2250Mhz cas 9 sound nice, but these speeds can only be easily reached on a Gulftown based CPU. Why you might ask ? With Bloomfield ( 920/30/40/50 and 960 CPUs) you need to run the Uncore at twice the speed of the ram, so 2250Mhz equals 4500Mhz Uncore speed. This is not achieveable on air with a bloomfield CPU, Gulftown only needs a 1.5 ratio to be maintained. In this case 3375Mhz to reach the rated ram clocks. Time for a test to see if we can get similar performance by tightening the timings and run at reachable speeds for your Bloomfield CPU.

How we tested :

133Bclock was kept and just by selecting a different ram multiplier we tested the tightest timings for each divider. First a few loops of Memtest ( before entering windows ) and then at least 100% of HIC memtest had to pass. If the ram is unstable it will easily fail on one of these two tests. Here are the best timings we could achieve on our setup :

1600Mhz CL6-8-6-24 Trfc 88 1T at 1.65Vdimm

1866Mhz CL 7-9-7-27 Trfc 98 1T at 1.67Vdimm

2000Mhz Cl 7-10-7-27 Trfc 100 1T at 1.71Vdimm

2133Mhz CL8-10-8-27 Trfc 120 1T at 1.68Vdimm (setting will normally be out of bounds for Bloomfield daily usage too, but I testged the divider anyway)

The above are all HCI Memtest stable. Eg 1866Mhz with 6-9-6-27 timings was also stable for the test chosen suite, but HCI Memtest would error out. Upping volts didn't make it through the Memtest. Looser timings however made it HCI Memtest. So 7-9-7-27 it had to be...

As always overclocking results can vary from kit to kit ( sad but true )

Not much to see here, similar results, all within the margin of error. The test duration is too short to pick a clear winner, but 2000Mhz CL7 is already my favourite...

Let's see if 32M brings some more clarity : Indeed, coming close to 2250Mhz performance ( considering the latter has a 70Mhz CPU and 250Mhz ram advantage ) 2133Mhz seems to be not so efficient as it should be. 2000Mhz seems to be a nice sweetspot ( keep in mind that your Bloomfield needs to run 4000mhz uncore then and will need propper cooling to so do ) Even Wprime1024 scales a bit with the run settings...

Of the Bloomfield selected speeds the 2000mhz Cas 7 comes out on top again for the synthetic 3D tests. Though scaling is minimal ( hence why some benchers usually limit to 2000Mhz with tight timings. Also due to enormous bandwith amount only dual or even single channel ops are commonly used. This in preference over triple channel to achieve higher overal CPU and uncore speeds.

Cinebench multithreaded clearly favours the 2000Mhz C7 setting. Again due to the loosening of the latencies, the 2133Mhz performance is somewhat crippled (not that it is of a real importance for daily OC'ed Bloomfield users)

Same story continues in PCMark05, 2000Mhz performing great, too bad TRCD 10 was required. It would have been fab if 9 was possible. But no matter what voltage or other setting I tried, HCI Memtest kept erroring out. Also once over 1.72ish Vdimm the sticks became unstable and started to crash in eg Hyperpi...

 Looking at the previous results , we notice the 2000CL7 setting is outperforming the other ram ratios. Maxmem shows us why : especially the Copy and Write tests are far better than at the other ratios. Almost besting the 2250CL9 results.

AIDA 64 also shows the Copy test to take a huge leap from 1600 and 1866 ratios. The rams at their rated speed are the fastest, but the lower clocked rams at tight timings are not that far behind...

Best output result at 4Ghz was again for the 2000 C7-10-7-27 setting. This time however there's a noticeable difference between the 4070Mhz of the 2250Cl 9 as it renders over one FPS and a half more than the 2000mhz setup. Quick cross comparing with the results of the previous page (Hyper ICs) the powerchips are outperformed, even when being run at tighter timings.

The 4 Game tests are not any different than the other tests run before. However the gain is minimal, it's not even worthwhile to run rams at high frequencies compared to regular 1600 or 1866Mhz frequencies.

Max clocks I could achieve on this triple channel kit was 2300Mhz. Loosening the timings to CAS10, upping voltages, nothing helped to pass HCI Memtest. Maybe some cold would unleash more potential, but that's not the purpose of this review...