2GB against 1GB: Performance Analysis
Synthetic Benchmarks: PCMark05, 3DMark06 and ScienceMark 2.0
We decided to start our tests of systems equipped with different amount of RAM with the popular synthetic benchmarks.
PCMark05 uses the same algorithms as real applications to estimate the system performance. However, it doesn’t require big memory resources: platforms with 1GB and 2GB of RAM onboard demonstrated pretty much the same results here.
At the same time the system with four 512MB DIMMs stands out a little bit. This memory configuration works slightly slower than the systems with only two memory DIMMs installed. This phenomenon is a great illustration of the performance drop caused by the forced 2T Command Rate timing setting that jumps in place of 1T in this case. With more memory modules in the system the number of managed memory banks increases and hence the memory bus of the Athlon 64 processor gets loaded more intensively. Unfortunately, in this situation the memory controller has to switch to a less aggressive working mode to ensure system stability. So, as you may see, the influence of the memory size increase can be twofold: installing more RAM can not only speed up the system, but also slow it down in some cases.
PCMark05 test contains an individual benchmark for the memory subsystem. Let’s take a look at the results we can obtain here.
It is evident that the subtest from PCMark05 suite also uses less than 1GB of RAM. Therefore we cannot notice any significant performance difference between the platforms equipped with different amount of RAM. However, when it comes to a system with four double-sided DDR400 SDRAM modules, we see the negative effect from 2T Command Rate setting very clearly.
The recently released synthetic 3DMark06 benchmarking suite serves primarily for the graphics subsystem performance analysis. However, the developers position it as a means for evaluating the overall gaming performance of the system, so we decided to run it today as well.
However, the results obtained in 3DMark06 proved just slightly dependent on the amount of RAM in the tested systems.
The CPU test from the same benchmarking suite measures the system performance during game physics calculation and game characters AI processing. These results are very similar to what we have just seen in PCMark05. As we can see, in this case 1GB of RAM is quite enough for comfortable functioning.
The last synthetic benchmark we are going to use for our today’s investigation is ScienceMark 2.0. It will allow us to evaluate if we really need more than 1GB of RAM for typical physical calculations (mostly used for molecular dynamics projects).
The situation is the same again. This test application doesn’t reveal any real advantages of having large amount of RAM installed. However, it allows to clearly see the evident drawbacks of having all four memory slots occupied.
I have to mention here that one of the arguments the CPU makers use in favor of the shift to x86x64 architecture is the possibility to have over 4GB of RAM in the system. However, at this time we don’t even see the need for 2GB of RAM. Strange, isn’t it? Anyway, it is still too early to draw any conclusions, especially since we have only run a few synthetic benchmarks. Let’s take a look at the results our platforms will show in real applications.