Most of the test results show less than 1% difference between memory modules. What's the margin of error on those tests? If it's 1% (or, more likely, more), the differences are completely irrelevant and probably not reproducible.

On the overclocking experiments, how long have you run your oc'ed RAM before declaring it works? An hour? A day? A week?

Like most RAM reviews, this is totally meaningless.

Each benchmark is run three times and the results of each test are averaged together. So yes, the results can be reproduced.

I posted a reply and then deleted it--first negative feedback I've ever gotten on a review, and I took it a little too far into a "fuck you so what I fucked your mom with a pipe" direction. Without further ado, a good response:

"Most of the test results show less than 1% difference between memory modules. What's the margin of error on those tests?"

It varies from games around .06% and with synthetic apps to .003%. That is orders of magnitude less than 1% and 1/1200th or so of the average stated margin of error of other review sites. I produced these numbers from my actual paper results, and some quick tests I could do right now.

"If it's 1% (or, more likely, more), the differences are completely irrelevant and probably not reproducible."

It is entirely reproducible, but I won't go as far as to say entirely relevant*.

"On the overclocking experiments, how long have you run your oc'ed RAM before declaring it works? An hour? A day? A week?"

I consider eight hours a successful overclock, since overclocks vary a great deal and it doesn't pay to go for more when the very next kit by the same company will overclock differently. I let things run overnight and check the time stamps the next day. Of the DDR2 I've reviewed only the (upcoming) OCZ memory failed to run all the way overnight, but was still stable for the first 20 hours. Even then, that was only on the stock timings/voltages overclocking test. By testing the RAM at such high voltages, it's easy to get the theoretical best possible clockspeed of the RAM, too. If it doesn't fail in the first ten minutes, it runs for hours. The downside is that 2.4V really will kill RAM, so it wouldn't really be safe without special cooling, and I can't predict how the reader will cool his or her computer.

"Like most RAM reviews, this is totally meaningless."

Tell me what you want and I'll adapt my reviews to follow suit.

*Unless everyone in the world is using the same model number Athlon X2 6000+, with their DFI NF590 SLI-MR2/G motherboards flashed with the same BIOS, also using a PNY XLR8 8800GTX OC, the results of a review can only be compared to like reviews. You have to use your noodle to guess how my results compare to other's results, and also how to apply those to your potential needs. The reason I don't include benchmarking results of the overclocked memory is that it's not really possible to scale the processor in parallel with the RAM.

Finally, the synthetic and real-world results of all these memory kits operate within a few tenths of a percent of each other. What this means is not that the reviews are useless, but rather that all DDR2-800 is similar. So similar as to be practically identical, and there's really no way of telling what kit is inside a computer without opening it up.

There is only one arena where memory behaves elementally differently: overclocking. And by the results I have, I can say that most kits are cut from different cloth. Because of this, I cover overclocking with a good spread of tests. But again, you can't say that everyone's processors will overclock 1:1 with the RAM, so it's fruitless to provide benchmarks of an overclocked system.

For everyone who's not interested in overclocking: just buy the cheapest RAM you can get, from a company you trust, with a warranty you like. It's otherwise indistinguishable.