Bij Ars Technica is vandaag het derde deel verschenen van hun RAM Guide. In dit deel wordt uitgebreid uitgelegd hoe DDR en Rambus geheugen werkt. Het artikel is een aanrader voor iedereen die alles wil weten over de Rambus performance issues en waarom het überhaupt ontworpen is. Het is wel erg technisch, maar als je bij deel 1 begint en als je hersenen een beetje meewerken moet het begrijpbaar zijn :
The previous editions of the Ars Technica RAM Guide deal with both obsolete and current SRAM and DRAM technologies: EDO DRAM, FPM DRAM, EDO DRAM, and SDRAM. This edition will cover the two next-generation DRAM technologies that are competing for dominance in the PC space. We'll open up this article by talking about DDR DRAM, and how it relates to the current generation of SDRAM. We'll also talk about its strategy for increasing bandwidth, and what makes it a next-generation solution. Then, we'll lay the foundation for a lengthy investigation into RAMBUS technology by considering the closely interrelated issues of transistor counts, pin counts, and granularity. After this discussion, we'll take a look at RAMBUS, starting with the core of the individual RAMBUS chip and working our way outwards until we've covered an entire RAMBUS system implementation.
[...] RAMBUS DRAM is a next-generation DRAM that, instead of taking the features of standard SDRAM and evolving one or two of them, takes DRAM technology and rethinks from the ground up how it should be integrated into a system. The DRAM arrays of memory cells that are at the core of all DRAM types are still there in RDRAM, but the way that a RDRAM-based system accesses those cells is very different. These changes in the way that DRAM cells are organized and accessed are the product of a series of design decisions (many of which were made almost a decade ago) that are aimed not only at providing the high bandwidth that next-generation media applications require but also at solving the granularity and pin count problems that RAMBUS, Inc. expected to accompany rising transistor counts. Assuming that pricing considerations aren't an issue (and with RDRAM they often are), the decision on whether or not to use RDRAM in a system is and should be made based on the actual bandwidth requirements of the system in question and whether or not the granularity and pin count problems that RDRAM tries to solve are actually relevant for that system. And when I say "system," I mean the hardware as well as OS and application software.