Users of HPC systems are always on the lookout for more performance for their applications. Some users turn to FPGAs (Field Programmable Gate Arrays, essentially a reconfigurable processor) to accelerate their algorithms but this options is time consuming and expensive. HPC users often look to the off the shelf market and in doing so found some kindred spirits: gamers.

Today, AMD released the newest revision of it's quad-core Opterons processors. These new processors, codenamed Shanghai, mark AMD's first product to be built with a 45 nm fabrication process.

The Shanghai processors bring some updates to the older Barcelona products. First, the L3 cache size has been increased to 6 MB and the on-chip memory controller has been upgraded to support 800 MHz DDR2 DIMMs which increase memory bandwidth up to 17%.

Our clusters are tailored to each users needs but the off the shelf storage options were rigid and limiting. A server with a large array of disks would be able to adapt to most any use but requires expertise and time to adapt to a changing environment. Dedicated storage systems were simple to manage but limited by either performance or protocol support. Advanced Clustering's Flexible Storage Server provides a simple web based interface without limiting the user to one protocol.

Ok, so the subject line is a little tongue-in-cheek. The purpose of this post is to discuss BLAS libraries and power consumption. And what a giant amount of variance in power consumption there is among the Big Four BLAS Libraries (BFBL):

• ATLAS
• MKL
• ACML
• GotoBLAS