The MemoryC Guide to Memory Page 3
However, at our overclocked setting of 400MHz, these dividers now provide the following settings:
(1/1) x 400MHz FSB = 400MHz (800MHz effective – PC2-6400 or better needed)
(6/5) x 400MHz FSB = 480MHz (960MHz effective – PC2-8000 or better needed)
(8/5) x 400MHz FSB = 640MHz (1280MHz effective – even PC2-10,000 isn’t enough!)
Now you can see why overclockers need fast memory for the best performance!
Fortunately, many chips use an FSB slower than 333MHz, with popular overclocking processors like the Core 2 Quad Q6600 running on a slower 266MHz FSB. At this FSB, the following dividers are available:
PC2-5400 memory (667 MHz effective, 333 MHz actual) = a 4:5 ratio
PC2-6400 memory (800 MHz effective, 400MHz actual) = a 2:3 ratio
PC2-8500 memory (1066 MHz effective, 533 MHz actual) = a 1:2 ratio
Some boards allow you to use any divider with any processor by having an adjustable FSB strap option. For example, if you install a 266MHz FSB processor, but set the strap to a 333MHz FSB option, you can set your 266MHz FSB processor up with 1:1, 5:6 and 5:8 dividers. This is very useful when overclocking, as it allows you to keep the memory speed under control at higher front side bus speeds and gives you more options to play with in order to get the best balance of clock speed and stability.
Just because a memory kit is rated to a certain speed, this does not mean it will be stable at any FSB/memory divider combination that provides that setting. That is because faster FSB settings put a greater strain on the chipset than lower FSBs and more aggressive multipliers. For example, a kit of PC2-8500 might not be stable if you are using an FSB of 533MHz with a 1:1 setting, but will work perfectly fine at 333MHz with a 5:8 setting. In general, in order to get the best out of an exceptionally quick kit of memory, you need exceptionally high end motherboards.
When buying overclocking memory it is best to have a goal in mind and buy the correct memory for the job. For example, if you plan on buying a Q6600 and running it at 3.2GHz (355MHz FSB), a perfect kit of memory would be a kit of PC2-8500 and running it with a 2:3 ratio. This will result in a memory speed of 1066MHz precisely. Do not worry if your goal doesn’t add up to the exact rated speed of a memory module, so long as it is under the rated speed, the memory should work fine.
Voltage – More is not always better
In many cases a higher voltage put through your components will allow you to attain a higher clock speed. This is particularly true of processors and motherboard chipsets. Memory on the other hand is a rather different beast, with some modules being very amenable to a higher voltage, allowing you to obtain higher speeds than you could at default, whilst others either don’t work at all at higher voltages or become unstable. When you buy premium memory from the likes of G.Skill, GeIL, OCZ, or other performance manufacturers, the hard work has been done for you. All of these manufacturers provide you with information on their memory with the correct operating voltage available on their website, or even printed on the modules themselves. Sometimes this is a single value which you should adhere to, or more often, it is presented as a range. For example, many 4GB dual channel kits work best at 1.8-1.9V. This is because the high-density modules they use do not really like voltages higher than 1.9V, becoming less stable and less able to overclock.
How to Test Stability
When installing new memory to an existing machine, or when overclocking, you should never use your main OS installation to test for stability. This is because memory that isn’t stable can badly corrupt data and cause irreparable damage to existing files. Fortunately, there are ways of testing memory stability other than just using your PC. One of the most useful tools is Memtest. The latest version of this application can be downloaded free of charge from www.memtest.org
in versions designed for either a bootable CD or floppy disk. After burning the image file to your chosen media, set this device to your primary boot device and allow it to boot. Memtest will then run through it’s default battery of tests which flag any errors detected in red. If you run through a kit successfully once, you can be fairly confident it is error free. To be fully confident however, you may want to consider testing it for six hours or more continuously.
Whilst Memtest is very handy, it isn’t infallible. Sometimes you get a kit that runs perfectly through the test without any errors but is unstable in windows. A handy windows-based utility for testing memory is Orthos. This can be downloaded from and is based on the Prime95 tester which is another alternative. Orthos is designed to test for both memory and CPU errors, and problems with either will result in errors in this program. If your system isn’t 100% stable when running Orthos, it isn’t 100% stable period! A run of an hour is cause to be confident that your memory is stable, but for full peace of mind, consider running the test over night. There are a number of different modes when running either Orthos or Prime95, and the correct mode for testing memory is Blend. This tests primarily memory but also CPUs, and is therefore an invaluable tool when overclocking your PC.
Optimising the Memory Performance of an Overclocked Machine
As you will have probably guessed by now, optimising the memory performance of a computer system is a careful balancing act of getting the right mix of optimal speed and timing settings, all the time staying within the bounds of the kind of CPU overclock you are aiming for. Often a memory kit will be capable of a higher maximum operating speed if you slacken the timings off, but you need to make sure the trade off in timings yields enough of a gain in raw bandwidth to make it work this compromise. There is no sure-fire way of finding the perfect balance apart from by trial and error, and there are a number of tools available that allow you to test memory performance to pinpoint that perfect mix. One of the best tools is Everest, which allows you to test bandwidth and latency. Another popular tool is Sandra, though the default test seems slightly biased towards a higher speed than tighter timings which is not necessarily indicative of real-world performance. A good test to see how your memory timings affects real world applications is to take a game with a built in benchmarking facility, set all of the video settings to their most basic and compare the average FPS returned. The CPU benchmark included in Crysis is a good example here, as well as some of the older 3DMark series of benchmarks available from futuremark (2001 being particular popular for comparing memory performance.
And you’re done!
Hopefully you should now feel confident enough to get the best out of the new memory you have purchased. As always, we welcome customer enquiries using our contact form and will endeavour to make suitable recommendations where possible.
: Ian Jackson
(part of the GoBeyond Group)