Slashdot Mirror
Intel To Rambus: Long Walk, Short Pier
NightHwk writes: "A article here reports: "We made a big bet on Rambus and it did not work out," admitted Craig Barrett, Intel chief executive. "In retrospect, it was a mistake to be dependent on a third party for a technology that gates your performance." [...] "We hoped we were partners with a company that would concentrate on technology innovation rather than seeking to collect a toll from other companies," Barrett said.
"
Rambus multiplexes the heck out of everything, to get lower pin counts. Address selection involves sending a string of values accross the bus. Data comes back in a burst of values. To get even similar performance, the Rambus architecture has to run their narrower bus several times faster and faster parallel busses have more problems - their harder to synchronize, have higher energy emissions and hence potentially more cross-talk and they are subject to resonant interference at shorter distances. This last was the problem that tubed the Intel three-slot controllers. Rambus thought they were clever enough to work around these problems, which aparantly they weren't.
They also seem to have failed to find out the frequency of different memory access modes (burst sequntial vs. random, what lengths predominate etc.) in a typical system, so they produced a system tuned for long sequential data burst. This resulted in high latencies which kill Rambus performance in typical useage (though it should really fly in streaming applications).
Rambus has a long list of nasties. Any one of them might be annoying, but collectively they make the whole thing a house of cards. Issues:
There are other problems, but that's a good starting point.
Lacking <sarcasm> tags,
Well - atleast the CEO seems to be able to own up and say 'Yes we screwed up and yes we realise this', rather than the obfuscated search for a scape goat that most other companies seem to participate in these days. Hopefully this will be a trend for the future. Would be nice to hear more people admit blame and move on trying to change it than do everything they can do cover over mistakes.
This is good news! Intel must have finally gotten around to reading Tom's Hardware.
Maybe it's the beginning of them getting their head out of their arse. I wonder how much behind-the-scenes wrangling there has been between Intel ("make your damn memory work") and Rambus ("give us money") has gone one leading to this moment. Maybe Intel Management woke up during one of the meetings and said, "Hey! These guys are a law firm disguised as a hardware company! Argh!"
Anyway, I'm happy about this. Now, if Intel can just help shut down the Rambus crusade against all other memory makers, it'll be perfect. Find the rock that Rambus came out from under, lift it up, and chuck them back.
________________________________________
Napster-to-go says "Fill and refill your compatible MP3 player", which is a lie. It's not MP3. It's WMA with DRM.
Why the hell is it so hard to make non-buggy products with Rambus? I first heard about Rambus when evaluating some new SGI graphics adapters (High Impact and Maximum Impact) that were among the first devices to use Rambus DRAM. Guess what? The products were announced in mid-1995 but ran into various bugs and supply problems which led them not to really ship till beginning-1996. When those supply problems became apparent, their stock hit an all-time high (~45) never to return and it has since dropped to around ~4.
I talked to a competitor (DEC)'s engineers around that time and they said that while they'd looked at Rambus, it was not a very stable memory technology; the complexities it introduced into their engineering were not worth the performance gain and cost hit.
The fact that Intel has had the same problems as SGI, albeit on a much, much larger scale, really leads me to wonder... what is it that makes Rambus memory controllers or interfacing chipsets so damn difficult to get working properly?
If someone could answer this question, I'd be really obliged. The "toll-keeper" problems were obvious from day 1 with Rambus... the techical problems were not.
--LP
First off, Intel have never OWNED Rambus - they merely license their technology for use in their various products. Anyone saying otherwise is simply talking rubbish.
However, they DID have a fairly serious financial deal that enabled them to purchase lots of Rambus stock at knock down prices once they had met certain requirements - this indicates more than a passing interest in the wellbeing of the Rambus company.
Personally I'm still not convinced that Rambus ever had that much to offer on the technology front, so Intel's claim that they were solely in it for the technology strikes me as doubtful. RDRAM has lots of extra bandwidth for sure, but when Intel forced the system on to the industry there was really no need for vast bandwidths such as these.
Moreover, RDRAM has really no electrical characterstics that are superior to those of SDRAM, and the gap in cost is considerable even today. Furthermore RDRAM isn't really any more 'future-proof' than SDRAM, especially when you consider the future potential of DDR.
So maybe Intel have finally just woken up and 'smelt the coffee', so to speak, realising that Rambus doesn't hold the answer to all the technical problems faced by the memory industry.
Quite apart from the technical problems CAUSED by Rambus - who remembers the i820 fiasco?
Lots of pro-Rambus zealots like to point at the success of the Playstation 2 (which employs RDRAM technology in favour of more traditional SDRAM or even SDRAM DDR technology) and intimate that this somehow relates to PCs and that Rambus is therefore "better". Saying that Rambus is "better" just because it is used in a certain product is ridiculous. Windows is fairly widely used but we all know that it's not necessarily better...
So to sum up:
1. Intel are wise to get out while they still can.
arnald