Study of Athlon 64 Overclocking Techniques - PAGE 4Tom Karpik - Wednesday, March 30th, 2005
In the following tests, I raised the LDT bus speed to specific values while keeping the memory speed constant through the use of memory dividers. The CPU multiplier was kept at 9x in order to see the effects of a faster LDT bus and CPU clock without the benefit of faster memory.
XviD encoding really likes CPU clock speed. As we scale the LDT bus and clock speed, frame rate rises close to linearly.

SiSoft Sandra arithmetic and multimedia scores scale as expected as we raise the CPU clock. Memory bandwidth scores are where it's all at. Memory bandwidth efficiency goes up at each increment of the LDT bus because we are raising the CPU clock, but the memory is kept constant. At 267 MHz on the LDT bus, memory bandwidth is at close to 6000 MB/s, at an impressive 91% efficiency. It is obvious that the K8 architecture depends on CPU clock speed to make use of memory bandwidth due to its onboard memory controller that scales with processor speed as was alluded to earlier.

MP3 encoding is quite dependant on CPU clock, and as such, we trim 7 seconds off our encoding time, bringing us to 20 seconds at 267 MHz on the LDT bus.

Half-Life 2 also sees some nice gains from raising the LDT bus and CPU clock speed. We get almost a 20 FPS increase going from 200 to 267 MHz.

Not unlike Half-Life 2, Doom 3 also gets a nice bump in performance - in the order of over 27 FPS.
What have these tests proven? They've proven that overclocking your K8 system while using a memory divider to keep that cheaper DDR RAM within spec is nothing to be ashamed of. We saw quite healthy gains in all departments, including memory bandwidth, merely by raising the CPU clock even though we kept the memory at ~200 MHz at all times. I repeat: You are not a weenie if you use a memory divider. :-)