RDRAM really the better technology?

[Benedikt]

Hi,

i've heard that RDRAM is, from the technical point of view, really the most modern RAM technology... In our school (electrical engineering) I've talked to some teachers and they agree, but please correct me if I'm wrong. I mean, why do they stop supporting the better technology sometimes, and support the worse one?
I have a few questions about RAMBUS:

1.) It is a serial technology, vs DDR (parallel). Since all modern buses/ports are serial ones (Hypertransport, USB2.0, FireWire, V-Link, Infiniband), is this the future or is it not? Why are the last parallel technologies DDR and the old PCI bus (and AGP), as long as I know...?
And I heard that they are going to use serial interconnects with DDR-II (or was it DDR-III?), I mean, it isn't possible to go up with the operating frequency as high as you want? I heard that parallel connects get difficult at higher frequencies....

2) Why does Intel give up the support of RAMBUS, if it's the most modern architecture? Why, because it isn't more expensive than DDR?

3.) If you would increase RAMBUS' bus width to 32bit, you would get a very cool RAM bandwith (6.4 GB/S with dual channels and 800MHz) *gg* wow!!! NO DDR could do this at the moment....

So please help me out and give me answers ... and sorry for my bad english...

Thanks in advance!

 

[Demon-Xanth]

Well, if you increase RDRAM's bus to 32bit and still run dual channel, you end up with 64 bits. This negates the whole pin count savings advantage of RDRAM.

A major disadvantage of RDRAM is that the modules are placed in series, adding RAM adds latency.

The only reason why RDRAM has become cheaper than DDR is demand. The demand for DDR is growing while the demand for RDRAM is shrinking.

Another issue is who holds the licensing rights: A company that has pissed damn near everyone off.

IMO, RDRAM's focus SHOULD have been the embedded market. In this arena low pin count means more, adding memory isn't an issue, and the type of data transfers favors continual throughput and isn't affected as much by latency. Look at the PS2, RDRAM works great there. I just don't want it in my PC.

 

[Sahakiel]

<< i've heard that RDRAM is, from the technical point of view, really the most modern RAM technology... In our school (electrical engineering) I've talked to some teachers and they agree, but please correct me if I'm wrong. I mean, why do they stop supporting the better technology sometimes, and support the worse one? >>


Older is usually cheaper and more widely used. aka, easier. This happened during the MCA vs ISA debacle. MCA was newer, better, but more expensive and ISA was already entrenched. EISA helped tide it over until PCI came into being. In the long run, the idiot consumer has more power over technological development than the ph.d's.



<< I have a few questions about RAMBUS:

1.) It is a serial technology, vs DDR (parallel). Since all modern buses/ports are serial ones (Hypertransport, USB2.0, FireWire, V-Link, Infiniband), is this the future or is it not? Why are the last parallel technologies DDR and the old PCI bus (and AGP), as long as I know...?
And I heard that they are going to use serial interconnects with DDR-II (or was it DDR-III?), I mean, it isn't possible to go up with the operating frequency as high as you want? I heard that parallel connects get difficult at higher frequencies.... >>


I think what happened was when the older technologies were developed, serial interfaces were slow and clumsy. The only way to really ramp up speed quickly was via parallel interfaces. Now, serial technology has gotten to the point where it's faster than parallel. I think it's due to traces becoming so small and close together that high frequency parallel architectures encounter too much crosstalk.



<< 2) Why does Intel give up the support of RAMBUS, if it's the most modern architecture? Why, because it isn't more expensive than DDR? >>


DDR is cheaper and an open standard. RAMBUS is proprietary. The P4 was designed to use very high memory bandwidth (another symptom of its focus on multimedia?). It can use high FSB's much better than, say, Athlon. At the time it came out, dual-channel RDRAM was the only technology to come close to giving the P4 all the bandwidth it was designed for. Never mind it didn't really make too much of a difference. Intel came out with DDR for P4 because at the time, it was almost as cheap as SDR. The whole point of the P4/SDRAM combo was cheap, so why not do DDR, as well. Draw more people from the AMD crowd who'd already invested in massive amounts of DDR. Besides, for most uses, DDR is enough, for now.



<< 3.) If you would increase RAMBUS' bus width to 32bit, you would get a very cool RAM bandwith (6.4 GB/S with dual channels and 800MHz) *gg* wow!!! NO DDR could do this at the moment.... >>


32-bit RDRAM is in the works, but RAMBUS is having trouble pushing it forward with PC1066 coming out around the same time. It'd basically be rerouting the second channel to the other side of the memory slot which (I think) is changed to accomodate 32-bit and higher density chips.

Dual channel DDR 333 would get 5.4 GB/s, not too far off from 32-bit RAMBUS and far above 16-bit RDRAM, even PC1066. Probably the reason RAMBUS is trying to get to 32-bit. At any rate, all that memory bandwidth may go to waste (see PIII coupled with DDR) until we get higher core clocks along with higher multipliers (anyone still remember when 3x multiplier was a bit high?). I wouldn't be surprised if the higher memory bandwidth of RDRAM starts to shine as the P4 ramps up past 3GHz and beyond and compilers take advantage of its odd architecture. Then again, I could be wrong and the P4 may have already reached its bandwidth limit (highly unlikely)
Besides, a lot of people seem to love their new P4's coupled with PC133 SDRAM. *insert utter disgust here*

I read somewhere a prediction that RAMBUS and DDR will eventually meet in the middle. RAMBUS will increase its bus width and DDR will increase its operating frequency until both are pretty much the same.

 

[ShadowWolf]

The key point in RAMBUS is the sacrifice of speed for bandwidth. RDR DRAM has enormous latency compared to DDR, which makes it not suitable for most applications such as Gaming and Rendering. What makes it LOOK good is the fact that the P4 has op codes designed to mask the RDR Latency. So basically, what would have to happen is that every computer manufacturer would have to mimic op codes to make RDR useful enough to actually co-exist with standard SDRAM memory banks.

To be frank, RDR is just a trendy thing that will never last. It's just not good enough and the company (RAMBUS) is just utterly terrible anyways. They're sleezy and outright just a disgrace and those companies never survive.

if you wanna find out more, search RAMBUS in google:

http://www.google.com/search?q=RAMBUS+MEMORY
http://www.google.com/search?q=RDR+MEMORY

Just for your information, serial is FARRR better technology ( in regards to it's ability to expand and other things of the sort ) than most other techs. Serial is the future of everything as of yet

 

[Ju1cyJ]

I remeber a while back a fellow talked about a company that was going to begin making TRAM, what are the benefits to that versus DRAM? If you are still out there and took that job at the company, give us an update!

 

[Smags87]

all this depends who you ask!!!

 

[ShadowWolf]

Ok, TRAM is Transactional RAM, basically it's a Non-volitile ( doesn't loose it's data when the power is turned off ) type of RAM that basically would be commonly used as a backup of some format to replace the operations of hard-drives. To be frank, it's something that would come most in handy when you loose power. Like, in ext2fs when power is lost it has to fsck the disk to try to recover the inodes, well if you had TRAM in the system it wouldn't be necessary. You could just have a kernel that wrote to TRAM memory and just stored each system change. Thus if you lost power, the entires would be marked with their locations and information and the system would just pull that information off the TRAM and write it to Hard Drive.

In some cases, TRAM could also REPLACE a Hard-Drive. Think about it: you could just move your OS & common apps to TRAM and just use the Hard-Drive as a mass storage device instead of a random access device. TRAM is, from my gatherings, synchronous, so it runs at BUS speed. That means it would be parrelel with your standard RAM ( unless you have DRD RAM ) and you could use each parrelel with each other

Basically TRAM would rock if they could put it out, but I don't know exactly how or where it's going.

I haven't heard anything 'bout it.

reference: http://www.daemonnews.org/200012/tram.html
and always: http://www.google.com/search?q=TRAM+RAM
http://www.google.com/search?q=Transactional+RAM


Enjoy

 

[andreasl]

The only problem with the existing RDRAM design is the huge number of open pages the spec supports. I say this because none of the chipsets take advantage of this and never will. It only adds to the complexity of the design and therefor increases production costs. We will always hear people using the latency argument against RDRAM, but 99% of all those have no clue what they are talking about anyway.

Dual channel DDR-SDRAM is feasable, that's for sure. And it will be very interesting to see how it will compare with dual channel RDRAM with both performance and price considered. Intel, SiS and VIA all have products planned for this. SiS should appear first in Q2 or Q3, VIA in Q3 or Q4 and Intel most likely sometime in (early) 2003. I'm talking about desktop products here. It's too bad that Clawhammer will only support 64-bit DDR333....

 

[andreasl]

<< Hi,

i've heard that RDRAM is, from the technical point of view, really the most modern RAM technology... In our school (electrical engineering) I've talked to some teachers and they agree, but please correct me if I'm wrong. I mean, why do they stop supporting the better technology sometimes, and support the worse one?
I have a few questions about RAMBUS:

1.) It is a serial technology, vs DDR (parallel). Since all modern buses/ports are serial ones (Hypertransport, USB2.0, FireWire, V-Link, Infiniband), is this the future or is it not? Why are the last parallel technologies DDR and the old PCI bus (and AGP), as long as I know...?
And I heard that they are going to use serial interconnects with DDR-II (or was it DDR-III?), I mean, it isn't possible to go up with the operating frequency as high as you want? I heard that parallel connects get difficult at higher frequencies....

2) Why does Intel give up the support of RAMBUS, if it's the most modern architecture? Why, because it isn't more expensive than DDR?

3.) If you would increase RAMBUS' bus width to 32bit, you would get a very cool RAM bandwith (6.4 GB/S with dual channels and 800MHz) *gg* wow!!! NO DDR could do this at the moment....

So please help me out and give me answers ... and sorry for my bad english...

Thanks in advance! >>



The reasons why DDR (and later DDR-II) will succeed over RDRAM in the marketplace are political. I don't mean this from a grass-root level where people are dissing RDRAM in forums such as Anandtech.. But all memory manufactorers have chosen DDR as their preferred standard (including even though they will still produce and push RDRAM in the high-end). It's the same from the OEM's, they want DDR instead of RDR.

From all I have read about DDR-II suggests that it will expand upon DDR and add features that were excluded from DDR for cost reasons. It will boost higher clock speeds and lower voltage but will remain 64-bit. Unfortunately the memory manufacturers aren't done arguing over the specs for DDR-II anytime soon so we won't see any products based on it until well into 2003.

About the '32-bit' RDRAM. It isn't really 32-bit. It's 2 16-bit channels on the same module. So it will still be dual channel, but will only require 1 installed module in a system minimum.