Is Fast Memory Really Worth It?
Memory vendors have become excellent at marketing their latest high-end products: DDR3-2000 speeds are currently considered state of the art for enthusiast Intel platforms based on Intel’s P35, X38, X48 chipsets or the new Nvidia 7 series. But how much sense do these products really make? While mainstream DDR2 memory has reached almost ridiculously low price levels - you can get two 2 GB DDR2-800 DIMMs for less than $80 - DDR3 memory at 1600 speed or faster easily costs five times as much, without delivering even double the performance. In fact, for the vast majority of users, the difference between mainstream and high-end memory turns out to be extremely small.
The importance of Random Access Memory (RAM) has changed a lot over time. There were noticeable performance differences between CL2 and CL3 timings back at the turn of the millennium, when first generation SDRAM at PC100 or PC133 speeds were popular. But now, the performance delta between quick and very relaxed timings is almost negligible in using fast DDR2- or DDR3-SDRAM. Although memory latencies seem to have increased from one memory generation to the next (CL2/3 with DDR1, CL3-5 with DDR2, CL5 and up with DDR3), the latencies haven’t changed much, as the clock speeds have doubled with each generation change. The effective latencies hence remained very much the same, while throughput has increased considerably. (Compare Prices on DDR)
Enthusiast memory does have another justification that is only indirectly based on performance: overclockers need maximum flexibility from components when they want to squeeze out maximum performance from their systems. Increasing the system speed is often the only way to increase the CPU clock, which will automatically accelerate the memory as well, as its clock speed directly derives from the system clock speed. Since you don’t want to be performance-restrained by limiting memory speed, fast memory may be necessary to achieve maximum system performance. This scenario is only valid for hardcore overclockers, however, as the benefit of fast memory over slower has become very small if the other components and parameters remain unchanged.
We wanted to know how important RAM speed and timings really are. To test this, we assembled a Socket 775 system, which we operated using two different processors: a brand new 3.16 GHz Core 2 Duo E8500 based on the 45 nm Core 2 Duo Wolfdale core with 6 MB L2 cache; and a 3.73 GHz Pentium 4 Extreme Edition single core processor. We decided to include the old single core Netburst P4 processor since it offers a smaller and less efficient cache memory than the Core 2 Duo. Both processors were benchmarked at DDR2-667, DDR2-800 and DDR2-1066 speeds as well as DDR3-1066 and DDR3-1333, each time using slow and fast timings. The P4, however, could not be benchmarked at DDR3-1333 speed, as it would have required FSB1333.