HI Plaz; Hi Plaz; You're right, my use of the term double buffering was incorrect. (Perhaps I should note that I type these things in at a very high rate, and don't always proof them carefully.) Allow me to rewrite that paragraph, with the changes in italics:
>>>
A simple way of expressing the graphics bandwidth problem is that each bit of graphics display memory, has to be read from or written to once for each frame, assuming that the memory is needed for the display, and that it must change when motion occurs. More modern high end graphics controllers can use more memory to describe the image than the completed image actually contains. As an example, the Nvidia DDR system uses 256Mbits = 32MBytes of memory. In order to read and write all that memory 60 times per second, you would need a bandwidth of 32MBytes * 120/sec = 4GBytes/sec, which is about how much bandwidth the Nvidia card actually provides. Thus with more modern technology, the memory bandwidth problem doesn't get better. It still boils down to something like 120 or so reads or writes per bit per second.
>>>
Incidentally, the above statement, that the memory has to be read to or written from about 120 times per second for each bit, also holds for the first IBM text only display. That display used a ROM to convert the text to an image, and so it required only about 4KBytes of memory, far less than a graphics display would require. Yet the bandwidth constant applied to that display as well. That is, you still had to be able to read the memory 60 times per second, and you wanted to be able to also write the memory 60 times per second. Thus, even in that very ancient, very non-graphical type display, my bandwidth constant of 120Hz/bit held. The same statement applies to other graphic display technologies, such as LCDs. Note that with LCDs, the graphical image may be refreshed at a rate far lower than 60Hz, due to the limitations of the display medium, and this reduces the bandwidth requirement. But it also reduces the amount of motion that the technology can display.
Another way of seeing this beautiful constant, is that the unifying fact of video displays is that they must produce an image for the eye. The eye has a bandwidth limit of something near 60 frames per second, and because that image comes from memory, with some sort of translation in between, you have to have that in order to support full motion video, the memory must have a bandwidth of about 120Hz per bit. It just doesn't matter if it is text, dumb frame buffer, 2D, 3D, or, likely, some future technology.
I hope that this was a more complete and compelling explanation. It is a fundamental fact about displays. Most of the last 20 years I have spent designing memory systems for various kinds of displays, everything from LCD to stuff that I can't name, and they all have to access memory at this same rate, 120Hz.
There are some caveats to this. If the display has large amounts of memory that are not changing, such as a cursor or primitive character definition, then that memory doesn't have to be written to. And if some memory has to be read repeatedly, such as a text based display where the font is stored in the same memory as the text, or when the memory must also contain software code that is repeatedly used, (i.e. inefficient cache) then that region of memory has to be read more than would be otherwise indicated. But overall, it is a good rule of thumb, especially when you are trying to illustrate the things that have stayed the same during the last 20 years of radical display card design changes. You see, the point is that as the amount of memory used in a display card increases, so does, proportionately, the bandwidth need of the display that uses it. This is NOT something that applies to general purpose CPU memory, for instance.
A lot of time, theory doesn't match practice too well. But the display industry has followed the rule I gave reasonably well for the last 20 years, from text based displays through the various vector graphics chips, on through the general purpose graphics processor chips, and now on to complicated 3D rendering stuff. I think that's cool. If you carefully read my explanation, I think that you, too, will see why this has happened.
You wrote: The "moderness" of the technology has nothing, absolutely nothing, to do with the bandwidth limitations. We're bandwidth limited because we are expecting so much more from video cards these days and, at high resolutions and pixel depths, most any 3d chip (GeForce, VSA-100, TNT2U), V3, etc)is able to outperform current, cutting edge memory technology.
Let's remember that this the Rambus thread, and that the subject of conversation here is memory chips, not graphics technology. Graphics technology has, over the years, desired a pretty much constant 120Hz (i.e. 60 to 180Hz) bandwidth per bit. I think that this is adequately explained by the bandwidth of the human eye. I don't think that it is going to change much. Any memory chip that is going to be used by the graphics industry is very likely to have to put out about 120Hz per bit. We could go back through history, and we would see this. I have an explanation for it. You call it a coincidence. It really isn't a big deal.
You wrote: I'm not qualified to judge most of your technical posts as I'm not a DRAM engineer, but this one leaves a lot to be desired.
If you work in an industry for long enough, you are bound to notice little coincidences that eventually make sense to you. If you were a long-time graphics DRAM engineer, you would have likely noticed this same coincidence, assuming you're the type that notices coincidences like that, and, if you were the sort that thinks about things like that, you might have come up with the same explanation. But it really isn't a big deal. It just means that as soon as I know how much memory the graphics guys want to include, I have a pretty good idea how much bandwidth I will have to provide them. If they ask for a lot more memory, but claim to not need the bandwidth, or vice versa, I know to start asking questions.
Sorry for the confusion re "double buffer," my mistake.
-- Carl
P.S. You quoted the word "moderness" as if I used it. Even late at night, I try to avoid words like that.
P.P.S. If you can read an advertisement under this line of text, but still within the box, well it doesn't have anything to do with me, and is likely to either waste your time or cost you money if you click on it. |
