Test results

Our tests demonstrated that AMD Phenom-based systems can reveal full potential of high-frequency dual-channel DDR2 memory. This time we decided not only to test memory modules, but also to compare performance of memory controllers in B2 and B3 revisions.

We run our tests with the integrated memory controller in Phenom processors operating in ganged mode, which provides higher results in a single core access mode, that is in tasks critical to execution speed of a single thread. If you run multi-thread applications with equal priority, you'd better use the unganged mode. Command Rate is set to 2T. All timings, except four main values, are chosen by BIOS automatically.

^* Block size is 32 MB.

As it makes little sense to buy high-frequency modules and use them at reduced frequency, we decided to use DDR2-1066 with the main timings as the minimal mode. The initial voltage was 2.1 V, but it turned out that our modules worked fine even when we reduced voltage to the standard DDR2 level of 1.8 V.

The next stage was to test our memory modules in the DDR2-1142 mode with 5-5-5-15 timings at 2.1 V. We had to overclock the CPU a little. However, the system was stable without raising voltages and reducing multipliers.

As an additional test procedure, we decided to determine minimum stable voltage for each mode. As is well known, reduced voltages noticeably affect heat release, power consumption, and contribute to reliability and service life of memory modules. We were pleasantly surprised - our memory modules were stable at maximum frequency, while the voltage was only 1.94 V, a starting value in other overclocker modules!

However, attempts to overclock the modules above the recommended maximum level revealed an insignificant frequency potential. Nothing helped to increase it - softening timings or raising voltage to 2.3 V. In practice we managed to increase bus frequency by a single MHz (from 214 MHz to 215 MHz). Solely for explorative reasons we tried to reduce voltage again and unexpectedly obtained stable operation at 216 MHz and 1.94 V. However, this mode is useless, because softer timings hamstring bandwidth gain, provided by the increased frequency, because of the increased latency. We noticed performance gain only in case of multi-threaded access, which utilizes memory bandwidth to full extent.

Comparison of results demonstrated by Phenom B2 (without TLB Patch) and B3 allow us to conclude only that both memory controllers operate identically.

We also published results of the previously tested 2GB memory kit from Chaintech, which was also successfully overclocked to DDR2-1150. On the average, Corsair memory demonstrates higher results. To be more exact, Chaintech memory performs a tad better in single-thread mode, but Corsair memory wins in the multi-threaded mode. So we can say that Corsair managed to raise memory capacity and preserve its characteristics on the level of similarly overclocked 2 GB modules.

Conclusions

Corsair again demonstrates its focus on products for people who don't take any compromises and want it all. Namely for those willing want to install 4 GB of memory in two modules and raise its frequency to the level of 2GB overclocker modules. We never doubted that these modules could do it, even before we started the tests. However, it turned out that hard-driving consumers may have another wish - unexpectedly, these modules showed operating stability at maximum frequency with the voltage of just 1.94 V.

Write a comment below. No registration needed!