Memory performance
Per our custom, we begin with some synthetic memory benchmarks. These numbers aren't clear indicators of how application performance will look, but they can tell us some interesting things about the memory subsystems of these processors.

The integrated memory controller on the AMD chips here makes for an unusual dynamic: memory bandwidth rises dramatically with CPU clock speeds, despite the fact that all of these CPUs are talking to the same DIMMs, using the same timings, over the same motherboard. The Opteron 165, with its relatively low 1.8GHz clock speed, demonstrates throughput considerably lower than the Opteron 180.

Notice that the performance of the Opteron 180 and Athlon 64 X2 4800+ is, for all intents, identical. This is the first of many graphs you'll see where these two chips' performance is separated only by the margin of error in the test itself (and perhaps some minor revision differences in silicon between the two.)

Our version of Linpack isn't optimized enough to serve as a true scientific computing benchmark, but it does show us the impact of moving through different stages of the memory hierarchy, from L1 cache to L2 and then into main memory—at least on one of the two cores on each CPU. You can see how the Opteron 165's larger L2 cache boosts throughput at larger matrix sizes compared to the Athlon 64 X2 3800+. The Pentium D processors, by contrast, all have larger L2 caches that can hold all of the matrix sizes in this test.

Memory bandwidth and latency are interrelated, so these results shouldn't be a big surprise. The Opteron 180 matches similar dual-core Athlon 64s and easily outpaces an Intel chip thanks to its integrated memory controller. The 165, though, is a little bit slower due to its lower clock speed. Still, the 165's access latencies are lower than any Pentium we tested.