Memory Subsystem Performance
After analysing platform performance first where we didn't see the increases expected, analysis of the memory system showed the main reason why.Memory Bandwidth
A 15% increase in memory clock resulted in less than 4% more bandwidth compared to the HEXUS reference score. The memory in the SCAN 3XS Chameleon system, while running at 1T command rate, has latency clocks of 2.5-4-4-8. And while latencies don't affect overall performance too much on AMD Athlon 64-based systems, it's the reason why the measured deficit occurs here. Tighten up the latencies at the same clocks and bandwidth goes up.
Memory Latency
When the CPU is made to go out to main memory the 3XS Chameleon sits at a disadvantage compared to the HEXUS system. The lacklustre memory bandwidth and this latency deficit is why we don't see higher scaling from the CPU's overclock.
Pifast
Pifast backs it up with a less than 6% advantage.
