Rambus for G800?

[Ali]

Just read this: http://www.hardwarecentral.com/hardw...eports/2243/1/

It might be possible. Matrox have ben playing with a 64bit bus, why not stick Rambus on it?

Ali (should this be in general hardware)?

[Stringy]

I don't think it is at all out of the Question... Though I would think the royalties and poor yields may stop them from using it, but it is an interesting alternative...

Not sure anyone would want to associate themselves with the Evil Empire that RAMBUS is portrayed to be...
though when they concider things like this it's hard to like them...

Craig

[Wombat]

From the real-world perspective, Rambus won't be around long enough. As soon as their patents are knocked down, they're history.

From a tech-side, they're ignoring latency issues. Rambus takes a while to get going, and I can see some real architectural challenges because of that.

[Galvin]

Plus adding rambus would make the card way overpriced. Rambus is extremely expensive and not as fast as DDR SDRAM.

[Nuno]

Matrox have ben playing with a 64bit bus, why not stick Rambus on it?

Because they can´t. RDRAM is 16-bit technology

[Ali]

Nuno, that link talks about the bandwidth at different bus sizes. They go up to 128bit which gives 12.8 GB/sec.

All the rumours about the G800 (well, some of them) state 6.4Gb/sec bandwidth, which is the same as a 64bit bus:

64-bit / 16-bit = 4 : 4 x 1.6 = 6.4 GB/s for a 64-bit databus

I know its silly saying if Matrox have a 64 bit bus, put Rambus on it. Im sure they have a 8bit bus somewhere from the dark old says of ISA cards.

I dont think cost will be a problem. Rambus has to get money fast for their lawsuits, and Video card makers are about the only people they havent pissed off yet.

I dont know about the latency issue though. They seem to ignore it.

Ali

[pchoi]

Putting Rambus on any graphics card will shoot the price through the roof! I would expect prices like $800-2000!!!!

Plus it's not all that fast anyways, the specs tell a different story than real world performance.

[Nuno]

Yes, Ali, but you know, I never miss a good chance to bash Rambust

[dZeus]

rambus is _very_ fast, but since it runs at 600 - 800MHz, it is no use for current front side busses like on a P3 with a 100Mhz- 133MHz FSB.

Just look at the memory performance of the P4.

The 3Dfx Rampage (or whatever it will be named) chip is said to use Rambus. Undoubtedly it will be very fast

[Thundrchez]

If Matrox does put rambus on their next video cards, it's time to shop elsewhere. You cannot test the individual rambus memory chips, so they would have to assemble an ENTIRE board with GXXX chip and then see if it works. If just one memory chip has a single error in it, they would have to scrap the entire card. Also, they would have to put bigger heatsinks on the memory chips than on the GXXX chip. I'm not sure if even AGPpro can feed it enough power for rambus.

[Thundrchez]

The guys at rambus prove that they are not real designers because they do not how to do design for test.

The architecture of rambus is such that the data has to travel through all of the chips on a rimm before it goes out. It is very screwey, IMHO.

That is why 800MHz flavor of rimms are extremely difficult to make. A memory maker has to put all of their chips on the rimm board and then test to see 1. if it works and 2. how fast will it go if it does work. If it doesn't work, they have to scrap it.

It would be the same problem if video cards had rambus chips directly mounted on them. I suppose that they could have a rimm socket on them...

[Ali]

Wouldnt you just use 1 chip per path then?

Have 4 paths of 16 bit, with 1 Rambus chip each.

Rimms use about 4 or 8 chips in series dont they. (ie 4*32Mb chips to get a 128Mb Rimm), and if they work in series, the signal has to go through all 4 chips to get an output.

Im not a EE, but it seems VERY stupid that you couldnt test the chips before soldering them onto a board.

Ali

[dZeus]

Im not a EE, but it seems VERY stupid that you couldnt test the chips before soldering them onto a board

That's exactly what I find hard to believe too..

[dZeus]

You cannot test the individual rambus memory chips

Why?

[Wombat]

Actually, I am an EE ( CoE, even better). And testing the entire RIMM isn't so bad ( I think they test SDRAM DIMM's once assembled as well). Except you still have access to the chips on a DIMM.
The stupid move on RAMBUS's part is that they didn't communicate enough. The RIMM cannot run when it's assembled. The fabbing process coats everything in a heat-distributing epoxy. This isn't something that can be removed for access, or repaired.


Bad designers! No treat!