To be thorough I ran the memory under SPD, Manual, and Auto with very little difference among results. Since the average PC-owner most likely will not adjust their timings manually, all tests were run on Auto. This equates to CL4-4-4-11, I also attempted to tighten timings several times without success. I believe Abit's AA8 BIOS 17 leaves something to be desired, although in their defense the DDR2 standard is still fairly immature. I'm sure this isn't an Abit issue specifically, as other DDR2 motherboards suffer from similar BIOS problems. The inability to lock the PCI-express while troublesome isn't memory related, however, being unable to change the memory divisor without experiencing a system lock-up is.

Beginning with SuperPi I ran the system at default speed 3.4GHz overclocking to 240FSB or 4.0GHz, and finally 250FSB or 4.2GHz. In the SuperPi chart below, the numbers represent the amount of time in seconds required to calculate Pi to 1-million places.



Next I ran Cachemem which measures latency, and bandwidth among other aspects. Cachemem seems to squeeze bandwidth fairly tight; resulting in lower scores compared to Sandra or Aida32 all things being equal.



At 4.2GHz the READ result was just slightly lower then at 4.0GHz. I repeated the test and in each case the result was the same. The memory was just running at its rated speed therefore I can only attribute this to a possible NB-MCH (North Bridge Memory Controller Hub) anomaly. Our next benchmark is a feature of ScienceMark 2.0 which records many memory nuances. For A64 user's I suggest running L1, and L2 cache options as well.



Next we move onto the synthetic benchmark Aida32 which as your about to see represents an interesting dichotomy. Aida32 consistently produces the highest bandwidth score compared to other tests.



For our final bandwidth specific benchmark SiSoftware Sandra 2005 the results consistently fall midway compared to Aida32 and Cachemem regardless of the memory tested. I believe SiSoftware Sandra is by far the most accurate (albeit synthetic) test for determining bandwidth.



I used two Futuremark benchmarks: PCMark04 and 3DMark2001 SE in the charts below.




Due to the high latencies native to DDR2 I thought it unnecessary to run real-world gaming tests. In a stand alone review such as this it would be difficult to determine where system memory is pertinent given the ATI X800 XT runs 256MB of on-board memory itself. In my upcoming DDR2 round-up, however; I will employ games such as FarCry, DOOM3, Halo, Half Life 2, as speed and latency differences should produce tangible results.



Conclusion:

Mushkin PC2-5300 Enhanced Memory offers an excellent product for anyone seeking DDR2-667MHz memory. Mushkin sells its memory direct and the memory tested today in 1GB kit form is available for just $274. Running at a constant 4-4-4-11 this memory was competitive to a few other brands which suffered from a propensity to lax timings running slightly beyond their rated speed, and some even at speed.

When the package advertises specific latencies at such and such a speed, the memory in question should maintain those timings under SPD or Auto BIOS settings. While such issues could be corrected via a simple manual BIOS adjustment, this is somewhat presumptuous. Most PC-users even those purchasing PC2-5300 do not make such daedal BIOS tweaks they want to plug and play. Mushkin has offered DDR2 devotees a high performance product, easy to use, offering a lifetime warranty and does so at an affordable price. Highly Recommended!

I'd like to thank Lee at Mushkin for making this review possible.

Questions/Comments: Forum thread