Hi JD_Canuck; Re "RDRAM is faulty and doesn't work" You're putting words in my mouth. What I wrote was "When it turned out that their designs were faulty, we ignored them and moved on to DDR designs." The word "faulty" does not mean "doesn't work", except, perhaps, to someone who's idea of engineering is simplified to that of designing something that works. Actual engineers have to design things to be cheap, and it is mostly in the expense that RDRAM failed. By being expensive, RDRAM was incompatible with the role it was designed for, that of the world's mainstream (and therefore cheapest) memory. By the way, for today's English lesson, here's the definition of "faulty" from Webster's Revised Unabridged Dictionary:
\Fault"y\, a. 1. Containing faults, blemishes, or defects; imperfect; not fit for the use intended.
dictionary.com
Here's a reminder of what the Rambus story was back then about RDRAM:
...
The market for Rambus DRAMs will take shape in 1999. Intel, which holds at least 70% of the world semiconductor market, has selected the Rambus chip as the main memory for next-generation PCs. Therefore, the Rambus chip will quickly become one of the mainstream products of the world DRAM market from 1999.
...
samsungelectronics.com
For that matter, what Rambus sold to Intel was a design that was supposed to handle 4 RIMMs per RSL. That was reduced to 3 by the time of the Camino fiasco, but they weren't even able to achieve that goal, and since then RSL channels have been restricted to only 2 RIMMs. By contrast, registered DDR allows up to 4 DIMMs and is shipping just fine. Here's a reminder from back in 1999:
news.com.com
I may as well answer your comment today to Skeeter Bug: "The last benchmarks i saw showed RDRAM leaving DDR(as it presently exists) in the dust on a performance basis as the P4 speeds ramp up, the pricing is similar, and isn't this what INTC told the world all along?" #reply-17156907
(1) Those benchmarks you were looking at were comparing a single channel DDR mainstream desktop chipset (the i845) with a dual channel RDRAM performance desktop chipset (the i850). So where's the dual channel DDR performance desktop? Wait until later in the year.
Mainstream chipsets are for boards that are cheaper than the boards that are made from performance chipsets. For example, on Pricewatch, the lowest 12 prices for i845 and i850 motherboards are currently:
i845: $ 73 75 76 79 80 80 81 84 85 85 85 85
i850: $ 85 95 95 97 100 102 107 107 107 109 109 110
pricewatch.com
pricewatch.com
When Intel decided that they had to "become memory agnostic" (which since Intel's support was the only thing keeping RDRAM alive at all really meant that they were dumping RDRAM) they naturally had to do it in the most important markets first. For them, that meant that the first thing they absolutely had to do was to get out a replacement for the horrible i820 Camino in the value segment. The value segment has more sales than any other, so that was where Intel put their efforts first. The replacement for that chipset was the i815, which used PC133 SDRAM.
After the value segment, the next largest segment was mainstream desktop, and that is where Intel brought in the i845. That segment has now been mostly cleaned of RDRAM, as a trip to a few retailers will attest.
The remaining segment, performance / workstation, was smaller than the others, and so the chipsets for that market were given lower priority (and in addition require more testing anyway) than the value and mainstream desktop. So that's why Intel was delayed in that segment until the Granite Bay chipset comes out later this year.
The network processing segment is smaller still, so it could be years before Intel replaces their NPU that uses RDRAM. Intel's server chipset group never really bought into the Rambus mirage, so their DDR server chipsets were not particularly delayed, and came out contemporaneously with the same stuff from Broadcom / Serverworks.
This whole thing has been a horrible embarassment for Intel, which has been forced to trod the path that was blazed by AMD. I doubt that they're going to make that mistake again, at least anytime soon.
But have no fear, you'll be seeing benchmarks where dual channel DDR chipsets wipe the floor with Rambus soon enough. What's going on is that Intel is slowly and methodically wiping RDRAM from its product lines. It only makes sense for them to do this starting with the product lines that ship the most product.
(2) The pricing is not yet "similar", in that RDRAM is still considerably more expensive than SDRAM or DDR. To get to similar pricing, they need to have RDRAM within 10~15% of SDRAM. This simply is not going to happen. Back in 1999, Rambus (and Intel) promised that this would happen in 2000, but as Intel said, trying to put the best face on a decision that cost them half the chipset market:
"Hindsight is wonderful. The technology is not problematic. The technology is still the best technology out there for showing off the microprocessor. The problem is the cost structure. The cost structure never met the goals we had. I look back and say for the time, I really think it was the right technology."
news.com.com
And what were those "cost structure" "goals" they had? Funny, but Samsung told us exactly what that goal was, and that it was unattainable, back in April 2000:
Intel loosens timing spec to spur Rambus usage
Anthony Cataldo, EE-Times, April 20, 2000
...
[Pete] Mueller [engineering manager for the platform components group at Intel] said he's pushing DRAM makers to get the RDRAM premium to 20 percent about SDRAM prices by the end of the year [2000]. That could be possible if more DRAM makers come into the market this year, and if SDRAM prices continue to rise throughout the year, which would help close the price gap between the DRAM types, he said.
But one of the current RDRAM vendors, Samsung, denied that prices are artificially inflated, and said there is no great difference between the RDRAM selling price and its cost of production. "I've heard that 20 percent figure from Intel about 10 times," said Jay Hoon Chung, manager of DRAM marketing for Samsung Electronics Co. (Seoul, South Korea), currently the largest supplier of RDRAMs. "But 20 percent is not probable by our point of view. We expect the price gap will be 1.5x by the fourth quarter."
...
eetimes.com
In other words, back in April 2000, even Samsung, Rambus' best buddy in the DRAM industry, was saying that RDRAM wasn't going to get down to Intel's cost goal. By that time, Intel was already hard at work coming up with RDRAM alternatives. But it takes time to design, debug, and test chipsets, so it took until this year before Intel actually had any DDR chipsets available. Until then, Rambus morons could sit there and amateurishly sneer (and did) that Intel was still supporting RDRAM only. BWAHAHAHAHAHAHA!!!
Now do you see why it was obvious to the industry that RDRAM was dead years ago? It was doomed by its higher cost of production. All this was simply labeled as FUD by the Rambus faithful and ignored. And they got slaughtered in the stock.
So no, what Intel "told the world all along" (your words) was that RDRAM would sell for within 20% of SDRAM. It hasn't got there. It may briefly reach that point, but since the majority of the world's DRAM makers weren't interested in making RDRAM when it was selling at 100% over SDRAM's price, why would we be so stupid as to imagine that they're going to suddenly ramp it up when it's fallen to 20% over SDRAM's price? They didn't ramp RDRAM all last year when SDRAM prices dropped through the floor, and as a result they all lost money. The simple reason that they didn't ramp RDRAM up is because their cost for making the stuff was higher than RDRAM was selling it for. This reason doesn't apply to Samsung, because Samsung is the lowest cost producer for very high end memory. As an example of Samsung's expertise, note their recent announcement of 800MHz DDR SDRAM chips.
By the way, remember back to June when Winbond was supposed to start ramping up RDRAM production? Here's a reminder:
Press Release, Winbond, June 27, 2001
Winbond to Provide Rambus RDRAM Memory Devices
Winbond Electronics Corporation and Rambus Inc. today announced Winbond is planning to ramp up 256M RDRAM production volume from Q4 2001. Winbond will support the growing demand from OEMs for RDRAMs and RIMM modules for use in performance desktop PCs and workstations.
...
winbond.com
Here's Winbond's DRAM and SRAM product lines now (nearly 2 months after the end of 4Q01), notice that it has DDR but still no RDRAM:
winbond.com
If someone had been paying attention to Winbond's press releases they wouldn't have been surprised that they didn't follow through. Here they are promising to make put RDRAM into production back in 1999:
winbond.com
The Winbond story has been repeated with most of the rest of the industry. No one can make RDRAM at all cheaply except Samsung, and even for Samsung it is an expensive memory type.
Because of this simple fact, RDRAM is stuck in niche status. If Dell, HP, and Compaq were to suddenly (and stupidly) reverse their course and drop their DDR computer lines in favor of RDRAM, the resulting demand for RDRAM would drive the price back through the ceiling. Samsung would make good money, but the other chip makers wouldn't react to the high prices by turning on the RDRAM pump. The other memory makers lose money on RDRAM so they won't make it.
Intel knows this, the other design engineers know this. It's obvious to us all, that's why we're not out there converting DDR designs over to RDRAM. And for that matter, RDRAM makes other parts of a design more expensive, but that's another story.
-- Carl |
