Test results - bandwidth and latency

First in our benchmark suite are the bandwidth tests for which we use SiSoftware's Sandra application. The System ANalyser, Diagnostic and Reporting Assistant is an information and diagnostic utility which has a very handy benchmark suite, in which you can compare your system to a range of other reference systems. The Bandwidth is one of the available benchmarks. Alternatively for bandwidth measurements, we can use the Everest application from Lavalys, which consistently gives higher results, but the scaling is (almost) the same as Sandra's :



As you can see, the best results are obtained by our 2Gb kit, the Corsair Dominator. This is not really that surprising, as the timings for this kit were lower than the other test candidates. For the second in row there's a difference between Sandra's results and the Everest result : I would have expected the 2x2Gb kit to come in second, as there is a latency penalty for going from 2 modules to 4, however, Everest thought different and gave the 4x2 Gb modules the better score.

In the above graph you can also clearly see the (largely theoretical) benefits of a higher front side bus: going from 333FSB to 400FSB gains us almost 700Mb/s (in Sandra), going from 400Mhz memory speed to 500Mhz (both at 400FSB) gives us another 500 Mb/s (again in Sandra).


Next up is the Latency result, for which we use Everest once more:



Here you can see the reason why Everest shows a higher bandwidth for the 4x 2Gb test: the latency is very slightly lower for the 4x 2Gb kit, resulting in more bandwidth. The latency results are always closely related to bandwidth results, at least when comparing memory modules of the same kind (in this case DDR2), our test results are no exception.

Test results - synthetic benchmarks

Let's get started with a rather new benchmark, the PC Mark Vantage suite. In this suite, there's a specific section for memory performance, and that's the test we'll be running today. PC Mark Vantage tries to mimic real-life applications, and does this by using 4 different memory tests, combined into their "Memory suite":



PC Mark Vantage respects the logical order, by rewarding the tightest latency memory with the highest score, and by penalizing the use of 4 modules slightly over a 2 modules setup.

Next up is 3D mark 2006 :



The results are ordered by cpu-score, but in this case the CPU test does not tell the whole story. When using memory that is fast enough, 3D Mark '06 seems to favor 2x2 Gb or 4x 2Gb over 2x 1Gb. I should conduct more testing to see if this is generally true, but in our tests this was the case.

Let's follow the 3D Mark with SuperPi, very well known to follow bandwidth and latency results rather closely:



SuperPi follows the results of the Everest tests earlier on very closely, favoring the 4x2 Gb setup over the 2x2Gb, when looking at equal memory speeds of course. The fastest setup remains the 2x1 Gb kit with tighter latencies.


Test results - real-world application benchmarks

Let's first look at the Cinebench results, which mimic the Cinema 4D application. We're up to version 10 now, have a look :



In the case of Cinema 4D, results are a bit mixed. 2x1Gb is the fastest, and between 2x2Gb and 2x4Gb the results are tied. I'd say this is good news, as the extra memory will come in handy when working with large files and movies.

Let's move on to Photoshop CS3 now. Everyone knows that Photoshop is a memory eater; it has been so ever since it came to the market. We're running a benchmark suite here which uses a rather small image file, when using larger files you'll come to a point where disk swapping will start to take place, and the application will slow down significantly. So it's clear that the more memory you have, the better. In our tests we never reached the point where swapping would start, not even with our 2x1 Gb memory setup.



Because we never reached a "disk swap state", results are predictable: they follow the Everest and SuperPi results closely.

Another important test lately is the H.264 test. H.264 has created a lot of momentum after the inclusion in Apple's Quicktime, and is considered an important video standard now. Encoding/decoding H.264 is a very cpu-intensive task, and we'll be measuring the impact of the memory subsystem on H.264 performance next:



The first three results are very much tied again: the 2x1Gb, the 2x2Gb or 4x2Gb results differ according to the "pass": the benchmark consists out of two "passes", each presenting a different test. In the first pass, the 8Gb kit is the fastest, in the second ass the 2gb kit.

Finally, a bit of Crysis to get a feel for gaming performance. We measured the "cpu-test" here with medium and high video settings at a high 1920x1200 resolution, and we stuck to the DX9 setting.



With medium quality, the 8Gb kit outperforms the 4Gb setup once more, whereas in high quality, where the video card limits things further, this is not the case. The fastest performance you still get from the 2x1Gb setup, but the differences are very - very small, to say the least.

On to the final words and conclusions ->