For the life of me, I've googled about a hundred permutations of rank, channel, DDR3, benchmark, -forum, -haswell, and I cannot find whatever that damnable article was.
As Grall mentioned, I'm almost positive it had to do with total open active pages, but also dealt with the clocking limits involved with dual ranked, dual-channel, and two-dimm-per-channel configurations. The setup was specific to Intel Core i-series processors with their IMC, and there were multiple percentage points worth of differences in non-trivial workloads.
But, since I can't find the damned thing, I guess it doesn't matter. Whatever you do, get dual-rank memory (chips on BOTH sides of the DIMM) for your config.
As Grall mentioned, I'm almost positive it had to do with total open active pages, but also dealt with the clocking limits involved with dual ranked, dual-channel, and two-dimm-per-channel configurations. The setup was specific to Intel Core i-series processors with their IMC, and there were multiple percentage points worth of differences in non-trivial workloads.
But, since I can't find the damned thing, I guess it doesn't matter. Whatever you do, get dual-rank memory (chips on BOTH sides of the DIMM) for your config.