Most gamers are aware of memory bottlenecks on the latest 3D graphics cards, where the only way to unleash the full potential of the chip is to gradually increase memory clock speed, and use increasingly exotic (and more expensive) flavours of memory. What they may not be aware of is that with the latest 1GHz+ processors from AMD and Intel, system RAM has also become a performance variable. Motherboards have started to allow users to tweak RAM latencies, and the very way in which they deal with data. The performance memory market has grown out of this now widespread consumer need for "faster memory", and has been recognised by large hardware developers all over the world. Today, we're going to compare memory produced by two of the world's finest; Crucial, and Mushkin. Both companies now have UK/European distribution deals in place. Crucial supply direct and via resellers (most notably Overclockers UK), whereas Mushkin is only available in the UK via hardware e-tailor The Overclocking Store.
What do the settings mean?
If you feel you are fairly well versed in the ins and outs of memory and the various boring acronyms that follow it, feel free to skip to the next section. Otherwise, open your mind and get your head round this little collection of brainteasers. Front Side Bus (FSB) - at its most basic level, front side bus is one half of the equation which calculates your processor's clock speed. For example, AMD's latest Athlon uses a multiplier of 10, and a front side bus of 133MHz, resulting in a speed of 1333MHz. Column Address Strobe (CAS) - if you want the technical explanation, the CAS value is the amount of time it takes for a signal sent from a memory controller to arrive at a DRAM circuit to indicate that the column address lines are valid. Of course, all the user really needs to know is that a smaller number here produces higher throughput (an added complication is that memory can often be overclocked to a higher FSB if the CAS timing is less aggressive). By manipulating these two variables we can improve system performance. The question is, where the trade-off lies. In our experimentation, it became increasingly obvious that retaining a low CAS timing yielded better results than sacrificing it in favour of a higher FSB. Hence memory rated at CAS 2 is just as if not more important than outright overclockability.
Motherboard - IWill KK266R
Processor - unlocked AMD Athlon 1333MHz
Cooling - ThermoEngine V60-4210 with 6,000 RPM Delta fan
Graphics - ATI Radeon 64Mb ViVo
PCI Cards - none
Operating System - Windows 2000
Software - SiSoft Sandra 2001se
We chose the IWill KK266R over its nearest competitor, the ABit KT7A after initial testing showed it incapable of booting at clock speeds above 133MHz. This may well be down to the individual board that we have, but we wanted to eliminate it as a variable. The IWill proved capable of FSBs up to 165MHz, and reports indicate beyond that elsewhere. The processor needed to be unlocked (a procedure outlined here) so that we could explore obscure FSB and multiplier combinations in our testing. Intel chips are multiplier-locked and there is no known workaround. FSB and CAS timing adjustment is available on Intel-based motherboards, but since we could not reduce the multiplier as we increased the FSB, we rapidly reached the physical limit of the core before maxing out the memory. As a result of these findings, our entire test is based on an AMD platform. In terms of benchmarking, we relied upon the industry standard memory benchmark, SiSoft Sandra 2001se. A success at any given clock speed required full boot, no error messages within Windows 2000 and successful completion of the Sandra memory benchmark.
As you can see (or read on the mouseover caption), the difference in performance at 133MHz FSB is negligible, and certainly within experimental error (2%). This shows that with identical settings there is no appreciable performance variance between the two brands (a trend that holds true across the full range of FSBs tested). With this in mind, the winner will be the memory which can reach the highest FSB.
In testing speeds above 133MHz our original stick of Crucial had problems achieving high FSBs. After swapping this for an identical stick, we managed to achieve the speeds outlined below. This shows that though Crucial guarantee to operate at 133MHz CAS 2, speeds above this vary dramatically. It should be noted that the Rev. 3.0 Mushkin used here is guaranteed to 150MHz CAS 2. Mushkin claim to hand-pick the individual memory chips thus identifying the cream of the crop. After exhaustive (very, phew -Ed) testing, the maximum FSB speed achieved by the Mushkin was 157MHz, with the Crucial just pipping it to 160MHz. As you can see from the results, the actual difference in performance is negligible, as expected given the mere 2% difference in FSB. This was the limit using CAS 2 latency, but theory dictates that switching to CAS 3 may allow for higher FSBs. In practice, this increase was itself fairly small (8MHz for the Mushkin, and 4MHz for the Crucial). Despite these gains, measured performance decreased by approximately 6% for both sticks of memory, proving our theory of CAS latency being more influential on performance than FSB.
The overwhelming conclusion from this head-to-head is that FSB speed at CAS 2 is more important than ultimate FSB speed if it requires dropping the CAS latency to 3. This is evidence of a growing trend where raw MHz is no longer King. Having proved that there is no difference between brands at identical settings, the most important factor is the highest FSB at CAS 2. It is difficult to draw a clear conclusion as the decision is ultimately up to the user: the Mushkin is the more expensive choice, but is guaranteed to 150MHz at CAS 2, and there may be some headroom. Crucial is markedly cheaper at the moment, but varies considerably in its ability to clock above 133MHz at CAS 2, which is understandable as Crucial only guarantee it to that. We will be sure to take another look at this topic in the future as memory manufacturers refine their fabrication processes, yielding even higher clock speeds.
Get your first month for £1 (normally £3.99) when you buy a Standard Eurogamer subscription. Enjoy ad-free browsing, merch discounts, our monthly letter from the editor, and show your support with a supporter-exclusive comment flair!