The Brethren
We had the choice of either using an Intel Xeon platform or an AMD Opteron system. Obviously, we picked AMD for benchmarking. The first reason is that currently the AMD processors deliver more bang for the buck, both from a general performance and a performance-per-Watt perspective.
In addition, the Xeon has been suffering from two architectural disadvantages. First, the current platform requires both CPUs to share one front side bus to access the system. As a result, the bandwidth per processor is somewhat limited to 6.4 GB/s at FSB800. Second, the memory controller is part of the chipset Northbridge, which means that the memory bandwidth (6.4 GB/s) again is shared by both processors.
AMD's HyperTransport architecture on the other hand offers much higher bandwidth to the system (6.4 GB/s per HyperTransport link, which especially applies for processor-to-processor transfers). Each processor comes with its own memory controller, which helps to shorten latencies and to increase the total memory bandwidth available (again, 6.4 GB/s per processor).
The single-core Opteron 248 still is a 130 nm device based on the Sledgehammer core, which runs at 800 MHz HyperTransport and a 2.2 GHz core clock speed. In contrast, the Opteron 275 dual core is based on the 90 nm Italy core and again runs at 2.2 GHz. Here, the dual core has some advantages, because it features AMD's E revision core with an improved memory controller and it also supports SSE3, which the single core Opteron 248 does not support.
A comparison between a single-core and a dual-core workstation or enterprise class system should not be based on a clock-speed comparison. Most IT administrators or professional users would go for dual cores in order to increase the computing density, but hardly anybody expects to double performance. However, we compared the chips at the same clock speeds in order to come up with the most affordable dual-processing system at a certain speed grade.
