AMD Athlon II X4 620 Processor
Very good optimizations of Java applications for multithreading allow Phenom X4 9850 significantly outperform the triple-core processor and almost catch up with the Q8200. However, the 620 shoots forward even with DDR2 memory, and DDR3 practically completely makes up for the lack of L3 Cache. Excellent results!
Yep, if cache memory had been sold separately from processors, tests with archivers would have been a ubiquitous attribute of its advertisements. And the budget way to get maximum cache size per core is to buy 2- and 3-core Phenom II processors, which are manufactured by disabling cores in a quad-core die, still having the sterling L3 cache. We have this processor in the list of our contenders, and it takes the first place. What concerns the 620, according to this theory, it might have performed worse, because its cache size is definitely not to archivers' liking.
The fourth core yields maximum dividends in this test, so the 620 improves its results. Cache size does not play an important role in streaming tasks.
Results in this group of tests are traditionally illustrated with this table. It's indispensable in this case.
||Phenom X4 9850
||Athlon II X4 620 (DDR2)
||Athlon II X4 620 (DDR3)
||Phenom II X3 710
||Phenom II X4 810 (DDR2)
||Phenom II X4 810 (DDR3)
||Core 2 Quad Q8200
The total score is irrationally affected by the XviD test, which apparently spoils the performance picture of Phenom II and probably Phenom (regardless of the core number). We found it out in our first tests of Athlon II, when dual-core models demonstrated significantly better results despite the lack of L3 cache in comparison with equally-clocked representatives of the top family. Formally, we didn't clear the suspicion that the problem was in different L2 cache sizes per core. But the quad-core Athlon II also has 512 KB per core just like Phenom II, and we see the same picture again. In this case it's an illustrative situation, because, for example, Phenom II X4 810 differs from the 620 only in the additional cache level. So, an optimization error in the program is the only rational explanation.
In other respects, everything is more predictable. Encoding should not depend much on cache size, that is the 810 shouldn't be much faster than the 620 (but it definitely must not be slower). But as we can see, L3 cache still provides some advantages in most tests. Only x264 seems to be indifferent to cache size, so there's a formal defeat. But this situation does not compare with the defeat in XviD. It can be explained with different tweaks in memory controllers of these processors -- we've already mentioned it in several tests.
Write a comment below. No registration needed!