To estimate the complex performance rate of our test systems we used two testing suites: graphics Futuremark 3DMark06 and “general performance” CustomPC Benchmark. For our 3D test we chose not the latest version of the popular Futuremark application for a reason. Its results depend more on the processor and memory subsystem performance than the results of 3DMark Vantage. We chose CustomPC Benchmark to replace the traditionally used PCMark Vantage and SYSmark 2007, because the performance index readings it generates do not fluctuate too much. As a result, we can achieve higher measurement precision with fewer test runs when testing similar platforms differing only in memory subsystem.
It would be silly to expect 3DMark06 to demonstrate any significant performance difference between the systems equipped with different memory types. This test serves to measure the performance of the graphics subsystem, which works almost identically with any type of system memory employed, as we have already seen.
However, general system performance measured during work in resource-hungry applications does depend on the memory, and this dependence is often quite significant. The testing suite we used checks the system performance under several types of workload, such as: digital photo editing in free GIMP graphics application, encoding MPEG-2 video into H.264 format using Handbrake utility, data archiving and encoding with 7-zip, and high-definition video playback.
Overall, the performance difference between the tested systems equipped with the fastest (in our today’s test session) and the slowest memory makes almost 5%, which makes us choose the system memory seriously. The memory parameters matter most for its performance under multi-threaded load, when fast DDR3-1333 ensures a 14% advantage over the platform equipped with slow DDR2-800 SDRAM. At the same time systems equipped with DDR3-1333 memory cannot outperform identically configured systems equipped with the best DDR2 SDRAM. In fact, the tests in real applications repeat the results obtained in synthetic benchmarks. DDR3-1333 with 9-9-9-27 timings can only be regarded as a performance alternative to DDR2-800 with 4-4-4-12 timings, and DDR3-1333 with more aggressive 7-7-7-20 timings works as fast as DDR2-1067.