Archive 1997–2024

ASRock 775Dual-VSTA: Does DDR2 matter?

Memory Performance

Our VIA based ASRock motherboard provides two DDR2 and three DDR memory ratios. The majority of end-users will select the memory ratio that matches their memory speed. We are testing four ratios at the fastest stable timings we can achieve and still pass our benchmark test suite. With these set ratios, CPU speed remains the same at 1.86GHz in our test platform with memory speed being varied by selecting the different ratios.

There are some downsides to this approach. With the memory controller in the chipset, instead of part of the processor as in AMD Athlon 64 systems, there is a small performance penalty for speeds other than a 1:1 ratio (DDR2-533 in this case). However, the penalty is in reality very small, though the performance between various chipset designs can vary a great deal as we will see in a future article.

Due to performance reasons we did not test the DDR266 setting as we believe most people will not need this setting. Although the BIOS offers a 1T Command Rate, we never could get this setting absolutely stable without drastically raising the memory latency settings. We even tried our more expensive memory modules with the same results. Our memory settings were derived from extensive stress testing with a variety of applications. While certain settings that allowed lower latencies worked well with certain applications, the final settings we arrived at had to work with all applications.

Click to enlarge

The memory performance result at DDR2-533 (1:1 ratio) leads the other memory speeds in all cases. This particular setting delivered the best raw performance although we will see in our application and game benchmarks that this advantage is negated by other platform components.

Normally memory bandwidth improves with increases in memory speed and reductions in memory timings. To evaluate memory bandwidth SiSoft Sandra 2007 Professional was used to provide a closer look at scaling. As we have been saying for years, however, the Buffered benchmark usually does not correlate well with real performance in applications on the same computer. For that reason, our memory bandwidth tests have always included an UnBuffered Sandra memory score. The UnBuffered result turns off the buffering schemes, and we have found the results correlate better with real-world performance.

In this case, we find the combination of our memory settings at DDR2-667 and the VIA memory controller generates Sandra results that are up 32% lower than the DDR2-533 settings. The DDR2-533 results are up to 48% better than the DDR-333 scores with the DDR2-667 and DDR-400 scores being comparable. The Sandra memory score is really made up of both read and write operations. It is also a synthetic benchmark that does not always reflect real world performance.

To provide more detail on the impact of memory performance, we also compared pure number crunching with Version 1.5 of Super Pi, using the time to calculate 2 million places of Pi at the different memory speeds. Our other memory test is the latency measurement from the latest version of Everest. The results are interesting as the DDR2-667 ties DDR2-533 in Super Pi while DDR-400 trails slightly and DDR-333 brings up the rear. Looking at the latency figures, DDR2-667 is almost equal to DDR2-533 and DDR-400 trails slightly, while the results for DDR-333 are terrible and follow the Sandra bandwidth figures when compared to our DDR2-533 scores.

Originally published at https://www.anandtech.com/show/2058 (Wayback Machine).