Memory Tests
The memory tests are designed to highlight the general performance of the memory subsystem. We use a trio of tests to analyse said performance, Pifast is up first. It's our own custom test to 10M decimal places.Pifast
The lower access latency of Socket 939 FX-53 is the reason the Socket 939 systems, including the KV2, are a touch faster than the Socket 940 system. The KV2 is quickest.
Sciencemark 2.0 - Bandwidth
Sciencemark 2.0 measures peak CPU to memory controller bandwidth, the bandwidth available to the CPU for memory intensive tasks. It doesn't take into account any excess bandwidth available on the memory controller, that may or may not be usable by other peripherals.
With the memory controllers on the processor in the systems on test, the exact same processor used in each Socket 939 system, and the same basic memory timings, it's no surprise to see the two Socket 939 systems bunch up like that. The Socket 940 version trails due to its requirement for registered (slightly buffered) memory.
Sciencemark 2.0 - Access Latency
Again, the requirement for buffered memory means that the Socket 940 system is slightly slower than the Socket 939 systems. The KV2 puts up the correct numbers throughout in the memory subsystem tests, a good indication that it's working correctly and quickly.