The Empirical Data

SuperPi and wPrime

SuperPi 1.5 is a mainstay amongst benchmarkers to compare x86 performance. Two tests are common 1M and 32M which run multiple loops calculating Pi to 1 million and 32 million digits it uses Intel extensions in its code base so comparisons to AMD chips are not as valid. 32M is affected greatly by memory bandwidth and latency.

wPrime is another benchmark that is commonly used for x86 performance and also works on calculating Pi. wPrime is multithreaded and the two tests sizes available are 32M and 1024M. Version 1.55 is used as that is the accepted version at hwbot. Memory performance is still tied closely to final results, but bandwidth is less important than latency.

Results are great all around but knocking timings down to 8-8-8 and clocking to 1900 MHz shows a slightly larger performance gain due to BCLK adjustments that also raised CPU speed by 2%. wPrime 1024M results actually took longer at 2133 MHz due to increased latency by moving to a 2T command rate.

PCMark Vantage, PCMark 7

These are complete system benchmarks that are commonly used around the world. These software are the latest in a decades worth of Futuremark products and evaluate total system performance as it pertains to the average desktop user’s experience. From 3D calculations to storage performance and video transcoding. Futuremark also includes whitepapers on the test that go into greater detail about each benchmark. The Entire suite was run for both benchmarks with subcategory totals being listed in the charts.

PCMark Vantage actually sees stock speed and timings performing better overall very slightly despite improved performance in certain subtests. PCMark 7 results are more inline with what is expected. 1900 8-8-8 provides best overall performance due in part to the 2% system clock advantage however in most tests performance increased by much more than the 2% system clock relates to specifically in computation/productivity tests.

Performance Test 7

Passmark’s software is also used to evaluate total system performance but offers specific hardware related pages to refer to for testing specific components. The entire suite is run and the total Passmark rating is included, much like WEI ratings in windows the total Passmark score is limited by the weakest components in your PC. The individual tabs are included to breakdown hardware specific tests and the memory tab was selected for representation in the chart below.

Although the total passmark rating is highest with the fastest RAM multiplier used, memory performance shows the exact opposite. 1900 8-8-8 wins overall due to the enhanced bandwidth and reduced latency of the memory sticks.

AIDA64 Memory Test

AIDA64 2.2 engineer edition from Finalwire is used in our motherboard reviews and is a great indicator of both CPU and memory stick performance. As this is a memory review, we have eliminated the CPU/FPU results from the chart and focus solely on memory performance.

As expected the higher the speed of the RAM the greater the performance with approximately 5% better performance in every test using the 2133 multiplier instead of the 1866. 1900 8-8-8 performed right in the middle although latency was extremely close to the 2133 results being off by only .1 ns.

3dMark Results

3DMark Vantage and 3DMark 11 are used to evaluate synthetic gaming system performance. The focus on game tests and use a number of predefined presets to make comparisons easy for users. Both tests are run using Performance and Extreme Presets.

Vantage Performance results are skewed in favor of the fastest memory setting due to DX10 API and low resolution used in the performance test. Feeding the GPU is the overriding goal. In all other tests performance is very even with only a slight edge going to the 1900 8-8-8 setting.

Game tests

For gaming tests we looked at The Dukes video card review and followed his example. Lost Planet 2, Hawx 2, final Fantasy 14 and Alien vs Predator were selected. All used 1920x1080P resolution 8x AF 4x AA with all texture/details setting raised to maximum allowed values. This represents the gaming audience that already has a high def display and a 580 GTX video card.

Results here are confusing with one game showing the lowest performance with the the fastest memory setting, another game showing the best performance with the fastest memory. What is obvious though and what was entirely expected is that gaming performance is generally not affected by memory performance once the threshold has been passed. (The threshold is generally what the CPU manufacturer recommends for memory speed).

Pages: 1 2 3 4 5