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: Jdaasoc who wrote (35258)	11/26/1999 7:38:00 PM
From: unclewest	Read Replies (1) \| Respond to of 93625

It appears that DDR memory for graphics adapters will get us through 2000(4XAGP) and DDR-II will get us through 2001 (8XAGP). What type of memory do you expect for 16XAGP in 2002. john, great question. my guess is won't be ddr II or ddr III. i have some ee phd friends, who have told me that ddr II has more problems than rmbs will ever have. unclewest how's the new jag running?

To: Jdaasoc who wrote (35258)	11/26/1999 10:12:00 PM
From: Bilow	Read Replies (3) \| Respond to of 93625

Hi Jdassoc; Re graphics memory requirements... I would think that the Rambus people would let this sleeping dog lie, given that the DDR x32 SGRAMs, along with embedded, have pretty much cleaned Rambus out of new design wins in this area for some time. But this is a subject near and dear to my heart, so I am happy to expound on it. There are two kinds of bandwidths required for video. The image bandwidth is limited by the resolution of the display, and the vertical frame or field rate. The other bandwidth is that required to assemble the image, and I am going to only mention that limit in passing. Display bandwidth is the proportional to the product of vertical refresh rate, number of bits per pixel, and the number of pixels per display. Traditionally, the vertical refresh rate is 60Hz or somewhat higher, and the human eye obviates the necessity of changing this vertical refresh rate. (The flicker fusion frequency for humans is more or less a constant over time. More modern graphics systems do have higher rates than were used in the past, though. But beyond 120Hz, I doubt that there is anyone on the planet who will complain much.) There are two other contributing factors to video display bandwidth: the number of bits per pixel, and the number of pixels per screen. Bits per pixel is already maxed out, the human eye has trouble discerning the difference between adjacent pixel values as it is. Screen resolutions are limited by the display device, not the electronics, and have historically increased at a rate far under 2x per year. So video bandwidth, as measured at the DACs, has been increasing at a fairly slow rate, not the 2x per year that the AGP figures would suggest. On the other hand, there is no doubt that getting your graphics card to update at 60Hz is what you want to do. For complex images, this can mean a higher computing bandwidth than is available with current graphics controllers (at least the affordable ones). But embedded provides a much higher bandwidth than direct RDRAM, or any possible future version of RDRAM. Staying on the chip is the way to obtain speed. The design issue boils down to one of obtaining a very high bandwidth out of a given amount of memory. Unfortunately for Rambus, the size of the memory required is limited by the display device. There is no reason to put a 1G bit RDRAM chip (circa 2002) into a graphics controller that is limited by its display to a resolution of 2048x1500 or so. Embedded memory is ideal for smaller memories where very high bandwidths are required. Traditionally, video display memory has been too large to place in embedded memory, though this is changing. The rate of increase in embedded memory size is much much higher than the rate of increase in picture resolution. It's pretty clear that the curves are going to cross fairly soon. As an example, a 2000x1500 screen, with full 24-bit color requires 2000x1500x24 bits = 9MBytes = 72Mbits. This is within the realm of current embedded DRAM limitations even though 2000x15000 displays are still kind of high end. Because of the historically high rate of increase in embedded DRAM size compared to the historically low rate of increase in numbers of pixels per display, it is clear that embedded is going to solve the video memory problem, and, further, that it is going to solve it better and better as time goes on. The other part of the video engineering problem is the memory required for describing a synthesized image. That is a little outside my specialty, but I think that the scaling laws mentioned above will also put these solutions into the embedded class. We should note that Nintendo chose embedded DRAM for its next graphics controller. The previous version used the older variety of RDRAM. Perhaps my calculations above will illustrate just why that design win went to embedded. Simply put, RDRAM is scaling to a size too big for that market. Embedded wins on granularity concerns, as well as the fact that it has a much, much, much higher bandwidth than RDRAM, and quite a bit lower latencies and power consumption. -- Carl