kzeise.com

If we look at the GeekBench scores of the 6s in detail, we see clearly see where the improvements come from. Memory copy is roughly 45% faster, this indicates that Apple has done a good job improving the interface to the main memory. Taken together with the now 3MB of level 2 cache, it is no surprise that the LUA scores have made a huge jump, with the interpreter causing a lot less misses. The AES scores see the lowest improvement, since it is almost completely hardware accelerated, and so only profits from the increased frequency. It is also noteworthy to see the improvements in SHA1 and some of the filters; these indicate that Apple must have improved the pipelining of a few commands, reducing the stalls from waiting for a result to arrive. I believe these pipeline stalls are also behind Apples push to have apps stored in an intermediate form on the App Store, this would allow Apple to change the processor behavior even if it meant that code optimized for an older processor would run slower.