|OCZ Blade ST 2000MHz Low Voltage DDR3 Memory|
|Reviews - Featured Reviews: Memory|
|Written by Steven Iglesias-Hearst|
|Friday, 12 November 2010|
Page 6 of 8
Benchmark Test Results
Benchmark Reviews runs each tests at least 3 times to avoid special situations where scores are over or under exposed. Passmark Performance Memory Test benchmarks the memory and reports a test score. I´ve chose Memory Mark along with uncached read and write speed since they´re the most important benchmarks for memory testing on this suite.
The memory scales as we would expect in the synthetic Passmark Performance tests, that is until we get to our overclocked settings. Maybe we have an anomaly here or maybe the CPU factor has an influence on the results, either way we will see as we look at other tests. Performance gains at the higher speeds are minimal thanks to the tighter timings at the lower speeds.
AIDA 64 Edition offers three simple memory bandwidth tests that focus on the basics; Read, Write, and Copy. In order to avoid concurrent threads competing over system memory bandwidth, the Memory benchmarks utilize only one processor core and one thread.
Once again scaling is as expected in a synthetic benchmark, AIDA 64 Extreme rewards higher scores based more on speed than timings in these tests. Our overclocked settings only take a hit on copy speeds this time round. A meagre 1.65% performance increase in read speeds and 1.90% increase in write speeds are achieved with our overclocked settings.
Sandra is based on STREAM, a popular memory bandwidth benchmark that has been used on personal computers to super computers. It measures sustained memory bandwidth not burst or peak. Therefore, the results may be lower than those of other benchmarks. STREAM 2.0 uses static data (about 12M) - Sandra uses dynamic data (around 40-60% of physical system RAM). This means that on computers with fast memory Sandra may yield lower results than STREAM. It's not feasible to make Sandra use static RAM - since Sandra is much more than a benchmark, thus it would needlessly use memory.
A major difference is that Sandra's algorithm is multi-threaded on SMP/SMT systems. This works by splitting the arrays and letting each thread work on its own bit. Sandra creates a thread for each CPU in the system and assigns each thread to an individual CPU. Another difference is the aggressive use of scheduling/overlapping of instructions in order to maximize memory throughput even on "slower" processors. The loops should always be memory bound rather than CPU bound on all modern processors.
In the Sandra test suite we get the same story with a happier ending, scaling is in line with speed rather than timings. We have a 4.16% performance increase in the Int Buffered test and we have a 3.88% increase in the Float buffered test with our overclocked speeds.
In the next section we will focus on real world applications to see if the results gained will match those of the synthetic benchmarks.