All overclocker PC memory falls into two categories: one provides relatively low timings at a low maximum frequency like 233MHz (466DDR) and another allows working at 250MHz (500DDR) and higher, sacrificing timings. Both approaches are generally equal, providing similar performance under the same conditions, if we don’t specifically seek for exceptions to this rule.
The market is not yet accustomed to DDR500 and faster memory modules. However, there are several companies offering such products already – just enough for us to carry out a comparative testing session. We picked up memory modules from six manufacturers: Corsair, GeIL, Hynix, Kingston, Transcend and SimpleTech.
So, we are set to seek out the most overclockable memory module that would work at the maximum possible frequency. Sticks of the DDR500 type are physically incapable of keeping the timings low, whatever the memory voltage is. The minimum timings for the most of the tested modules to work in the DDR500 mode and at 2.9V Vdimm are 2.5-6-3-4. Whatever we did, we just couldn’t make our modules dance to the enchanting 2-5-2-2 time even with the Vdimm = 3.4V.
Meanwhile, there is a small speed difference between 2.5-6-3-4 and the DDR500-standard 3-8-4-4 timings, and it is well compensated by the higher operational frequency. Thus, at least for Intel processors and 32-bit Athlons, it’s better to have memory that works as DDR566 with 3-8-4-4 timings rather than one that keeps 2.5-6-3-4, but as DDR533 (this is not a mere supposition, but a real picture confirmed by benchmark applications; further information will be given in our exploration of memory performance with different frequency/timings combinations, now being prepared for publication). That’s why our testing is reduced to searching for the maximum memory frequency in the synchronous mode (1:1 of the FSB clock rate).
As for lower timings, all the tested modules were found capable of working as DDR500 (DDR533 for GeIL PC4200 and SimpleTech PC4000) with 2.5-6-3-4 timings. We also found that the manufacturers are very conservative about the nominal timings for their sticks. Otherwise, we left other timings alone and clung to 3-8-4-4.
Our exploration tool is the most fastidious benchmark for the system RAM: Lobby High Detail (Game 3) from 3DMark 2001 SE b.330 suite. We publish the maximum memory frequency at which the testbed went through the test, with a precision of 1 megahertz. The frequency may be made a couple of megahertz higher for passing SiSoft Sandra Memory Benchmark, but the system is not truly stable at that.
We carried out our tests in the dual-channel mode only. On the one hand, we reduce the randomness factor (modules differ among themselves – some are prone to overclocking, others aren’t), but on the other hand, we check the memory for compatibility in the most relevant operational mode for today’s computer systems. Moreover, some modules came to us in “coupled” packages, originally intended for working in the dual-channel mode. If you are an owner of an Athlon 64 processor, you will achieve higher frequencies since the processor-integrated memory controller is a single-channel one.