Performance

General Performance: SYSMark 2004 SE, PCMark05

SYSMark 2004 SE test measuring the system performance in real applications turned into a true awakening for those AMD fans who despite common sense have hopes that the new platform would prove a worthy competitor to Intel’s solutions until AMD comes out with their K8L architecture. The new AMD Quad FX platform loses not only to the quad-core Kentsfield processors, but even to the dual-core Conroe processors.

If you take a closer look at the diagrams, you will be able to notice another pretty strange thing: Quad FX platform based on two dual-core Athlon 64 FX working at 3.0GHz loses even to one dual-core Athlon 64 FX working at 2.8GHz. However, there is nothing unnatural about it: the reason lies with the organization of the memory subsystem in the AMD dual-socket platform. Although each CPU has its own memory subsystem, the data gets distributed evenly over all the memory banks independent of the CPU they are assigned to. Therefore, each processor has to address not only its own memory subsystem, but also the memory subsystem of the other CPU. These “long-distance?requests feature much higher latency and hence slow down the system. And if we add that the data transfer between the CPUs is performed solely through the memory subsystem, then the performance drop we observe in dual-socket Quad FX platform will be absolutely logical.

As a result, the applications working actively with the memory subsystem may turn into a so-called Achilles?heel for Quad FX platform.

Another test, PCMark05, produced better results on Quad FX. At least, this dual-processor platform with dual-core CPUs in it outperformed the single-processor system with one Athlon 64 FX-62. Although, working at 3GHz speed it cannot catch up with the quad-core CPUs from Intel. Looks like contemporary AMD processors cannot compete against Intel Core micro-architecture even within a dual-processor system.