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: Scumbria who wrote (40092)	4/17/2000 1:33:00 AM
From: Bilow	Read Replies (1) \| Respond to of 93625

Hi Scumbria; Re RDRAM and power consumption. This is in regard to the pump and dump site (run by a very recent employee of Rambus, who one would think, would have enough sense not to publish clearly false information about a company that he is a near insider in), "dramreview". I said that I would go through the site from one end to the other, starting with the power consumption figures, and show that they were largely fabrications. In particular, the power hype is silly: Maximum power consumption: (table) SDRAM: 11.6 Watts RDRAM: 4.6 Watts dramreview.com This would suggest that Rambus needs simpler thermal management than SDRAM, but the Intel site says otherwise, and, in fact, specifically states that thermal management of memory is a new problem, starting with RDRAM: Direct RDRAM Thermal Design Methodology Traditionally, the memory susbsystem has not required special attention with respect to thermal design. With the increase in sustainable power, this is no longer the case. developer.intel.com That's odd. If RDRAM has a maximum power dissipation of only 1/3 that of SDRAM, you would think that sustainable power would be decreasing, not increasing... Any explanations out there? It gets worse. Since RDRAM is thermally limited, this means that the designers have to be careful to prevent the user from melting the memory chips. And at 4x the cost of SDRAM, those are expensive ice cream bars. Thus the designers have to include software/hardware that throttles back the speed of memory when it is getting too hot. From the same link as above: The system integrator then can perform system design tradeoffs (ambient temperature, airflow, power consumption via performance limitations, RIMM pitch) based on maintaining the module Plate Temperature at or below that specified by the module vendor. We all know what design tradeoffs mean. Less performance, less power dissipation. It's a good thing that those RDRAM machines being sold into the small server and workstation market aren't going to get their memory worked very hard, otherwise their performance would end up suffering at the same time that they are being most highly needed. And the performance reduction doesn't come in the form of a reduction in bandwidth, which current applications don't need much of anyway, but instead, in an increase in latency, which will show up in performance measurements right now. For John Walliker, here is a link for the thermal management of RDRAM in the MCH, we used to talk about how many chips would typically be in the various power states at the same time: i82840 MCH (Memory Controller Hub Datasheet pp151-153 In a 82840-based system, RDRAM operates in one of three modes: active, standby, or nap. The number of devices allowed in each state at any given time is dictated by the heat dissipation budget specified by the system designer. From 1 to 8 devices may be in pool "A" and are configured to operate in the active mode. In addition, 1 to 16 device-pairs may be in pool "B" and are configured to operate in the standby mode. The rest of the device-pairs are in pool "C" and may be configured to operate in either nap mode or standby mode. Regardless of how many devices are configured into pool "A" and pool "B" or whether the pool "C" devices are in napping or standby mode, the system designer is responsible for providing adequate cooling for the number of RDRAM devices in the system. developer.intel.com My own guess is 1 or two active, the rest standby, at least when memory is being heavily used. But you can bet that Intel arranges for performance measurements on machines that haven't been on long enough to end up with thermal speed limits. Another interesting paper is: i840 Chipset Thermal Design Considerations (p28-29) Each RIMM is expected to incorporate a thermal sensor. For usage details, refer to the Intel 82840 MCH (Memory Controller Hub) Datasheet developer.intel.com By the way, I can't locate the "Direct RDRAM Thermal Stress Software Application Note" anywhere on the Intel website, but it is referenced in one of the above links. Anybody know anything about it? I can't imagine that what it contains will be completely pleasant for RMBS holders. -- Carl

To: Scumbria who wrote (40092)	4/17/2000 6:12:00 PM
From: Dave B	Read Replies (1) \| Respond to of 93625

Scumbria, Check out slide #8 the Intel slide presentation referenced in the following post: Message 13448008 It seems to support both of our positions. It shows the BX and 820/RDRAM800 combination as having slower latency than the Via chipsets or the 840 chipset at lower bandwidths, supporting your position. It also shows the Via and BX chipsets hitting an absolute wall in latency when the transfer rates hit about 500MBs (it doesn't even appear that they can go any higher than that). Hopefully, you, John W., and Tenchusatsu can make some sense of this for the rest of us. Dave