Video Transcoding Test

Few consumer applications will make good use of a six-core processor, or even a four-core processor. Extra cores can give you a system that remains responsive when performing a computationally-intensive background task, but will rarely accelerate the execution of an individual program. There are several reasons for this:

• Few users have more than two cores. According to Steam's August 2010 statistics, dual-core systems comprise 54.77% of its users, quad-core systems 27.49%, and six-core systems a miniscule 0.03%. Thus developers tend to concentrate their efforts elsewhere.
• Relatively few computational problems scale well with the number of threads available. (One common task that does is rendering, which is why modern video cards have dozens or hundreds of cores).
• Writing good multithreaded code is difficult and time-consuming. Developers generally realize a better return on their effort for other code optimizations.

All that said, media transcoding (converting to a different format) is something that does scale well with the number of available cores, and the free and open-source Handbrake 0.94 video transcoder is an example of a program that makes full use of the computational resources available. For this test I used Handbrake 0.94 to transcode a standard-definition episode of Futurama to the "iPhone & iPod Touch" presets. The times are presented as mm.ss since the charting program will not let me use colons as in mm:ss...

Here we finally see a real-world program that can fully utilize multi-core CPUs. The AMD processors do themselves proud, equalling or beating the Intel processors in all but one case, and the miniscule performance advantage the i7-980X has over the stock-clocked AMD 1090T in no way justifies its four-times-higher price. One might ask why the 980X didn't do better, since its Hyper-Threading feature gives it 12 virtual cores, and the answer is simply that a virtual core is an abstract contstruct wherein the processor tries to schedule code to make the best use of available execution units; it's not the same as a real extra core. The results depend on the task and the code, and Handbrake code just doesn't benefit much from Hyper-Threading.

You can compare the bang-for-the-buck performance of these CPUs in Handbrake 0.94 by making a "dollars per reciprocal second" measurement: in other words, we're looking for the best combination of low processor price and low encoding times.

The overclocked 1075T wins this comparison, with the AMD 965 Black Edition getting second place. It's telling that none of the AMD processors exceed 200, while none of the Intel processors are under 200. In this admittedly ad-hoc measurement, the base clocked AMD 1075T is 3.5 times "better" than the 980X, which is reasonably close to the 980X's 4.1x-more-expensive price differential. The overclocked 1075T gives by far the most bang for the buck, with the 965 Black Edition and stock-clocked 1075T coming in close behind.

<< Previous Page - Next Page >>