Encoding, Compression, and Final Thoughts

In an effort to see exactly where memory speed might make a bigger impact in typical use with this system, I benchmarked performance with a couple of common applications. First up is Handbrake. For this test I used a longer MPEG-2 video file (89 min), and encoded it using the “Normal” profile.

Handbrake finished slightly faster with the DDR3-2666, but did not scale noticeably from 1600MHz to 2133MHz. This application is almost completely CPU bound, so no surprises here.


Next we look at 7-Zip performance, using both the included benchmark and a ‘real-world’ compression test using a large folder. First up are the results from the built-in benchmark:

Higher memory bandwidth increased performance in both compression and decompression, though single-threaded decompression did not seem to scale past 2133MHz.

Next, I measured the time needed to compress a files on my OS partition. In this case I made a 3.1GB folder from the contents of my Windows DVD, and used the .7z format at normal compression.

Some good gains! In both the manual and built-in benchmarks 7-Zip compression benefitted greatly from the successive increases in memory bandwidth.

 

The X Factor: Integrated Graphics

It must be said that ultimately this review only covers half of the story. So far we have looked at the effects of memory speed on a very fast system with a dedicated GPU, and in the gaming benchmarks the additional memory bandwidth has been of little benefit. However, in a system built around an AMD APU, for example, DDR3 speed becomes of vital importance for gaming since the integrated GPU is using system memory for its frame buffer.

The results of the benchmarks on the Intel test platform may have been underwhelming, but the system was representitive of the kind of components highly-overclocked RAM like this HyperX kit might be paired with, as Kingston clearly targets the high-end market segment with a part like this. It would be worth revisiting this memory on an AMD platform to look at APU graphics scaling, though the budget-friendly APU’s from AMD would likely not be paired with this particular Kingston kit in the real world.

For now, I thought it might be interesting to look at graphics scaling in a couple of benchmarks using the Intel integrated graphics on the test system's Core i7 4770K. While the HD 4600 is a very low performance part compared to an AMD APU, it is still completely dependant on system memory bandwidth. For these quick tests I used the newest Intel graphics beta drivers for the HD 4600 (version 15.33.64.3464), running each at lower quality settings at both 1920×1080 and 1366×768.

Though very minor, the average frame rates did increase with memory bandwidth in both the Valley benchmark and Thief. I was expecting a little more than this, but with only 20 shader cores in the HD 4600 I don't think memory is the bottleneck here!

 

Final Thoughts

As tested, there were only a couple of instances where memory speed improved performance appreciably, such as the CINEBENCH OpenGL rendering benchmark. In this case the higher speeds made a significant improvement even over the short duration of the test. The 7-Zip file compression test showed how important memory bandwidth can be, though many other common applications (such as Handbrake) are going to rely almost exclusively on CPU horsepower.

In the gaming tests with a dedicated GPU, there was occasional benefit in frame rates, but nothing significant. This helps to illustrate that as long as there is sufficient overhead in your system, your performance in games is likely to be limited exclusively by the GPU. In fact, for a modern system with a dedicated GPU, I think system overhead is the biggest takeaway here. Even for a high-end system like this Z87 setup with a Core i7 4770K, 1600MHz is actually a very good memory speed as it is, and really doesn’t hold the GPU back.

The Kingston HyperX Predator 2666MHz DDR3 kit tested today is definitely fast, and it performed up to the rated specifications without any special adjustments. I simply set it to “X.M.P. Profile #1” in the BIOS, and it worked flawlessly over many hours of benchmarking with no system instability. Kingston uses carefully screened Samsung IC’s in these modules, and they are very likely capable of being pushed to higher frequencies than I was willing to experiment with for this review. If memory bandwidth is paramount for the workload that you have, I see no reason to discourage you from this kit.

Bottom line, memory is more than just a check box on a system build, but past a certain point any additional performance will come down to either the needs of your build (i.e. integrated or discreet graphics), or specific applications.

« PreviousNext »