Memory performance
This first round of tests is very much synthetic, measuring the performance of the memory subsystem, but it's the perfect setup for what follows. These results should illustrate how the CPUs differ in some key ways.

Linpack shows us floating-point math performance while the CPU is operating on data matrices that fit into the L1 cache, then the L2 cache, and finally spill over into main memory. The FX-57's large L1 data L2 caches combine to form an effective total data cache size of 1088K. As you can see above, performance drops off once the matrix sizes approach that barrier, because main memory is much slower.

The FX-57's rev-E heritage shows up here in a couple of ways, including the fastest performance we've seen from any of the Athlon processors as they reach into main memory. Also, notice how the Athlon 64 X2 4800+'s L2 cache bandwidth matches that of the FX-55, even though the FX-55 runs at 2.6GHz and the X2 4800+ at 2.4GHz. That's because AMD's 90nm L2 caches are plainly faster, clock for clock, than those on its 130nm chips. Judging by its scores at the middle matrix sizes, the FX-57 benefits from the faster L2 cache at 90nm, as well.

Here's a nice illustration of the benefits of the Athlon 64's built-in memory controller. Memory access latencies are much lower than those of the Pentium processors, because the Intel CPUs must access RAM via a memory controller on a core-logic chip situated behind a front-side bus. The A64's one-hop path to memory is almost twice as quick.