Chapter 5 Cache and Memory Optimizations 97
Software Optimization Guide for AMD64 Processors
25112 Rev. 3.06 September 2005
Figure 1. Simple SMP Block Diagram
The AMD Opteron processor implements a Cache-coherent nonuniform memory access (ccNUMA)
architecture when two or more processors are connected together on the same motherboard. In a
ccNUMA design, each processor has its own memory system. When a processor accesses memory on
its own local memory system, the latency is relatively low, especially when compared to a similar
SMP system. If a processor accesses memory located on a different processor, then the latency will be
higher. The phrase ‘non-uniform memory access’ refers to this potential difference in latency.
In an AMD Opteron processor system, each processor contains its own memory controller. Figure 2
shows an example of a two processor AMD Opteron system in a ccNUMA configuration.
Figure 2. AMD Opteron
Dual-Core AMD Opteron processors and AMD Athlon X2 Dual-Core processors share the on-chip
integrated memory controller and memory. Two or more AMD dual-core processors still use the
ccNUMA configuration. Figure 3 illustrates a dual-core AMD Opteron configuration.
CPU0 CPU1
Memory
AMD Opteron™
Memory
CPU0
AMD Opteron™
CPU1
HyperTransport™
Memory