Patriot Division2 Viper Xtreme 4GB DDR3 RAM kit
PCMark shows us surprisingly little information about the memory modules in slots 0 and 1, and we did not bother with slots 2 and 3, which for the purposes of these tests were unpopulated. CPU-Z shows us a great deal more, mostly confirming our specifications and specifically confirming XMP profile support.
What we have here is the first benchmark any user is likely to be exposed to, Windows 7’s built-in Windows Experience system. The memory is clocked at an impressive 7.5 on the arbitrary scale used by Windows. This is pretty darn fast, though it’s unfortunate that they provide so little information about the relevant metrics. SiSoft’s Sandra is rather more explicit about what it measures, however. Detailed at top right, the Cache And Memory benchmark shows us an overall score of 71.8 gigabytes per second of data bandwidth, though the “speed factor” of 72.0 is nearly as poorly explained as Windows Experience (though they say lower is better). The detailed view shows a breakdown of RAM bandwidth compared with data set size; these modules hit their stride at 64 kB chunks of data, at over four hundred and thirty gigabytes per second. They drop off pretty precipitously after that however, as significantly more memory pages are required per memory transaction – remember, this is a log(2) scale, the data set doubles in size with every additional notch on the horizontal scale, and the memory holds up better than the exponential decay curve one may expect under such circumstances. Clockwise from the previous, we have memory bandwidth, the ability to push around floating-point and integer variables. Oddly, the previous benchmark had significantly different values for each module; this time, the results are much more even. The results seem very slightly biased in favor of floating-point operations over integer operations, but otherwise are extremely close. Remember this graph, for we will be revisiting it later. Lastly, moving clockwise once more, we come to memory latency, in terms of clock speed and data-set size. We find a flat four clock cycles minimum for a memory transaction when dealing with extremely small data sets, though as data sets double in size, we find an odd stair step curve; this is at least partially the fault of the logarithmic curve on the horizontal axis, and the extremely strange scale on the vertical. The exact contours are unimportant, but once again, we will revisit this graph later.
Next we come to Performance Test 7.0. Much like Sandra’s memory-to-cache benchmark, this breaks the performance down by category, size, and unlike Sandra, by task. Once again, results are pretty simple to explain – lots of operations per second means low latency, and lots of bandwidth means data can get where it’s needed with minimal waiting. These factors contribute to a PassMark Performance Test score that’s really quite high, and a score of 9629 PCmarks in PCMark Vantage. This memory was tested in a potent PC with a high end motherboard and Intel Core i5, which certainly contributes a great deal to the score, but we can be sure that PCmark and Performance Test were not RAM-limited. How can we tell? Move on to the next benchmark suite, which represents the ‘smoking gun’ in this evaluation.
After throwing SiSoft Sandra at it, we moved on to AIDA64. In addition to providing speeds in MB/second and latency in nanoseconds, it compares the results to a crowdsourced database of example systems contributed by users. This provides some meaningful information on the Patriot RAM kit – it’s not just fast, it’s among the fastest. Two sixth-place finishes, a fifth, and a third. At stock speeds. That’s pretty fantastic, and in all cases, it was the fastest system powered by an i5 chip, suggesting that there may be some CPU bottlenecking going on. So if we aren’t the fastest right now, what’s any good hardware reviewer to do? Overclock it until something maxes (or burns) out! I’m sure it’ll be good, but just how good? Click through the next page.