Intel dropping Rambus?

[Joe DeFuria]

What's this I'm reading about Intel "dropping" rambus? The link on the B3D front page is a blurb about Intel supporting DDR (which we already knew they were doing). We already knew Intel was aiming for DDR in the server market.

Is this simply the case of someone mis-interpreting news, or is Intel actually abandoning Rambus in all future chipsets starting later this year?

 

[pascal]

Maybe they want to see the market reactions (to intel chipsets using DDR and rambus)before deciding dropping rambus.

Rambus still have a place with some enthusiasts but maybe the OEM market will decide that DDR is the one to go. Then Rambus maybe will not be viable with only niche markets.

With lower latency FSB (133MHz) probably the performance differences between DDR and Rambus will be smaller, but DDR will win the price (mobo and memory) war.

 

[pascal]

Just saw this:

i850E supports PC800 RDRAM only? (INTEL)
By Johan
Wednesday, February 27, 2002 7:54 AM EST
Intel spokesmen have confirmed to us that the new 850-E chipset will feature a 533 MHz FSB but will support only PC800 RDRAM. According to Intel, PC1066 is too expensive right now, but Intel will keep evaluating the price/performance ratio of PC1066. Intel indicated that it is still possible that the i850E features PC1066 support when it will be launched (around June 2002).

A few thoughts: it is rather obvious that Intel has been slightly "traumatized" by the huge price premiums that Rambus Rimms carried at the Launch of the Intel Pentium 4, back in November 2000. Back then, RDRAM costed up to 4 times more than PC133 SDRAM and it was one of the reasons why the Pentium 4 conquered much less marketshare than Intel would have hoped. It seems that They are determined not to make the same mistake again. Nowadays, PC800 Rambus costs about the same as good quality brand PC2100...

We have no accurate pricing right now, but it is weird that PC1066 carries a high pricetag. At the end of last year, some of our industry sources told us that Samsung was getting 35% and better yields at 1066MHz. Yields should have improved now and Rambus has been showing PC1200 rimms lately. We hope to get back with more information, stay tuned.

Maybe a PC with a 533MHz FSB and a 266MHz CL2 DDR will be as fast as using PC800, but the price will be lower.

 

[mat]

Xbit says that the i850e will be the last one with rambus support.

BTW: does anyone has an idea what the Springdale Chipset (see xbit story) is? (dual channel DDR ram maybe?)

 

[Saem]

I wonder while implementing the Plumas chipsets (I believe that's the dual channel DDR chipset for the Xeon) they came up with some cost-effective technology which allows them to implement aggressive dual channel and single channel buses for cheap.

If they did, through latency and frequency scalling of DDR they could beat out DRD, but the thing is, Intel is huge on an syncronous bus. That means, DDR100, DDR133, DDR200, DDR400 and DDR533, is where want to go. The last two won't happen and even if they were possible, the DDR-II architecture will likely be in mass prodcution.

I figure they'll have the current 100MHz and 133MHz variety, come up with a dual channel version of 133MHz, then when memory manufactuers start bitching about low margins and Intel wants to cash in on new chipset/mobo sales, they'll bring out a mid range edition that runs single channel 200MHz DDR and a dual channel 200MHz DDR for workstations/enthusiasts.

That's is what I think will happen if they drop DRD. I really doubt, they will, though. DRD has somewhat "repented" for the sins of the father or grandfather in this case. (can you spot the patent joke?) Even Tom is somewhat keen on DRD. I believe the rising prices of DDR -have they stabalized?- will mean that the two memory choices will cost about the same and I don't think there will be much to complain about. Well except for folks like me and my crazy thoughts on have open standards. The two big beefs about DRD were the price difference which has almost evaporated and the moral dillemma which falls to the way side -for some folks- when better performance can be had for a small premium.

Here is just a random thought, take a few DECpaqtel engineers lock them in a room with P4 and DRDRAM, say don't come out till -quoting Star Trek here- "You make us go, we like things that make us go." I'm not sure how much benifit the P4 would see from significantly reducing the latency on DRDRAM, but it might make a differnce in the case of some DB benches. Maybe that'll only happen for the mobile (save power) and Xeon ones (performance), but the Xeon already has 603 pins, can it spare to gain ~200 more?

 

[Dave Baumann]

I must say that going with RAMBUS for as long as they have had it does seem a little odd to drop them now just as they are beginning to gain a little ‘credibility’, certainly from the financial perspective.