Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Rambus (RMBS) - Eagle or Penguin -- Ignore unavailable to you. Want to Upgrade?

To: John Walliker who wrote (26086)	7/31/1999 10:56:00 AM
From: J_W	Respond to of 93625

John, In addition to the the DDR heat issues are performance issues. From the Dell white paper: dell.com Other emerging PC memory technologies address the need for improved memory performance with higher clock rates and/or transfers on both edges of the clock (for data only). While these techniques are effective, they do not scale well. A PC 133 DDR memory system with a peak bandwidth of over 2 GB/sec may be available within months of the introduction of Rambus; however, the effective bandwidth will still be less than that of Rambus due to Rambus' superior pipelining and command handling. SDRAM efficiencies will be approximately 60 percent while Rambus accesses can be scheduled to reach up to 95-percent efficiency. Rambus offers a major advantage for future bandwidth expansion. Adding bandwidth with Rambus consists of adding another channel with 33 signal pins. In comparison, an additional SDRAM interface requires 132 pins. Although an additional 132-pin interface may be a reasonable approach for expanding server memory, it is not appropriate for workstations or desktop PCs due to the component cost and system board space required. Moreover, increasing the clock rate or widening the data path beyond PC 133 parameters for a single SDRAM channel will increase the difficulty of controlling emissions, maintaining signal integrity, and meeting timing margins. What this says is that PC266 will have a peak data rate of 2GB/sec and a sustained rate of 1.2GB/sec. Rambus will have a peak rate of 1.6GB/sec and a sustained rate of 1.52GB/sec. For video and streaming applications, the AGP port(x4) needs, the sustained rate is very important. Now what happens when you run out of headroom using DDR? Do you up the transfer rate? That will bring out a multitude of new frequency related issues making implementation very difficult. Do you add another DDR channel? That will add another 132 pins creating yet more system board emission and signal integrity problems, not to mention the additional real estate on the system board. It is no small wonder that Intel decided to go with Rambus. Intel has to look beyond the current generation of processors and applications. What they saw is a dead end using SDRAM. Beyond PC266 DDR SDRAM simply does not scale well while DRDRAM does. Regards, Jim

To: John Walliker who wrote (26086)	7/31/1999 12:07:00 PM
From: Matrix_Man	Read Replies (1) \| Respond to of 93625

John, I truly do appreciate it when someone who is technologically savvy is able to explain the real retails of why Rambus has the currently superior solution. My undergraduate degree was a BS in physics and I enjoyed your explanation. I believe that the only solution which will be better is complete embedding of DRAM which is not yet possible. I believe this is why Intel selected this solution. I believe this is why it will dominate in the years to come. I believe that if they build more MHz, the applications to use it will come and ensure its success. FTR - long RMBS, and staying that way for now.

To: John Walliker who wrote (26086)	7/31/1999 8:28:00 PM
From: grok	Read Replies (1) \| Respond to of 93625

RE: <One of the reasons why there is a time delay between data transfers is that the signals on the interconnection bus oscillate backwards and forwards between any discontinuities in characteristic impedance> Yes, rambus is an elegant solution for driving signals but it has to be since they have up to 32 drams on the same signals running at 800 MHz data rate. A typical DDR system would have many fewer loads since the DIMM width matches the data bus width and usually a data signal only needs to drive 1/4 as many dram chips at, in this case, 1/4 the rate per signal. And sdram signals are also terminated so operating at the planned frequencies can and will happen. Not as elegant as rambus but they'll do the job. In fact, the DDR-II committee is working out how to handle 400 to 532 MHz data rates in the future. Of course, at that time the signal driving methods may look a lot more like rambus than they do today. Oh, by the way, I think that the rambus world has given up on supporting 32 drams on one channel due to electrical difficulties so sh*t happens on both sides. RE: <It is also wrong to suggest that doubling the data rate will not increase the power consumption. In CMOS circuits power is dissipated as a result of charging and discharging circuit capacitances. If the rate of doing this is doubled then so is the power consumption (if the voltage stays the same).> Yes, of course the power in the data lines doubles between 200 and 400 MHz. But if you reread my original post you'll see that I said something like ... DDR power is not worse than rambus. And, it is true that power if one of rambus' biggest weaknesses. Read up on wind tunnels and active pool miss latencies if you have any doubts about this. RE: <Another issue is that increasing the data rate of conventional RAM will considerably increase radio frequency emissions. Many aspects of the Rambus design minimise these.> You may be right on this. At least I've heard rambus advocate this and I haven't heard anyone refute it. But I don't really understand the rambus advantages on this issue. RE: <Yet another issue is that conventional clocking arrangements fail when the wavelength of the clock signal is of the same order as the dimensions of the conductors. Rambus overcomes this by using two clocks, one for each direction of signal travel.> This is the first I've heard of this wavelength issue. Sounds like something worth checking into.

To: John Walliker who wrote (26086)	7/31/1999 9:48:00 PM
From: The Prophet	Read Replies (2) \| Respond to of 93625

John: Based on your expertise, would you say that DDR can compete with RMBS for some indefinite period of time before RMBS' scalability ultimately defeats it? Or would you say it cannot compete at all? Prophet

To: John Walliker who wrote (26086)	4/9/2001 9:03:29 AM
From: Scumbria	Read Replies (1) \| Respond to of 93625

John, It is also wrong to suggest that doubling the data rate will not increase the power consumption. In CMOS circuits power is dissipated as a result of charging and discharging circuit capacitances. If the rate of doing this is doubled then so is the power consumption (if the voltage stays the same). The number of times that the transistors switch within a certain time period is determined by how much data is transferred to the CPU. DDR typically shows a 5-10% performance improvement, so it seems reasonable to expect a comparable increase in number of transitions per unit time. DDR also has more idle time than SDRAM, because the data transfers occur more quickly. Scumbria