|Mushkin PC3-10666 DDR3 1333MHz RAM Kit HP3-10666|
|Reviews - Featured Reviews: Memory|
|Written by Olin Coles|
|Thursday, 01 November 2007|
Page 4 of 6
Testing Methodology: DDR3 RAM
Before the testing really begins, there are several settings to be adjusted in the BIOS. Many combinations of the CPU's front side bus are matched to a specific memory multiplier, giving us the desired RAM (and CPU) operating speed. When high-speed DDR3 is tested, a 1:1 memory multiplier is combined with the highest CPU front side bus speed available to obtain baseline readings; the fastest FSB is presently 1333MHz found in select Intel Conroe and Kentsfield series desktop processors. When I first begin this process and start benchmarking for a baseline, the memory voltage and timing are always left at the manufacturer's default values.
After each and every change to the BIOS, the system is restarted and all RAM modules are automatically tested with MemTest86+ for one full pass. Since the modules are not yet overclocked, one pass is usually sufficient to determine if a bad part exists. Later, once I begin to overclock the memory, these tests are a good way of quickly pre-screening stable configurations. Once the test pass is complete, I move onto the synthetic benchmark tests. Here at Benchmark Reviews we use the following test suites and applications:
After all of the tests have been performed at the default memory clock speed (or highest JEDEC standard), which establishes the baseline reading for our comparisons, I then increase the front side bus to overclock the RAM without adding any additional voltage. My theory is this: a module that overclocks without additional voltage is going to perform better than a module which requires you to potentially damage it or void the warranty. Another way of looking at this is that if a module overclocks well without added voltage, it is likely that there will be more headroom if addition voltage is applied.
I am fully aware that many sites turn up the voltage and apply dangerous amounts of power to the modules so that they can unlock that last megahertz; but if I begin down that road where will it end? Do I then start writing reviews where I apply liquid nitrogen cooling to the RAM? Do I volt-mod the motherboard and apply so much power I ruin expensive parts and equipment? In the end, I decided that overclocking will be done at the stock voltage for benchmarks, because I believe most enthusiasts are going to aim for the same goal. If on some rare occasion I determine that adding voltage would potentially return large performance gains, then I will add up to 0.2V to the default volt rating.
In each synthetic benchmark test Windows XP was booted fresh and the application was opened. Task manager was then utilized to end the explorer.exe (Windows) process so that only the most essential services were running and a trim 16 processes total were consuming system memory. This procedure allowed the RAM to be as isolated as possible for our testing. Once the system was ready, I perform a single test-run of the benchmark, followed by three recorded tests. At the end of the series, I average the three results for the final score.
Sometimes it is very difficult to get broad results between the memory modules tested, so we use several different speeds of memory from participating manufacturers. Benchmark Reviews is very serious about performance, which is why we test the products against as many benchmarks as reasonably possible. Not every test suite is effective in these reviews, and some often calculate processor and other system components into the score. This is what makes a RAM review difficult: unless all of the modules compared have the exact same rating and the CPU operates at the exact same speed for each and every test, the comparison is always going to be subjective.