RDR:Ram of the Future?

[Lollan]

What do ya think? ive heard it referred to this several times, and have RDR on my new system.

[merlin262]

RDRAM is a distraction.

Rambus lost out on its oppurtunity to be the future through there attempted monopolization of the market, then there corrupt business practices. The only place Rambus executives deserve to be is in jail - and those that invested heavily because of the "promise' of the company, deserve to have lost there money. I have no sympathy.

There are 2-3 new memory technologies cropping up that will displace both RDRAM and DDR-SDRAM in the coming years. I believe that we will see DDR II, and dual channel DDR ram become popular soon.

After that, I think MRAM will take over - it's faster, uses less power, cheaper, etc...

[Sharkfood]

It depends on who you talk to.

If you talk to any AMD fan, they will tell you that RDRAM was abandoned years ago, and that even Intel stopped supporting it... and you should just pretend you dont see the new 533mhz bus RDRAM systems or the new i850/i860 derived chipsets that all will take RDRAM that havent even come out yet. :)

The bottom line is: RDRAM has substantially greater bandwidth than anything else out there at this time. Quad-data rate is obtainable (true 533mhz ram bus speed) which is the current pinnacle of bandwidth.

The downside is- the way it is accessed and allocated, latency is a real crapper and gets worse as you increase the size of the allocation map. After you get a good gig into the memory bus, your latency becomes that of PC-100 memory. :)

For many server needs, this isnt a problem. Say you are running Oracle 9i on a Xeon based system- you just cant beat RDRAM. It does queries and has massive rows of data to return to the requesting client. This functionality is very tuned for RDRAM and will excel on just about any RDRAM system.

Now take your average Web Server getting 10,000+ hits a second. RDRAM will sloth along as this is small chunks of data with completely random access (ie. lots of small cached pages). DDR ram will simply obliterate an RDRAM server in this scenario as bandwidth needs are low.

For games/home use- it's becoming more and more a case for the need for bandwidth. As textures and game data grow to 256MB->512MB requirements, and these consist of just a handful of large chunks of data, bandwidth becomes more important than latency. In the "old" days, latency was a driving force since data bits in memory were tiny and tuned for smaller memory models. So the convergence is just starting to appear, which is why RDRAM benches are now surpassing DDR memory benches in newer technology. Instead of 10,000 operations working on 2k-4k data, it's more like 5,000 operations working on 90K-120K data.

I think RDRAM is here to stay until we can get memory with similar or surpassing bandwidth speed and significantly lower latency. If other quad-data rate memory ever evolves, it's anyone's guess what the latency from allocation will become. People seem to think you can just double current DDR memory without having to rethink the allocation and addressing of such technology. Unfortunately, that isn't a realistic vision in the short/medium term. If actual memory speeds dont double, then we are talking going wider (which is what RDRAM is all about).. going wider means more complex addressing/allocation, more bus arbitration to be concerned about, and more complex memory controllers to handle this. This all adds up to more latency. Surely RDRAM isn't at it's peak for how this is handled, but it's the current best available.

Cheers,
-Shark

[Mike1_2001]

Well said, Shark :)

[merlin262]

Originally posted by Sharkfood
II think RDRAM is here to stay until we can get memory with similar or surpassing bandwidth speed and significantly lower latency. If other quad-data rate memory ever evolves, it's anyone's guess what the latency from allocation will become. People seem to think you can just double current DDR memory without having to rethink the allocation and addressing of such technology. Unfortunately, that isn't a realistic vision in the short/medium term. If actual memory speeds dont double, then we are talking going wider (which is what RDRAM is all about).. going wider means more complex addressing/allocation, more bus arbitration to be concerned about, and more complex memory controllers to handle this. This all adds up to more latency. Surely RDRAM isn't at it's peak for how this is handled, but it's the current best available.

Cheers,
-Shark

The legal implications - and Rambusts manipulation of the markets have changed this though. While RDRAM IS some good tech, there is curerntly a stigma developing around it. (Note that Intel's last BIG chipset announcement - did not - suprise - include RDRAM support.) RDRAM is being shuffled to the side of the market.

With Dual Channel PC3200 RAM, you get more than enough bandwidth, and there are HIGHER clocked DDR technologies on the way. The DDR-II specification was just solidified by JEDEC, which I expect will become the standard memory technology in the future. RDRAM might be good tech, but theRambus slit it's own throat.

PC4300 memory will soon be out - giving a high bandwidth, low latency solution - and that's with only a single channel. Dual channel DDR-SDRAM chipsets(ala nVidia)..... well, you get the picture.


And here, from the EE times(this was released today) you get an idea of the what we'll be seeing in 2004:


JEDEC officials said the memory panel will now turn its attention to a DDR-III specification for a chip to follow in the 2004 to 2005 time period. Sources said DDR-III could be combined with the work of the Advanced DRAM Technology industry group, which includes Intel Corp. and a number of leading memory makers.


EE times chip center -
http://www.ebnonline.com/digest/story/OEG20010625S0104