Time to put these babies where they belong, plugged in the dimm slots of a high end socket 1366 system. For the out of the box performance we will put the Kingston 2250CL9 vs common used ram clocks. Versus 1600Mhz CL8, 1866Mhz and 2000Mhz CL9 setups. Hyper kits with better/tighter timings are still being used, but pretty rare. But anyway, for the enthousiasts I compared the Kingston set with Corsairs' GTX2 at the same speeds a few pages further on.
The test setup comprises of the following hardware :
Asus Rampage III Formula X58 board (bios 0505)
Intel 990X @4Ghz, uncore 3600mhz watercooled
Asus GTX480, 266.58WHQL drivers
Windows 7 64Bit Professional
The RAM clocks were tested like this :
1600 and 1866MHZ clocks :
133Bclock x 30 for 3990Mhz, Uncore at 3600Mhz. Simulated by using Corsair GTX2 dimms
1600/1866 divider selected : 1600Mhz CL8-8-8-24 1T & 1866Mhz CL9-9-9-27 1T TRFC 82
2000Mhz Clocks :
143Bclock x 28 4004Mhz. 1866 ram divider : 2002Mhz with 3580 Uncore speeds. 2002Mhz CL9-9-9-27 1T TRFC 88. This simulated by using Corsair GTX2 dimms.
2250Mhz clocks :
140Bclock x 29 for 4070Mhz, Uncore at 3674Mhz for 2250Mhz Cl9-9-9-27 1T TRFC 120
As you noticed the 2250Mhz seting has the highest CPU and uncore speeds. Sadly comparing rams isn't as easy as with eg the Sandy bridge platform. More variables come into play with socket 1366. Therefore Back 2 Back was also locked at 4 setting.
So plz keep in mind the extra CPU and uncore speeds when looking at the results.
No real surprise as the Kingston kit steams ahead of the other tested configurations. Pi 1M and WPrime being very CPU dependant loving every available Mhz. Timings and ram speeds hardly matter.
Similar picture for the extended test versions. Highest CPU speed wins, but you can already spot a nice scaling going from 1600Mhz to 2000mhz (similar CPU and uncore speeds)
Time for some 3D synthetic action :only a few 100 more marks due to increased ram speeds. For a bencher critical, for the daily user absolutely insignificant.
Cinebench Release 10 64bit was used to test the single and multi core performance of the setup. Again slight noticeable scaling with higher ram clocks... the mere 70Mhz extra CPU speed and high ram clocks really pull ahead.
Futuremark's PCMark05 tests overal system performance, including CPU, GPU to HD performance. Gradually the score increases if we push ram clocks to higher limits. Even the 2000C9 score is pretty close to 2250Mhz, explainable due to the higher performing, but so fragile Elpida Hyper chips on the downclocked Corsair GTX2 kit.
Maxmem has been in use for a while now at the HWbot website and most glitches/bug are rectified. This makes it pretty reliable for analysing multithreaded ram performance. Magic words : bandwith scaling with more ram speed ! A solid boost from 1600 to 1866Mhz, from there on the jump in bandwith is less pronounced.
With AIDA64 comes a very handy ram bandwith benchmarking utility. Measuring not only the ram performance but also the Level 2/3 cache of the CPU. Though in the above chart we only listed ofcourse the ram performance. Similar scaling as with Maxmem. Take note we still use the 1.5 version to keep the data of previous tests aligned.(so your output can vary slightly)
For the encoding test I still favour X264HD, compiling a video fragment into HD quality. Being multithreaded it nicely demonstrates your setup's potential. Higher ram bandwith and speeds allow you to render more frames per second. Again it might not look like much, but every frame counts here. Alike when benching hard.
Synthetic stuff aside, time for some games. Tested at 1600 x 1200 resolution with high to very high detail selected, just like you would run it to actually play the concerned game.
Far Cry 2 is known for nice scaling with extra CPU/RAM power. And indeed this is no different here, though we are reaching a GPU wall here. One extra FPS going from 1600 to 2000mhz ram speeds, Even with extra CPU speed and even more ram speed, the result of the 2250 setup is similar to the 2000Mhz one. With similar Resident Evil test results for most dividers. Wickedly enough the 2250Mhz was retested 5 times and the average outcome was always lower than the 2000mhz setup.
Even though we could have lowered the resolution to make the games less GPU dependant, thus to show off the benefits (if any) of higher ram clocks. We still opted to test the games as how they should be played with loads of eye candy, DirectX10 and co enabled. Very little performance gain to be expected if the GPU is the limiting factor...