Cas 3/1T, 400FSB 1:1 testingNow we're getting to more serious business. As mentioned previously, we're going to put the memory through its paces now, by tightening the timings at 400Mhz as much as possible. Furthermore, we're letting front side bus and memory run in sync with each other, which should give the best possible bandwidth and latency results, and the best performance for our motherboard overall. For tightening the timings, we chose a cas 3 setting, and a command rate of 1T. We started with 3-3-3 timings, but for some modules we had to relax these to 3-4-3 or 3-4-4. Luckily, all our modules were able to run at a command rate of 1T, but for the OCZ PC6400 it was a tight call indeed.
After that we just booted to windows again, and started testing. Again, no further tweaks, no hidden performance enhancements, just straight test results.
First the overview table (click on it to open in a new window) :
Again, the best results are highlighted in green. Let's look at the individual results now in the following graphs.
A new look at the bandwidth of Lavalys and Sisoftware's SANDRA applications:
As you can see, we're talking about very different bandwidth results here: all modules score well over 10,000 Mb/s now, which is a very nice result. The Micron kits clearly take the lead here, with our high end kits up front. The Everest result for the PC8888 Dominator is rather low, but the Sandra result is not. Both the dominator parts were struggling though with the command rate of 1T, and this seems to be reflected in the results.
On to the latency results :
Again, the dominator parts and the OCZ Elpida kit finish at the back of the pack. The other kits have absolutely great results which can only be obtained on the 680i chipset.
Let's look at SuperPi :
Logic is followed here... almost. The best performing bandwidth and latency parts come out on top, except for the OCZ PC6400 and G.Skill parts which seem to have switched places, strangely enough.
Cinebench next :
Another strange result: the top end Dominator part comes out on top here, while all the rest share the same score. Cinebench is obviously hardly dependent on memory programming.
Time again for our 3D Mark 2006 benchmarks, first the cpu result:
Differences are once again very small, with the OCZ micron kits coming out ahead. Results are mixed though, and seem unrelated to bandwidth and latency results. The complete 3D06 score is as follows:
Again a strong showing from the OCZ Micron kits. Differences are hardly worth mentioning though.
Now for PCMark 2005:
Again very good results for our OCZ kits. The Dominator PC8888F finishes last here.
Finally, our Prey tests:
Mixed results here, without a clear winner. The OCZ PC9200 comes out on top and has very convincing results overall for the cas 3/1T benchmarks.
Our 2T results are on the next page ...>