SI
SI
discoversearch

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 ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  
To: Dan3 who wrote (83060)12/15/1999 1:57:00 PM
From: Tenchusatsu  Read Replies (3) of 1570954
 
Dan, <I think I learned something about pin costs from Carl's posts - he was pretty persuasive that pin costs were dropping very fast, so that this benefit of Rambus didn't overcome its drawbacks.>

Haven't you read my response to him? He used IBM's 5000 I/O pad package for the Power4 multi-chip module as an example of cheap pins. That's like using the Ford Excursion (the Suburban Assault Vehicle) as an example of cheap gas. Pins certainly do not come for free.

<I'm not going to rehash the "efficiency" claims again, but I'd like to know what network, disk i/o, or video technology you expect to absorb anything other than a burst from 16 Rambus channels.>

Most of that bandwidth will be consumed by the Alpha processor core itself. The Alpha in particular has always been architected with very wide memory channels. Perhaps the designers like Pete Brannon feel that anything less would be uncivilized.

As for I/O, well, PCI-X is up and coming, as well as Infiniband (the new name of the merged NGIO and FIO standards). Both will have rather huge memory bandwidth requirements, depending on the devices. Of course, that will add up to only a fraction of the bandwidth provided by 16 Rambus channels, but like I said before, most of the bandwidth will be consumed by the processors anyway.

But don't take my word for it. Take the word of the EV7 designers who decided that four RDRAM channels per processor was necessary.

<Is your goal the design of systems that swap blocks of memory back and forth all day?>

If you have to ask, then you know NOTHING about servers.

Tenchusatsu
Report TOU ViolationShare This Post
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext