Tests

The benchmarking procedure (the list of software and test conditions) is described in this article. To make the diagrams easier to read, results are represented in percents (100% stands for the result of Intel Core 2 Quad Q6600 in each test). Detailed results in absolute values are published in this Microsoft Excel spreadsheet, which includes all previously tested processors.

3D Modeling and Rendering

We can call it a draw in both groups of quad-core processors. What concerns the Q8200 lagging behind the Q6600 (especially Phenom II X4 810), it will happen in most tests. Victory of the Phenom II X3 720 in the segment "two cores versus three cores" looks quite convincing, although not all programs in this test can actually use more than two cores.

CAD/CAM

Dual-core processors run the show in this group of tests, and even Athlon 64 X2 6400+ is faster than the new and old budget quad-cores. So the E8200 shoots forward, and the triple-core Phenom II takes the second place. This group of tests seems to favor as few cores as possible. But it's actually the effect of higher frequency and cache size per core. These tests can use more than two (or even one core) only occasionally.

It's hard to tell whether this tendency remains after these programs are updated and/or new programs of this class added. It should be done to keep our test procedure up to date, but as such software is notable to its inert treatment of innovations in processors, the situation is not likely to change cardinally.

Compiling

Multiple cores do not yield much dividend to compile tests either -- Phenom II 720 and 920 demonstrate identical results, and they differ only in the number of cores. However, owing to architectural differences, more effective data exchange with memory through the integrated controller, Phenom II processors win in all three groups of tests. Results of the inexpensive triple-core processor are the most impressive from the practical point of view.

Professional photo processing

As usual, Adobe Photoshop comes to the rescue of Intel processors. This program is good at distributing load between cores. So quad-core processors from Intel become the leaders. It's actually the only test where the Q8200 outperforms the Q6600. And the explanation is not far to seek -- the effect from increasing the bus frequency actually must appear where all four cores are actively used (traffic between core pairs, implemented in Core 2 Quad via the external bus and chipset, grows, and it becomes a bottleneck).

However, when we compare processors with different number of cores, Photoshop is also adamant: three is better than two.

Scientific applications

We've stopped expecting rational results from MATLAB long ago. It usually returns only irrational verdicts to processors. This time MATLAB drops the Q8200, its total result is even 20% as low as in the Q6600. Perhaps, AMD fans will be pleased that MATLAB has finally turned against its perpetual favorites, Core 2 processors. So Phenom II X4 810 even outperforms its direct competitor. However, we decided that "the guard is too tired already". So it's time to try and calculate the total score in this group of tests using the other two programs, if not remove MATLAB from our test procedure completely. By the way, they do not show any subterfuges and even don't contradict to each other (a processor that earned a higher score in Maple usually gets a higher score in Mathematica as well). However, as these two programs give us three marks that are used to calculate the total score, while MATLAB gives as many as six, they couldn't possibly outargue it.

Here is the score without MATLAB results. The Q8200 is outperformed by the Q6600 by 5%, which does not contradict to the overall situation in other tests. We can see very well that AMD K10 core in Phenom II gets a "proper" chance to reveal the power of its expanded floating-point unit, which must peak in mathematical programs. By the way, K8 was also good at mathematics. What concerns Opterons (based on a similar microarchitecture), everybody knows that they are a very popular choice for supercomputers. So, we've got expectable logical results.

Fortunately, a result in one group of tests does not affect the total score very much, so even serious deviations do not change the final CPU rating owing to the big number of applications in our test procedure. You can see that it's true, if you compare total scores obtained with our method including MATLAB and without it. So all general conclusions made in previous articles hold true. And what concerns the group of scientific software, we've always recommended to look through detailed results and do not trust MATLAB completely.

Write a comment below. No registration needed!