is actually considerably lower than for a SX-9 supercomputer (=2.5), however the Interconnect Bandwidth to node performance expressed in Byte/Flops exceeds the SX-9 by a factor of two.

In combination with a carefully designed network topology the new Intel Nehalem processor is capable of running weather forecast applications with a very reasonable efficiency.

These attributes can add to the initial hardware cost of the system, but can minimize the overall TCO (Total Cost of Ownership) with usage, such as in lower electricity, smaller footprint, higher reliability, and lower application development costs for certain classes of high-end applications.


You must be a Registered Member in order to comment on Cluster Connection posts.

Members enjoy the ability to take an active role in the conversations that are shaping the HPC community. Members can participate in forum discussions and post comments to a wide range of HPC-related topics. Share your challenges, insights and ideas right now.

Login     Register Now

Author Info

Dr. Thomas Schoenemeyer has been working in the area High Performance Computing for 13 years, and at NEC for 8 years. His main tasks are application porting, optimization and parallelization, with a special focus on applications in the Earth Sciences Community. He contributed to the development of the OASIS coupler development. Currently he is deeply involved in the evaluation of Intel Nehalem architecture for weather forecast applications.