CPU overclocking and it’s effect on perfomance is a well covered subject but memory overclocking is less well known and is usually just a side effect of CPU overclocking. I though I’d take a more indepth look.

I decided to run do my overclocking and run my benchmarks on an overclocked CPU as it’s unlikely that you would overclock the CPU but not the memory. So I set my E2160 to 3 ghz with a 333 mhz fsb and a 9x multiplier which allowed me access to memory settings of 800, 1066 and 1333 mhz which are common memory clocks. I then did a further test with the fsb set at 375 mhz and the multiplier dropped to 8x, this kept the CPU clock speed at 3 ghz but allowed me to access a 1500 mhz memory clock.

Benchmark Testing

To test the performance of these overclocks I ran some of my usual benchmarks. For a measure of raw memory speed I ran the Lavalys Everest memory benchmark, to measure whether this leads to increased CPU processing power I ran 1 million SuperPI and to test real life performance I ran the Custom PC media benchmarks and the Futuremark 3DMark05 and 06 benchmarks. I also ran the half life lost coast video stress test but the fps results were so similar that I couldn’t really deduce much, however there was a defined increase seen by moving up from 800 mhz to 1066 mhz others by just over 1fps and a slightly less clear increase to 1333 mhz but the margin for error was so small I decided not to report these.

First off then is the everest memory benchmark. This just measures raw memory speed with regards to writing, copying and reading to and from memory. You would expect these to show the biggest change as the memory frequency increased.

As expected there is a definate increase in performance as the frequency increased. It is interesting however that the read performance increased markedly with frequency increase but the write performance didn’t. In fact when you consider the write results it seems that they only really respond to the increase in the fsb which was required to run the memory at 1500 mhz. So if this is the case then the increase in the write speed shouldn’t be regarded as part of this test as I am not looking into how fsb affects performance in this study. Of course don’t be suprised to see a study into how the fsb affects performance cropping up on this site sometime soon.

The second test I performed was my favourite raw cpu power test, SuperPI. I like this test as it only takes a very short time to complete, especially now that CPUs are getting quicker and quicker. I run the 1 million decimal place test as this seems to be the community standard.

As you can see from the graph, the overclock on the memory does provide a performance increase. It is far less than the increase in performance due to CPU overclocking but it is an increase all the same. It is also clear that there is a bigger difference in performance between 800 and 1066 (1.420s) than there is between 1066 and 1333 (0.344s) even though the increase is 266 each time. This suggests that there is a larger performance gain to be seen up to a certain point and then less beyond that point. This is after all what you would expect as the memory reaches a speed where it is supplying the CPU with data as quick or quicker than the CPU can deal with it so any additional memory speed is wasted. Of course if you increase the CPU speed then you get back to a point where the increase in the memory speed makes a difference again and this test was done using a E2160 CPU @ 3ghz which although decent enough it isn’t exactly the fastest setup.

So after completing the synthetic benchmarks I moved on to the more useful real life benchmarks. Obviously it’s impossible to have a truely real life benchmark as it’s still a benchmark and isn’t real life! However, the media benchmark from CustomPC is one of the most real benchmarks that I’ve come across as it does some image editing, a bit of video encoding and then multitasks by playing back the encoded video whilst archiving some files.

It would appear that the memory overclock makes quite a difference for image editing and for the multitasking tests but not so much for the video encoding. The media benchmarks also mimick the SuperPI test in that the performance increase lessens after 1333mhz has been reached. The results even show a decrease in the video editing performance when going from 1333 to 1500 although I would expect that this is an anomaly due to the margin of error in the test.

My final benchmarks for this study are the Futuremark 3dmark benchmarks to see if gaming performance is affected by memory overclocking. My expectation here was that there would be a small increase but nothing significant. I ran 3dmark05 and 3dmark06 on the standard settings for the free version so that it’s comparable for the majority of people.

And the results bare out my expectations, there’s very little difference between the results except for the 800mhz setting which yet again trails behind. This test is almost entirely graphics card dependant its actually a suprise that the system memory overclock makes any difference at all. The differences are slightly larger on the cpu test as you would expect and the 800 mhz setting again shows it’s weakness.

Conclusion

Overclocking your system memory does increase the performance of your PC, however the performance increase is not a great increase even with what seems to be a massive overclock. This is because the memory doesn’t perform any calculations like the cpu and gpu do, it just supplies the cpu with the information required. The speed at which it can supply that information has been seen in this study to have an affect on overall performance but only by a small degree.

So it seems from the results of my study that it’s important to have memory that can keep up with your CPU or at least to be able to overclock your memory to keep up with your CPU but any extra isn’t really worth paying out for. I’m not sure how far DDR2 can go with regard to memory clock speeds but if my E2160 benefits from memory clocked at up to 1333mhz then I’m sure that the later and higher ranged CPUs will benefit from 1600 and above, especially in the short term future. This I guess is a good thing as the price of DDR3 has just recently plummeted and DDR3 motherboards now have high availability, so it’s looking like DDR3 should be included in any current and future motherboard upgrade plans and when overclocking, concentrate on CPU speed and once that’s been found optimise the memory for that setting.

Popularity: 1% [?]