The new benchmarking procedure (the list of software and test conditions) is described in this article. But here we use the old principle of grouping applications for relative scores (by application similarity). Later on we shall use a new principle (in compliance with task classes). To make the diagrams easier to read, results are represented in percents (100% stands for the result of Intel Core 2 Quad Q9300 in each test). The detailed results in absolute values are available in this Microsoft Excel spreadsheet.
3D modeling and rendering
The group of 3D Modeling applications unexpectedly get no performance gains from DDR3, they even slow down a little. It's very strange, because triple- and quad-core processors from Series 700, 800, and 900 always demonstrated performance gains from using DDR3 in this group of tests (about 3-5%). Perhaps, memory bandwidth in 3D modeling packages does not matter for dual-core processors anymore, while DDR3 memory latencies may be higher in some modes and deteriorate results. But leaping ahead, it's an only example of its kind. DDR3 is at least no worse in the other groups, and sometimes it does noticeable good.
It looks like we cannot expect scalability and high relative scores in CAD/CAM programs (conservative applications that do not favor multithreading). It concerns not only quad-core processors, but also dual-core models.
Phenom II is formally outperformed by its competitor here. The same concerns Athlon II with DDR2. However, Athlon II effectively makes up for the lack of L3 cache and small total cache size by upgrading to DDR3 memory.
AMD processors outperform their competitors in the compile test by an unattainable value. Athlon II almost catches up with the E7400. And we remember that compiling loves cache! But in fact, it's not about a large cache, it's about a wider sense: dependence of this task on capacity of a processor to interact efficiently with the memory system on the whole.
Image editors demonstrate the opposite picture, what they value is integer arithmetic performance, so Intel processors take the lead. There is only a small performance gain from DDR3 memory, so higher memory bandwidth is not relevant here. And what concerns latencies, the progress is not that big, as we use DDR2 with minimal timings, while DDR3 has to use higher values so far. As a result, some operations in Photoshop are performed faster, the others are slower, so the average advantage of DDR3 is minimal.
AMD processors are victorious here. The most striking effect from DDR3 is demonstrated by the cheaper model, which was only a bit faster than the E5300 with DDR2, but caught up with the E7400 with the new memory type. We haven't seen such performance gains from upgrading to the new memory type yet! On one hand, the lack of L3 cache and mediocre cache size should reveal interaction peculiarities of the memory system. But on the other hand, we didn't expect that the integrated memory controller (even if it's the third generation of such controllers for AMD) would be so effective. In fact, taking into account the difference in frequencies with Phenom II X2, we can say that DDR3 fully makes up for the lack of L3 cache.
From the ideological point of view, improving the controller and overclocking memory looks like a better approach than the traditional method of smoothing the difference in bandwidth of inner CPU blocks by adding cache memory of various levels to the processor. Especially for the budget segment, by the way, where a large cache is impossible for economic reasons. No matter how large a cache is, there will always be situations when even short-term data won't fit there. Besides, as DDR3 grows cheaper, users of Athlon II processors will often choose high-speed 1333 or even 1600 memory instead of the officially recommended 1066. Fortunately, there are no problems with setting such frequencies with a multiplier or a base frequency. Besides, this processor will probably be affected by the increase in CPU NB frequency as well.
Write a comment below. No registration needed!