AMD to Intel????

[Alpha_One]

Currently, it puzzles me why AMD with their revolutionary* on-die memory controller is reducing cache. In contrast, the E6600 and above are very fast because of these 4MB. It is predicted they will double to 8MB or 16MB after the 4x4 (Kentsfield Quadcore).




*Actually intel had an on-die prototype pre-P4 days but decided to bin it. Not sure why. AMD is first to market it and it indeed proved to be revolutionary.

AMD cutting back cache: Supply-and-demand. AMD has a HUGE disadvantage over Intel when it comes to production capacity, not to mention that they're slightly behind in smaller manufacturing processes. Dual cores = less production = more expense. AMD is simply FORCED to cut back cache to save space on their chips. Otherwise, either their chips will be too expensive, or they'll just run out of chips to sell.

Intel not implementing the on-die controller. Two main reasons, I would guess:

a) Technical: Fabrication technology wasn't quite advanced back then. The first Intel P3 and AMD Athlon Classic had the L2 cache off-die for a good reason: the on-die L2 caches were BIG. BIG means EXPENSIVE. Any on-die memory controller would significantly add to the size of the (back then) very large dies, making them very expensive. Another thing is that a memory controller produced to partner SDRAM, would only partner SDRAM. Just like 939 and AM2 processors, which is, for all intents and purposes, the same thing with different memory controllers.

b) Politics. This were the days when Intel was aggressively pushing RDRAM. If RDRAM flopped (as it did), it would be EXTREMELY expensive to switch back due to reasons outlined in A.

P.S. On-die memory controllers are hardly revolutionary, in the same sense that variable valve timing isn't revolutionary. I mean, by World War II lots of gizmos and witchcraft on a modern internal combustion engine were already thought up - it just doesn't make sense to make them. On-die memory controllers are not a new idea, some small companies during the Pentium One days (was Cyrix one of them?) were attempting to implement what was a primitive example of what the A64 has today and then some. i.e. EVERYTHING (cache, memory controller, drive controller, PCI interface, the lot!) on one chip. Not practical (back then) though.

P.P.S. Benchmarks or no benchmarks, the E6600 clobbers the FX-62 on the most important measure of all: price. The performance can go at most a few percent either way depending on the application, but the E6600, at 1/5 the price of the FX-62, would comparatively be a steal. I'd go for the E4300 though. Or just stick with my loud, hot P4 530 (it still works, anyway), and see if Core or K8L or whatever new thing there is in a few months would be attractive.