OK, I found this interesting, and very much in-line with Apple's decisions; the L3 cache on my 17" 1 GHz PowerBook doesn't affect speed particularly much, at least in certain benchmarks. If you have the developer tools installed (including the CHUD tools), you can test these things for yourself, YMMV. Anyway, I see very little difference between L2 and L3 cache enabled and just L2 cache enabled in CacheBasher or SkidmarksGT (using the Hardware system preference installed by the CHUD tools to selectively enable or disable the caches). Using L3 alone (with no L2) does make a huge difference over using no L2 and no L3, however. Anyway, the bottom line: many people are going to have improved performance from the increase in L2 cache and the loss of all L3. Only a few applications will work better with the third level. FWIW, I got up to about 12 GBps of bandwidth in CacheBasher, by doing a load of vector data, with a stride of 16. With no cache, this drops to about 700 MBps. With no cache using integer data, it's only 170 MBps. I guess this is probably more interesting to me than to the average person; in one of my classes right now, we are discussing how to build a memory system, including how to make a cache. Our final project for the class (we started about 2 weeks ago, we have another month and a half to finish it) is to build a 16-bit microprocessor. For me, that's LOT of fun. For the vast majority of my friends, that would probably be worse than chinese water torture. ;-) Happy Halloween, everybody!