Formerly About Advanced Micro Devices Message Board

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Politics : Formerly About Advanced Micro Devices

Public Reply Prvt Reply Mark as Last Read File

Previous 10 Next 10 Previous Next

To: Daniel Schuh who wrote (93590)	2/16/2000 5:15:00 PM
From: Jeff Fox	Read Replies (2) of 1572365

Dan, re:"The thing that still puzzles me most about the (limited) Willy info is the Rambus angle" Thanks for your well written post. You frame these memory systems problems well. Let me have a shot at illuminating Intel's motives with Rambus. Believe it or not Rambus was chosen as the best, perhaps the only technical solution to a problem of how to get high memory bandwidth into desktop systems at the best cost. It is obvious that there are rocks in the way of achieving these goals. Rambus requires major cost reduction to meet the requirements. Pin count cost is key to understanding the logic... Simply put, to go faster with conventional architecture requires more and more pins. Sure DDR helps by doubling throughput with current pincount, but that's about the end of the road for this architecture. The big problem lies with the need to have memory access signals provided from outside the memory chips. They simply cannot scale with higher speeds and memory chip capacities. Rambus reduces pincount for very fast memory systems contained within only a few chips. The memory array signals are contained inside the memory, thus they can scale up in both speed and width without external constraints. Intel sees this as a key for desktop systems as they evolve through this decade. To summarize - its all about cost and packaging constraints. Pin count is key to understanding the choices... Now server systems have huge memories, thus lots of individual memory chips. The cost of memory dominate over the cost of glue chips. Here it is clearly most cost effective to use conventional architecture - very wide memory busses with centralized memory controllers. With high RDRAM cost premiums so high things aren't clear. Either Rambus memory gets cheaper or Intel will have to bail elsewhere, but I don't see any other realistic solution as of yet... Jeff

Report TOU Violation

Share This Post

Public Reply Prvt Reply Mark as Last Read File

Previous 10 Next 10 Previous Next