|Mushkin PC3-10666 DDR3 1333MHz RAM Kit HP3-10666|
|Reviews - Featured Reviews: Memory|
|Written by Olin Coles|
|Thursday, 01 November 2007|
Page 5 of 6
Test Results: Mushkin 1333MHz DDR3
Testing RAM is a subject which requires a bit of technical knowledge. I have recently found many other sites using nothing by video games to benchmark the RAM product they are reviewing, and this whole practice made no sense at all t me. Video games are GPU and CPU dependant, and RAM has a very small impact on framerate performance as I will prove to you in my own tests. So keeping all of this in mind, I use the tools that belong in a system memory review; I use system memory tests for my benchmarks. Additionally, I don't spend three (usually tiny) pages discussing how I overclocked or how I made it to tighter timings. Every memory module comes with its own limit, so one size doesn't fit all.
Without adding additional voltage, and maintaining the CL6-7-6-18 timings, this kit would only overclock to 1372MHz. This kind of improvement almost doesn't qualify. However, with a little leniency this Mushkin DDR3 1333MHz kit later overclocked to 1500MHz with default voltage and the sacrifice of latency timing values of 6-7-6-18 loosened to 9-9-9-24 in order to remain completely stable. Obviously a 167MHz increase in speed without adding to the voltage is worth mentioning, but it's also a good idea to keep in mind the default memory voltage for this kit is only +0.3V over spec at 1.8V; which means that there could be more room for safe voltage increases and higher overclocks or even tighter memory timings if desired.
Our first results were recorded from Lavalys EVEREST using the Cache and Memory Benchmark tool. The results shown below represent the average measurement obtained from the Mushkin PC3-10666 CL6-7-6-18 HP3-10666 DDR3 1333MHz 1GBx2 1.8V RAM kit at the 1:1 RAM to CPU multiplier of 1333MHz and then overclocked to 1500MHz with open timings. The results for the average read, write, and copy bandwidth from EVEREST are displayed below.
Everest is among my most trusted benchmark programs, and the Cache and Benchmark tool is one of the more reliable in terms of consistent results. The chart above shows that the extra 167MHz over the 1333MHz baseline had nearly no impact on the Read test, but it certainly amounted to a significant increase in added bandwidth with more than 27% improvement in the Write test.
With both CPU-Z and EVEREST reporting memory clocks at CL6-7-6-18, I will believe that the ASUS P5K3 BIOS settings were configured correctly. While Mushkin rates this HP3-10666 kit for 6-7-6-18, I have discovered that the ASUS P5K3 motherboard sets the clocks to 9-9-9-24 when using the AUTO setting.
Next up was the PassMark Performance Test benchmark which runs several different system memory tests in a row. Although some of the tests are specific to the performance of the RAM, others take the CPU clock speed and front side bus into account when developing a score. Most important are the memory read and write tests, and the score based tests are bias towards CPU speed and other hardware factors.
Passmark's Performance Test offers the most consistent memory test results of the entire group, with each of the test runs resulting in a score nearly identical to the previous test run. Ideally, all of these programs should be this consistent, but until they are I would consider Performance Test to be the best tool available for testing system memory bandwidth.
SiSoftware Sandra Lite XIIc offered marginal results regards to increased performance; according to the chart above. Although the bandwidth tests are of a different nature, the improvement of 1500MHz over 1333MHz still resulted in only 7% in both the integer bandwidth and float buffered tests.
The last of our memory testing applications to run is RightMark Memory Analyzer. This program may not offer the same level of consistent test results that Performance Test does, but offers a more technical approach to testing the system memory. Plus, this is a shop favorite because it's free.
RightMark Memory Analyzer offer nearly the same tests that EVEREST and Performance Test have, and could be considered a bit redundant, but then again this is a technology article and us geeks like our redundancy. While it's true that Memory Analyzer hasn't been updated in over a year and seems out of development, it's still not a bad tool since it gives the second most consistent results every single run (unlike Sandra). It also offers an average "real" RAM read and write bandwidth result which most enthusiasts don't appreciate because it combines the results of dozens of tests. This is the layman's alternative to the Sciencemark v2.0 test suite, which is also gone but not forgotten.
Finally, I tested with the memory multiplier set at 1:1 for 1333MHz against the overclocked 1500MHz setting in the game World in Conflict. Realizing that games can be either CPU or GPU bound, this made it difficult for me to compare all of the memory sets since the clock speed of the processor would change as I adjust the front side bus for the desired RAM speed. In the end, a 167MHz system memory improvement gave the framerate in World in Conflict a mere 1 FPS more in both the average and maximum frame rates while minimum framerate was identical. This result is evidence of how insignificant the system memory speed is in relation to video game performance, irregardless of timings.
But don't misunderstand me, because system memory could have a much larger impact on game performance if you use it to overclock the processor. Obviously, if you are using the Mushkin PC3-10666 CL6-7-6-18 HP3-10666 DDR3 1333MHz RAM kit you aren't going to keep the CPU at the stock speed anyway. Additionally, it is very likely that you will use a video card that more than makes up for the very small gains of overclocked system memory. This means that faster RAM allows for a faster CPU, and in turn produces a faster framerate. But in the world of system memory benchmarks, comparing the different sets of RAM in a game is pointless.