Results: SiSoftware Sandra
The $1200 and $1600 machines employ different models of the same quad-core Haswell-based die. Don's Core i5 doesn't benefit from Hyper-Threading though, and 2 MB of its shared L3 cache is disabled compared to my Core i7. The i5 is also clocked 100 MHz slower by default, but Don was able to overclock his -4670K 100 MHz higher than my Core i7.
Still, the i7 wins this face-off, demonstrating the benefits of parallelization, even through a technology like Hyper-Threading, where two threads are scheduled to each physical core.
My more conservatively-overclocked high-end machine also enjoys an advantage in Sandra's Cryptography module, though we know that this is related to memory bandwidth available for shuttling instructions to the processor's cores.
Sure enough, there's the source of my advantage (at least in the AES-NI-accelerated Encoding metric). Memory bandwidth charts should never scale proportionally to system price. But this one does. That’s partly because Don doesn’t overclock his memory, and lets it run at underclocked defaults as a baseline. At the other end of the chart, Paul can blame AMD's integrated memory controller for the $600 machine’s poor results.
Fortunately for both of my fellow editors, real-world software rarely reflects memory bandwidth with a 1:1 relationship. It's more likely that we'd see slightly lower performance in certain tests attributable to these numbers.