Overclocking potential :
Since this is the first time I've worked with 4Gb PSC powered dimms on the socket 1366 platform. I was looking forward to see if these babies could clock near as good as on the 1155/56 motherboards.
To keep all settings/speeds aligned we opted to stick with the 133 Bclock and to just change ram divider (1600-1866 and 2133). So total CPU clock and uncore speeds remains the same for each opted ratio.
So again using : 133Bclock with a 30 Multiplier, totalling in a 4000Mhz CPU speed. The Uncore was kept at 3600Mhz. Back to back locked at 4 and the Command Rate if possible set to 1T.
We just tested the stock dividers, so 1600 vs 1866 and 2133Mhz.
With HCI Memtest Pro we tested how stable the sticks were able to run.
We got the following timings out of Gskills latest creation :
1866Mhz CL8-9-8-24 1T at 1.65Vdimm
2133Mhz CL9-11-10-27 1T at 1.67Vdimm
2133Mhz CL9-11-9-27 was not possible, not even at elevated voltages of 1.75Vdimm. Yes they passed the entire test suite but would fail in HCI's memtest before passing 200%
While these might seem average to some, don't forget these are 4Gb dimms, putting a lot of stress on the Integrated Memory Controller (IMC)
Also we needed added cooling for 2133Mhz operations, without they would fail in HCI Memtest.
Time for the usual test suite. Starting off with SuperPi 1M and Wprime 32M.
Both being very short tests, but heck we include them anyway. Though overal CPU speed rules tighter timings. Not much to see here, let's move on...
The overclocked 1866Mhz Snipers are taking a small advantage over the stock kit. Being it SuperPI32M or Wprime 1024. Those who are expecting some more gain with more ram speed, are wrong. With the 2133Mhz divider, the RTL (Round Trip Latency) values were pretty loose set. So we corrected them to 59-60 to get performance back up. Secondly the timings had to be pretty loose 9-11-10-27 to pass HCI memtest.
Both our 3DMark tests show us the same outcome. 1866Mhz slightly topping the stock kit, 2133Mhz trailing again.
AIDA64 clarifies a bit more what is going on : at the 1866 divider we get a small increase in overal bandwith. 2133Mhz however is struggling to keep up with the stock SPD settings. And this after correcting the Round Trip Latency. Without it was set at 56 and performance was sloooooooow.
Maxmem, one of the benchmarks being used at Hwbot.org tells the same tale. We already discovered similar lack of efficiency on X58 : higher ram speeds with looser timings is usally just as fast then slower ram speed with tighter timings...
1866Mhz crawling ahead of the stock settings. The 2133Mhz setting still can't make the difference. On X58 best option is to run these sticks out of the box, or slightly overclocked... though it doesn't even seem worth the hassle
Hardly any scaling, 30 points is negligable between 1600 or 2133Mhz. Flashback : seeing simularities when I tested the Kingston 2250CL9 kit... lower ram speed with tighter timings was as fast, if not faster then higher speeds with loose timings...
Our 4 game tests follow up on the results of the above tests. Slight advantage for the 1866Mhz divider... but it will be more worthwhile to overclock that GPU to increase FPS...
Conclusion : 1600 or 1866Mhz is optimal on X58... 2133Mhz is far from. Unless you enjoy lots of subtiming tuning you can maybe get the performance on par. On Sandy Bridge these sticks are best run at high speeds...