It's been a few weeks since HardwareCentral's initial look into the RDRAM vs. DDR debate, and since then there have been some changes and announcements. How does the future look for Rambus now? HardwareCentral has the scoop!
A very nice, unbiased, objective article, unlike some certain website *cough*tom*cough*
Rambus vs DDR provides an interesting situation. Intel is trying to cram rambus down our throats, while AMD is going to evolutionary route and giving us DDR. If AMD plays their cards right, they may get tons of business from uses who just don't want to pay for rambus ram. Of course, if a DDR chipset doesn't come out soon, it could spell trouble.
RDRAM, is a little out of the normal users cost for performance range. Is there anything on the Tech front that might suggest another memory provider that can take advantage of the rise and fall of a clock cycle?
[This message has been edited by B/K (edited 04-01-2000).]
The "only 2 percent" portion about manufacturing cost is kinda deceptive. 2% is off the sales. If you go by Micron Tech. last quarter they had sales of 1.4 billion dollars, 2% of that is 28 million. According to the press release the profit was 161 million so if everything they sold was Rambus the royalties would be 17.4 % of the profit! quite a different percentage I would say!
1. Motherboards apparently require 6 layer pcb construction and have to be accurately manufactured to work satisfactorily with RDRAM - this increases costs
2. RDRAM in its present form appears to be pretty much maxed out - PC800 RDRAM appears to be bleeding edge in terms of production yields.
3. The trend towards on-chip L2 cache has the side effect that latency time becomes more important since cache misses will be more frequent with the smaller cache size.
Some other thoughts.
The article suggested that the i820 is a wretched chipset and should be blamed for RDRAM not performing better. Maybe. However, didn't the i820 have a new internal architecture that was supposed to remove bottlenecks? Also, I have the impression that its release was held up while it went through several revisions, some of which must have included performance optimizations. Also, while the performance of the i820 with the MTH and SDRAM is poor, it is not so bad as to suggest that the whole design is bad as one would expect a performance hit in any system using such a configuration. To compare it with the i840 is a bit unfair. Sure, one should be able to get better performance by using more memory channels. However, this negates what was supposed to be a selling point of RDRAM - low pin count. One could design a chip set using two channels of PC133 memory and do better than existing SDRAM (or RDRAM) implementations.
Maybe Willamette will show up with a super integrated RDRAM controller that will do miracles, but this remains to be seen.
I think RDRAM makes some sense for something along the lines of a PS2; a system with a small low chip count motherboard which is designed to work with a fixed memory configuration using memory from a single vendor. However, this model, if applied to PCs would be a fundamental change, and would have to provide convincing advantages to be justifiable.
Geez!. I knew latency was an issue to due the bus size difference between Rambus and Sdram or DDrRAM but I had no idea that the MCH would have to cascade latencies until they would equalize. So much for servers with 8GB of RAM. I would like to see two systems equal evrything using PC600 Rambus going head to head, one with i820 and the other i840. Just to validate that comment on the i820 chipset quality.
A.M.D.? Assassinating Microsoft Diligently
Simple and easy: Rambus technologia is not good:
very expensive, not better benchmarks, even with the i840 (you say... is better...) does not justify the very very price against its performance...
And one thing more: if Rambus suceed maybe we will have another monopoly of not very good technologi... if we cry for microsoft and intel... another tear...
Technology must be: cheaper, better and widespread acces...
visit my site (in spanish)
Carlos A. Elenes G.
So if we assume that RDRAM is way out of reach as far as real usability then is DDR SDRAM a viable alternative? Has anyone put out info on chipsets or motherboards contianing the ability to support DDR SDRAM?
If not then is it fair to compare RDRAM(a working market available product) to some dreamed up unavailable technology(DDR SDRAM for system memory use)? So far I have only seen this type of RAM (ddr) used in Video cards as of yet.....If anyone knows of a good motherboard that supports AMD K7, 4x AGP, UDMA33/66, AND DDR SDRAM, then drop me a line or some links at firstname.lastname@example.org
Simply put, the promise of Rambus has been denied. It's too expensive, the latency is way too high, and PC133 DDR SDRAM even has higher bandwidth. If Intel didn't have their heads so far up Rambus' A**, it would be dead already. There is no benefit to choosing Rambus over DDR SDRAM. It will eventually be forgotten and discontinued, but Intel is trying hard to convince everyone that it is the future. I think the computer world is becoming smart enough to recognize a smokescreen when they see it.
"Slipped my mind that I could use my brain..."
Of course, all the bandwidths of these different types of memories are theoretical and rarely does anything in eletronics live up to it's theory of operation. DDR SDRAM may fail, as well, but we don't know that yet. We do know that the performance boost from Rambus is far from acceptable, especially when you consider the cost.
"Slipped my mind that I could use my brain..."