Articles: Mainboards

Bookmark and Share

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 ]

Performance Comparison

As usual, we are going to compare the mainboards speeds in two different modes: in nominal mode and during CPU and memory overclocking. The first mode is interesting because it shows how well the mainboards work with their default settings. It is a known fact that most users do not fine-tune their systems, they simply choose the optimal BIOS settings and do nothing else. That is why we run a round of tests almost without interfering in any way with the default mainboard settings. But today this approach was only applicable for Intel DX58SO2, which we have included into the article for comparison purposes. The CPU works the way it is supposed to only on this particular mainboard, while on Gigabyte GA-X58A-OC all power-saving technologies are disabled by default, and in the BIOS of Gigabyte G1.Sniper the switching to deep power-saving modes is also off. As a result, Intel Turbo Boost technology will only partially work on these mainboards: even when only one processor core is utilized, its clock frequency multiplier will only increase by 1x instead of 2x as it normally would. It means that during tests in single-threaded applications both Gigabyte mainboards will be slower and in idle mode both of them will consume more power than necessary. This comparison is obviously not going to be in Gigabyte’s favor, so we made an exception and manually enabled all power-saving technologies before commencing the tests.

However, it would be incorrect to claim that Intel DX58SO2 mainboard is a perfect ideal product. While reviewing this board I overlooked the fact that it didn’t use information from the memory modules SPD and used its own settings instead. The Kingston KHX12800D3LLK3/6GX memory modules that we use work at 1600 MHz frequency with 8-8-8-24-1T timings only when their voltage is increased to 1.65 V. to ensure that the computer starts successfully at the DDR3 nominal voltage of 1.5 V, the modules are introduced to the system as 1333 MHz memory with 9-9-9-24-1T timings. Both Gigabyte mainboards used these specific parameters, while on Intel mainboard the memory by default worked at 1066 MHz with 8-8-8-20-1T timings. One of the reasons why I didn’t uncover this issue when I reviewed Intel DX58SO2 was because for comparison purposes I chose Asus P6X58D-E. We didn’t detect any performance differences, but only because Asus mainboards also do not use the SPD settings for the memory and configure all parameters on their own. It could be that Asus mainboard set the same frequency and timings. So, we decided not to correct the memory frequency and timings on Intel DX58SO2 this time. On the contrary, we were curious to see how it would affect the performance. Note that the results of our testing participants are arranged in a descending order on the diagrams.

We used Cinebench 11.5. All tests were run five times and the average result of the five runs was taken for the performance charts.

A small video in x264 HD Benchmark 3.0 is encoded in two passes and then the entire process is repeated four times. The average results of the second pass are displayed on the following diagram:

We measured the performance in Adobe Photoshop using our own benchmark made from Retouch Artists Photoshop Speed Test that has been creatively modified. It includes typical editing of four 10-megapixel images from a digital photo camera.

In the archiving test a 1 GB file is compressed using LZMA2 algorithms, while other compression settings remain at defaults.

Like in the data compression test, the faster 16 million of Pi digits are calculated, the better. This is the only benchmark where the number of processor cores doesn’t really matter, because it creates single-threaded load.

There are good and bad things about complex performance tests. However, Futuremark benchmarking software has become extremely popular and is used for comparisons a lot. The diagram below shows the average results after three test-runs in 3DMark11 Performance mode with default settings:

Since we do not overclock graphics in our mainboard reviews, the next diagram shows only CPU tests from the 3DMark11 – Physics Score.

We use FC2 Benchmark Tool to go over Ranch Small map ten times in 1920x1080 resolution with high image quality settings in DirectX 10.

Resident Evil 5 game also has a built-in performance test. Its peculiarity is that it can really take advantage of multi-core processor architecture. The tests were run in DirectX 10 in 1920x1080 resolution with high image quality settings. The average of five test runs was taken for further analysis:

The performance difference is not very dramatic, but Intel mainboard is often the last of the three, because its memory worked at a slightly lower frequency and the more aggressive timings couldn’t really make up for that. We see a slight lag of about 6-7% in 7-zip test, which is extremely sensitive to the memory sub-system parameters. However, do not forget that we manually fully enabled Intel Turbo Boost on both Gigabyte mainboards. This allowed them to avoid disaster in single-threaded SuperPi and Adobe Photoshop tests.

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 ]


Comments currently: 2
Discussion started: 08/04/11 08:06:07 PM
Latest comment: 03/08/16 03:55:13 AM

View comments

Add your Comment