Silicon Investor (SI) -- The First Internet Community

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.

Technology Stocks : Rambus (RMBS) - Eagle or Penguin -- Ignore unavailable to you. Want to Upgrade?

To: Ian Anderson who wrote (48770)	8/2/2000 7:01:07 PM
From: jim kelley	Respond to of 93625

Ian Thanks for your excellent and very clear post on this subject. I know there are a lot of good engineers like yourself who have remained quiet in the midst of the negative propaganda blitz. But sometimes it is necessary to speak up and set the issues straight. Regards, Jim Kelley

To: Ian Anderson who wrote (48770)	8/2/2000 8:14:39 PM
From: Bilow	Read Replies (1) \| Respond to of 93625

Hi Ian Anderson; The DDR guys have are taking into account the fact that the boards have to run in so many different installations with so many different DIMM populations &c. I think that if you listen to the industry presentations you will be impressed. Not the high level view graph stuff, but instead the audio presentations of the DDR guys. I'll post some links later, when I get to my other computer. You will have about 4 hours of lecture to listen to, but there really isn't any point for me to talk to you about this without your having listened to the other side. You've made some comments on the difficulty getting SDRAM to operate reliably in DIMMs. (And believe me, the comment about the FCC compliance guys hits home with me.) It is possible that you are not taking into account the signalling differences between SDRAM, and DDR: (1) DDR went to SSTL-2 signalling, not CMOS. This means that voltage swings are drastically reduced (about half amplitude). Since EMI is typically proportional to something like the square of the voltage swing (with the same rise and fall times), this means that EMI is probably going to be easier with PC2100 than with PC100. The termination has been standardized and improved quite a bit. It still isn't as sweet as a decent ECL SSI board (I used to be in supercomputers), but the signal levels are beautiful and the EMI has been reduced. In short, the engineers learned from your PC100 EMI experience, and they took this into account. Also SSTL-2 uses relative threshold voltage levels instead of the ground reference. This helps make signal integrity less susceptible to ground bounce and reference problems, allowing lower voltage swings. This is not your daddy's TTL. (2) DDR went to source synchronous timing, while SDRAM has a fully synchronous system clock. This means that the timing margins are huge, relative to SDRAM. (That is, much less time is getting eaten up by clock skew.) Consequently, the designers were able to leave the edge rates alone. DDR is sweet stuff. As far as power supply problems, well it uses considerably less power than SDRAM, at the same bandwidth, so that shouldn't be an issue. I think that DDR266 is going to be fine for the PC industry for the next three years or so, and after that, I think other trends will take over and alleviate the necessity of running large bandwidth interfaces over long PCB traces. One of those trends is embedded, another is SOC &c., the third is the explosion in cheap high pincount packages, and the fourth is a slow but sure reduction in memory chip count per system (and thus encouraging point to point designs). So DDR doesn't have to have legs that far into the future. If it were really the case that DDR wasn't scalable to large systems, as you are suggesting, then you must find it very odd that essentially all the server chipset design companies (including Intel) are making DDR chipsets for servers. The basic fact is that DDR has sewed up the low end, point to point market, as well as the 2000 design wins in the high end, buffered system market. The middle market (including desktop workstations) now has dozens of chipsets in development, it will convert next year. Looking forward to the granularity issues that will arise from 1Gbit DRAM chips, the DDR guys are designing memory interfaces as wide as x64. So DDR definitely has enough legs to satisfy the industry for quite some time. -- Carl P.S. Glad to welcome another VRAM survivor to the thread... That stuff had the nastiest collection of timing requirements of any memory I've dealt with. No one who has not dealt with it first hand really understands how bad it was. You can still get data sheets on it from OKI, who still, believe it or not, makes the stuff. I'd give a link but you have to sign up for their data sheets.

To: Ian Anderson who wrote (48770)	8/3/2000 12:49:22 AM
From: Ali Chen	Read Replies (2) \| Respond to of 93625

<Rambus solves the problems of high speed PC main memory because it is a complete specification of the memory subsystem. The tight tolerances imposed on RIMM manufacturers will guarantee inter-operability.> Solves? Complete specifications? Guarantee inter-operability? Are you honestly saying this, Mr. Anderson? Didn't you forget that the original Rambus "solution" was 3 RIMM slots, but the third one was later abandoned by Intel? Is it because of those "complete specification of the memory subsystem", or because the spec failed to stand the reality? Short memory I guess... More, why would Dell had so much hassles all the previous year with Rambus systems? Some systems were only available with scarce PC800, while other were "recommended" with pc700 or pc600? Maybe because of this "Errata 28" of Intel's 840 board? ftp://download.intel.com/support/motherboards/workstation/or840/24528806.pdf Let me remaind you the text: "In system configurations with a single 4 or 6 device PC800 RIMM installed in each channel, the system may experience memory errors.. observed behavior can range from being completely transparent to the user to causing system hangs. Workaround: Configure systems with PC800 RIMMs with 8 or more devices... or use PC600 RIMMs" Inter-operability? What a joke! <The tight tolerances imposed on RIMM manufacturers will guarantee inter-operability.> And you are saying that you have designed something... I can tell you what: the "tight tolerances" are imposed because your buddies at Rambus have very little clue on what's going on with signals. Tight tolerances are just a kludge, an insurance that sets up a huge price tag but does not really work. I hope you will stop repeating that "party line" lie covering it with your former achievements. Thank you.