Hi tinkershaw; Thanks for the reply.
Re: "Intel has one, one, one, desktop CPU on its roadmap ..."
By "desktop CPU" I presume you meant to write "desktop DDR chipset", LOL. If this is the case, you should note that Dell sells computers that use one, one, one desktop PC133 chipset, namely the i815. So your point about there only being one is pointless. You're suggesting that Intel should make more than one chipset for the "desktop" PC market where only one is needed. Intel only has one single channel RDRAM chipset for the desktop market, the i820, and it's pretty much moribund. The dual channel RDRAM chipsets (the i840 and i850) are "workstation" chipsets, and are not mainstream. For higher end DDR chipsets, Intel has other solutions, plus the many chipsets from VIA, ALi, SiS, ServerWorks & whoever.
Re: "... it has been stated as an interim solution at best ..."
But Intel already stated that the PC133 wasn't going to be a good idea. A simple explanation for this is that Intel makes it a practice to not recommend a memory technology unless they're currently selling it. Doesn't that make sense, from a sales perspective?
Re: "... a solution not due out until 2002 ..."
1Q02 is less than 6 months away. That is about the right amount of time for Intel to get ready to ship a major new chipset. Yeah, all the motherboard makers have DDR 845 boards demonstrating, but it takes longer to actually get the stuff ready to produce. I ought to go look up the delay between when the first i820 boards were sampled and when actual production began.
None of this industry is capable of turning on a dime. The products that will be for sale in 2002 are already decided on, and DDR for Intel's desktop is going to be there.
Re: "Now, if, if, IF, for some reason consumer out cry is so great that they demand DDR, DDR, DDR ..."
Customer's won't cry for DDR for the same reason that they haven't cried for RDRAM. They don't care. The reason is that memory is a commodity.
The primary attribute of memory is, well, memory. The speed at which that memory can be read is a secondary attribute, one that is less significant to (typical mainstream) system performance than the amount of memory. This is why DRAM is a brutal commodity based market. The bottom line with memory is dollars /MByte, and that is exactly where RDRAM sucks.
This is not the case in niche memory markets, like graphics, which is where Rambus was once supposed to shine. The reason they didn't shine in graphics is that Intel didn't have 80% of the graphics chip market available to lose in the quixotic process of trying to make RDRAM mainstream there. The only only only reason Rambus made progress in the PC area was Intel. If RDRAM was a better technology, the other companies would have all dumped into it (followed Intel) over the past 5 years, and we wouldn't be having this conversation.
You agree that Intel's the only horse pulling the Rambus wagon, and you agree that now they're looking at DDR where they didn't look before. Wall Street already figured out what that means, and Rambus' stock (which hit single digits briefly today) shows it.
Re: "Except for the niche players these companies are all followers. The only way they sell product is to follow Intel and then try to undercut it."
If all these companies are followers, how come they didn't follow Intel into RDRAM? How is it that Intel failed to make RDRAM the standard among all companies designing with memory?
Re: "Intel is the clear innovator on the desktop, the rest of the industry follow."
It was possible to argue that Intel was the innovator in RDRAM when they brought the i820 out in the fall of 1999. But now it's the summer of 2001, and still, not a single "follower" company has even hinted that they are going to someday make an RDRAM chipset for PCs. Doesn't that seem a bit strange to you? I mean really. Intel's decision to use RDRAM dates to long before 1999. How long do you think it should take AMD to decide to follow Intel into RDRAM? Isn't RDRAM supposed to be easier to design with???
Re: "For AMD DDR was an easy decision, it made Intel's pioneering move to RDRAM harder and it enabled AMD to take a crack at Intel while it was vulnerable during an architectural change."
You're mixing your metaphors. At the beginning of the paragraph you had Intel as the leader whom everyone would naturally follow. Now you've got the tail wagging the dog. Some tail. Some dog, LOL.
Re: "This company's choice was therefore made not because of a preferred technology but a fear of not obtaining enough supply."
Big companies like Motorola and PMC-Sierra sign contracts directly with the memory makers for guaranteed supplies. The real problem with RDRAM is that it's too expensive for what it provides. It's more expensive than other memory types, and it doesn't provide performance better than DDR. Samsung isn't willing (or able) to sell the stuff cheap enough (i.e. at SDRAM prices) to make it cost effective. Here's some quotes from industry figures as to why they went with DDR:
Net processor vendors dragged into DRAM war
Anthony Cataldo, EE-Times, May 4, 2001
...
PMC-Sierra Inc., for one, said it is joining the DDR camp and has promised to provide a MIPS-based processor for the control plane with an integrated DDR controller this year. The company chose DDR "purely because it's more mainstream and has as high or higher performance" than Rambus, said Tom Riordan, vice president and general manager of the MIPS processor division at PMC. The company's processors should benefit from DDR's low power consumption and higher maximum bandwidth, especially since the memory bits will be channeled directly into the processor.
PMC also had concerns about the first-access latency for Rambus DRAM, a sentiment expressed by several NPU providers. "Latency is much more important than bandwidth," Riordan said. "Networking is not pipeline-able; when the header comes in, you need to immediately look at it."
Motorola's C-Port operation, which put hooks to both SRAM and DRAM in its current network processor, faces similar concerns with Rambus. "The real performance parameter for DRAM memory technology is access time," said Dave Husak, chief technology officer at [Motorola's] C-Port. "We're using the memory as a managed buffer system that requires haphazard random-access patterns. Paying the time penalty on access to various packets in buffers will drive performance to a fraction of the burst performance. Rambus doesn't address that problem."
...
Under the agreement, Denali is including the Rambus controller interface as part of its online tool for generating memory subsystems.
The move comes six months after Denali started offering a similar tool for DDR, demand for which "has exploded," said Sanjay Srivastava, president and chief executive officer of Denali.
...
csdmag.com
-- Carl |
