RAMBUS still king!

[oops]

but you people just don't realize that dual channel DDR is going to be as expansive as RDRAM when it comes out.....so at least it's not going to take any advantage on price.

 

[Placid]

If you are upgrading and already have ddr memory then it is good.

 

[metra]

All i can say is that we really dont know. People make predictions and some predictions are backed by good points (most arent =[ ), but we just dont know. Only time will tell.

And... haha, i love how people post some statement and dont support their opinion.

 

[larva]

System efficiency is similar and only a small factor. The numbers I gave do hold relevance.

The thing is that system efficiency would affect is the difference between theoretical bandwidth and realized bandwidth. The point is that bandwidth is but one aspect of a memory subsystem's performance. The notion that RDRAM must be superior because it has greater bandwidth is pure nonsense and as such has no relevence to a educated discussion of the effectiveness of the two architectures as a whole. It's like trying to point to Sandra numbers and and say they tell the whole story, except you just calculated the result instead.

The above quote truly is pseudo-knowledge. The 6ns vs 32ns numbers have no indicative relation to real latency. Judging from this text of yours, you apparently think Rambus latency is 4 times that of DDR...

Try again, I said only that the large difference in latency of the devices was indicative of the substantial difference in latency of the memory subsystems. Unlike your above attempt to fashion numbers that support your arguement and exclude the complexity of the subsystem as a whole, I did not attempt to extend this one particular fact into a statement of the realized latency difference.

Hardly:

http://www.tech-report.com/reviews/2002q4/i845pe-ge/index.x?pg=5

A 23% difference is more like it. Compare this to the DDR333 2.7Gb vs 4.2Gb RDRAM bandwidth (a 56% difference). That is a favorable trade-off by all accounts.
*Dual channel is not taken into account here because it's latency and efficiency haven't been properly measured yet..

Huh? How exactly do you achieve the 4.2GB RDRAM bandwidth without a dual channel implementation? And the attempt to declare the bandwidth discrepancy more important than the latency difference is short sighted. The particular application's needs will determine which is a bigger factor, not a "56 is bigger than 23" leap of logic. Most home user tasks are far more sensitive to latency than bandwidth, meaning that the two percentages need not be equal to have an equal (or more) affect on the application performance.

I did not say outlive. I said both would become obsolete in about a year. Also, claiming that an installed base will somehow postpone obsolesence is irrelevant to the current discussion we have on the memory's performance.

Well, let me quote you on this one:

2) DDR memory is possibly more dead than rambus is.

To say it is "more dead" describes a lack of lifespan. You started out saying RDRAM had a longer lifespan to look forward to than DDR, but now backtrack to say that you didn't. Obsolence has little impact on how long DDR will continue to be produced and used. You may declare it "dead" from a technical viepoint, but the continued evolution of this memory type is assured along with the production of the current type long after RDRAM will be just a memory.

 

[FIZZ3]

The thing is that system efficiency would affect is the difference between theoretical bandwidth and realized bandwidth. The point is that bandwidth is but one aspect of a memory subsystem's performance. The notion that RDRAM must be superior because it has greater bandwidth is pure nonsense and as such has no relevence to a educated discussion of the effectiveness of the two architectures as a whole. It's like trying to point to Sandra numbers and and say they tell the whole story, except you just calculated the result instead.

And I did not claim bandwidth was all there is to it. See below.

Try again, I said only that the large difference in latency of the devices was indicative of the substantial difference in latency of the memory subsystems. Unlike your above attempt to fashion numbers that support your arguement and exclude the complexity of the subsystem as a whole, I did not attempt to extend this one particular fact into a statement of the realized latency difference.

Sure you did not state it explicitly, but the strong suggestion lurked beneath the surface, especially in conjuction with words like "huge" etc. My little link provides clearer information than you gave IMHO.

Huh? How exactly do you achieve the 4.2GB RDRAM bandwidth without a dual channel implementation? And the attempt to declare the bandwidth discrepancy more important than the latency difference is short sighted. The particular application's needs will determine which is a bigger factor, not a "56 is bigger than 23" leap of logic. Most home user tasks are far more sensitive to latency than bandwidth, meaning that the two percentages need not be equal to have an equal (or more) affect on the application performance.

Rambus on the P4 is always dual channel. thus the 4.2Gb.

I did not declare that bandwidth weighs more. I said 56% bandwidth is more valuable to me than 23% latency. And I might add that the benchmarks are with me on this one, however small you might deem the difference.

To say it is "more dead" describes a lack of lifespan. You started out saying RDRAM had a longer lifespan to look forward to than DDR, but now backtrack to say that you didn't. Obsolence has little impact on how long DDR will continue to be produced and used. You may declare it "dead" from a technical viepoint, but the continued evolution of this memory type is assured along with the production of the current type long after RDRAM will be just a memory.

Note that I also said "possibly". I'll explain why I said this: DDR-I will not evolve to higher speed grades than 400 at the max. Rambus has lots of plans to go beyond current PC4200. This goes from PC1333 ( http://www.gamepc.com/labs/view_content.asp?id=stateofpc1066&page=6 ) to much higher (see rambus' website). Now we already discussed the issue of chipset support. There will be DC DDR-I chipsets for a limited timeframe still. And I am aware of Intel's '850E is the last' stance. I might add though that SiS also has a Rambus license and is planning to develop chipsets.
So this somewhat more detailed picture is why I said Rambus is "possibly more alive", in the sense that there is a chance that the new modules will be accomodated in future chipsets, whereas DDR-I ends at 400Mhz no matter how you look at it.

 

[larva]

And I did not claim bandwidth was all there is to it. See below.

Um, yes you did. To quote:

Conclusions:

1) DDR can't even match PC1066 stock speed at 183Mhz 3:4 which makes the ram run at DDR 488 already.

You conclude that DDR cant match speeds with RDRAM because it has to run at 183 3:4 to provide the same bandwidth. This is false. Due to the latency advantage DDR posses it does not need to match the raw bandwidth to provide a level of performance equal to RDRAM.

Sure you did not state it explicitly, but the strong suggestion lurked beneath the surface, especially in conjuction with words like "huge" etc. My little link provides clearer information than you gave IMHO.

Read into what you will, but the statement was not made.

Rambus on the P4 is always dual channel. thus the 4.2Gb.

I am well aware of this fact. Rambus went dual channel in the P3 days because of the miserable failure of single channel systems, as even their substantial edge in bandwidth still did not raise their performance to SDRAM levels. So why this statement, surely it must confuse more than me...

*Dual channel is not taken into account here because it's latency and efficiency haven't been properly measured yet..

I did not declare that bandwidth weighs more. I said 56% bandwidth is more valuable to me than 23% latency. And I might add that the benchmarks are with me on this one, however small you might deem the difference.

No you infer that the two weigh equally, and therefore the 56% edge in bandwidth is more important than 23% in latency. The truth is home user tasks are vastly more sensitive to latency than bandwidth. In this setting (the one the users of this board almost exclusively are in) the latency and bandwidth differences are a wash. For applications that even equally benefit from bandwidth, (not seen in the vast majority of home user tasks) RDRAM shows an edge. And if you run SPEC ViewPerf 7 for a living, RDRAM is god. I guess you've found a usefull income from running SPEC ViewPerf

Note that I also said "possibly". I'll explain why I said this: DDR-I will not evolve to higher speed grades than 400 at the max. Rambus has lots of plans to go beyond current PC4200. This goes from PC1333

Rambus can plan what it wants, without Intel pressuring the entire industry to march to a RDRAM beat few enough people will care enough to make the production of such memory econimically feasible.

And as pointed out before DDR-1 that runs at 650MHz has been comercially available for over a year. The fact higher clock rates haven't been used for system memory DDR is due primarliy to a lack of demoand for the bandwidth it provides. The manufacturers have increased DDR speed implementations as fast as needed to keep up with the bandwidth demands of the audience they are playing to.

 

[OC-Master]

RD-Ram is pathetic compared to any Dual Channel DDR Configuration!

As soon as Granite Bay hits the market, Intel will close the door forever with RD and forget it even existed.

Second, DDR ram is fully backwards compatible (PC2100~PC3500) and also DDRII which also uses 184-pins.


OC-Master

 

[FIZZ3]

You conclude that DDR cant match speeds with RDRAM because it has to run at 183 3:4 to provide the same bandwidth. This is false. Due to the latency advantage DDR posses it does not need to match the raw bandwidth to provide a level of performance equal to RDRAM.

larva, you are reading selectively here. The conclusion that you quoted was a conclusion on the topic of bandwidth! Concluding something about bandwidth does not infer anything about latency like you claimed it did.

I am well aware of this fact. Rambus went dual channel in the P3 days because of the miserable failure of single channel systems, as even their substantial edge in bandwidth still did not raise their performance to SDRAM levels. So why this statement, surely it must confuse more than me...

If you were indeed aware of this fact you would not have had the need to "huh?" the 4.2Gb. The reference to the P3 platform is *utterly* irrelevant here and I think you know that very well.

No you infer that the two weigh equally, and therefore the 56% edge in bandwidth is more important than 23% in latency. The truth is home user tasks are vastly more sensitive to latency than bandwidth. In this setting (the one the users of this board almost exclusively are in) the latency and bandwidth differences are a wash. For applications that even equally benefit from bandwidth, (not seen in the vast majority of home user tasks) RDRAM shows an edge. And if you run SPEC ViewPerf 7 for a living, RDRAM is god. I guess you've found a usefull income from running SPEC ViewPerf

Apparently, I "read into" your HUGE 6ns versus 32ns statements some things that are not there, but what is it that you are doing here? I *never* said latency and bandwidth are equal. I said the bandwidth advantage overpowers the latency disadvantage in this case. This is what the benchmarks have shown all along.

So you think that the latency advantage of DDR makes it the king of "home user" applications (whatever those are)?

Maybe you mean things like games? Maybe 3DMark or Jedi Knight will show the advantage? Or UT2003?

http://www.gamepc.com/labs/view_content.asp?id=p4pe&page=6

http://www.gamepc.com/labs/view_content.asp?id=p4pe&page=7

http://www.anandtech.com/mb/showdoc.html?i=1738&p=9

http://www.anandtech.com/mb/showdoc.html?i=1738&p=10

Hm guess not.

Photoshop then, who doesn't draw some pics every now and then? Surely your latency theory will prove itself here.

http://www.gamepc.com/labs/view_content.asp?id=p4pe&page=9

Nope.

Well maybe SYSmark is a better gauge of "home user" apps then.

http://www.anandtech.com/mb/showdoc.html?i=1738&p=8

No, Rambus still wins. Where is the vaunted latency advantage then? Heheh certainly not here:

http://www.anandtech.com/mb/showdoc.html?i=1738&p=11

Rambus can plan what it wants, without Intel pressuring the entire industry to march to a RDRAM beat few enough people will care enough to make the production of such memory econimically feasible.

I agree that the chances are somewhat unfavorable, that is why I formulated it as I did. =)

And as pointed out before DDR-1 that runs at 650MHz has been comercially available for over a year. The fact higher clock rates haven't been used for system memory DDR is due primarliy to a lack of demoand for the bandwidth it provides. The manufacturers have increased DDR speed implementations as fast as needed to keep up with the bandwidth demands of the audience they are playing to.

Again you bring up the graphics memory dream. It is time for you to realize that DDR333 is the FINAL spec. Higher than semi-official DDR400 it will NOT go. You can fantasize about graphics ram on your chipset all you want, fact is that it will never happen.

 

[FIZZ3]

RD-Ram is pathetic compared to any Dual Channel DDR Configuration!

GB at DDR133 (PC2100) and PC1066/4200 are about equal. So are GB DDR100 and 850E PC800.

http://www.tbreak.com/reviews/article.php?cat=&id=166&pagenumber=5

http://www.anandtech.com/mb/showdoc.html?i=1748&p=18

http://www.anandtech.com/mb/showdoc.html?i=1748&p=19

Second, DDR ram is fully backwards compatible (PC2100~PC3500) and also DDRII which also uses 184-pins.
OC-Master

DDR-II is not compatible with DDR-I modules nor chipsets.

By the way, PC3500 is just cherry-picked PC2700 that can be overclocked. Kind of like OCZ Rambus PC1200. The real DDR-I list goes from DDR100 to DDR133 to DDR166 (PC1600, PC2100 and PC2700).

 

[larva]

Funny, one of your links contains the following statement, under the section "The Final Word"

"With some high-quality DDR-333 memory sticks and tight motherboard BIOS DRAM timings, Intel 845-PE motherboards can give gaming and application performance within a percentage or two of PC-1066 RDRAM products, at a much lower cost. "

http://www.gamepc.com/labs/view_content.asp?id=p4pe&page=10

This was the whole point. Even the relatively pedestrian 845PE with good ram an intelligent bios settings virtually equals PC1066 RDRAM's performance for "gaming and application performance". Since we well know the it does not posses the raw bandwidth of RDRAM, this can only happen because of the latency advantage enjoyed by DDR, and it's relevence to home user tasks. Pay as much as you want for a percentage point or two, it doesn't make sense. And hope you don't run into a really hot 845 rig, as it can and will eclipse the perfromance of your treasured RDRAM no matter how highly you think of it.

 

[FIZZ3]

Funny, one of your links contains the following statement, under the section "The Final Word"

"With some high-quality DDR-333 memory sticks and tight motherboard BIOS DRAM timings, Intel 845-PE motherboards can give gaming and application performance within a percentage or two of PC-1066 RDRAM products, at a much lower cost. "

http://www.gamepc.com/labs/view_content.asp?id=p4pe&page=10

This was the whole point. Even the relatively pedestrian 845PE with good ram an intelligent bios settings virtually equals PC1066 RDRAM's performance for "gaming and application performance". Since we well know the it does not posses the raw bandwidth of RDRAM, this can only happen because of the latency advantage enjoyed by DDR, and it's relevence to home user tasks. Pay as much as you want for a percentage point or two, it doesn't make sense. And hope you don't run into a really hot 845 rig, as it can and will eclipse the perfromance of your treasured RDRAM no matter how highly you think of it.

I am aware of that, larva. You said specific things about latency being more important than bandwidth that I addressed. We had already agreed before that it's a matter of personal choice whether you want to pay for the extra performance or not.

Maybe it will interest you to know that my most 'treasured' ram is actually a 256Mb stick of Mushkin Rev.2 PC133 SDRAM that did 160Mhz at 2-2-2, 5-7t!

 

[larva]

I said exactly what your lind states in my original post. You started a tirade saying that wasn't the case. I said the advatage of RDRAM varies between insignificant and inconsequential when compared to a optimized 845 platform using good ram. This is exactly the conclusion your link states. The point about the latency is that it is a bigger factor, which it must be in order for DDR systems to in any way compete with RDRAM ones, because as you pointed out the RDRAM bandwidth advantage is larger numerically than the latency edge of SDRAM. Where it not for the fact that the tasks most use these machines for are more sensitive to latency than bandwidth the results could not be as close as a percentage point or two.

 

[FIZZ3]

I said exactly what your lind states in my original post. You started a tirade saying that wasn't the case. I said the advatage of RDRAM varies between insignificant and inconsequential when compared to a optimized 845 platform using good ram. This is exactly the conclusion your link states. The point about the latency is that it is a bigger factor, which it must be in order for DDR systems to in any way compete with RDRAM ones, because as you pointed out the RDRAM bandwidth advantage is larger numerically than the latency edge of SDRAM. Where it not for the fact that the tasks most use these machines for are more sensitive to latency than bandwidth the results could not be as close as a percentage point or two.

Hm I was looking forward to some comment about my treasured SDRAM... oh well.

I think I've made my points sufficiently clear, no further tirading required.

 

[Goo Kenson]

but you people just don't realize that dual channel DDR is going to be as expansive as RDRAM when it comes out.....so at least it's not going to take any advantage on price.

No...dual channel DDR can use any DDR sticks we have now. Assuming we already have some DDR RAM, the only cost will be the DCDDR mobo and possible a second stick if we only have one.

 

[Fluxer]

256mb Corsair c2 pc2700 DDR - $114

256mb Dane Elec - Samsung pc1066 RDRAM - 16bit - $113

You would need to buy two of each sticks to work with either dual DDR or dual RDRAM. I am using the exact RDRAM found on the link above(it has Samsung pc1066 chips, with no mention of Dane Elec and a Samsung sticker on it -- plus the RIMMs are single sided and don't get that hot), which I bought from newegg.com (100% stable, 24+ hours of Prime 95 without any errors and no detection or compatability issues).

I haven't tested the Corsair pc2700 cas2 but I think it can be agreed that you can't go wrong with Corsair.

Yes, RDRAM is a lot more expensive, right? In addition, dual DDR has gone through years of testing to fully optimize data transfer like RDRAM, right?

Anyway, feel free to correct me on prices or whatever -- I don't keep up with DDR prices as they change nearly every day. Maybe i'm wrong, but even if RDRAM is more expensive (which I suspect) it's still a great solution for a pentium 4 system.

 

[larva]

The best DDR value I have found is the 256MB Kingston Value Ram PC2700 I bought recently for $74. The corsair is good if you can't find a reasonably priced alternative that is proven for your chipset. On my 845e and pe machines the Value Ram ran at 430+ MHz at cas 2 (PC3500 speeds).

I agree the cost delta has never been less between DDR and RDRAM at present. The performance delta has never been less as well though. RDRAM certainly is high performance solution, the whole point I was making was niether RDRAM nor granite bay has a striking advantage over the level of performance 845 series chipsets have attained at 333MHz memory speeds, and for those who are looking to overclock 845 has proven capable of running even inexpensive PC2700 to PC3500 speeds and beyond. Given this kind of bandwidth DDR systems (even single channel ones) are extremely compeitive with all other solutions and do so at the lowest cost. As an added bonus they achieve the highest FSB of the architectures present, a useful characteristic in light of the quality of Intel's procduction these days.

 

[Maxvla]

this thread is full of crack heads.

anyone saying rdram is more expensive than ddr hasn't checked prices in a LONG TIME.

also. all these people saying its stupid to buy a rdram board now when ddr-II is just around the corner are also stupid.

rdram is the best right now! there is no dispute.

also... rambus has been doing dual channel for quite some time now and they've pretty well figured it out. DDR is just now attempting dual channel so they will certainly be having problems with chipsets, etc for at least a few months after it's release. i'm not sure about you but i do NOT want to beta test hardware. especially memory which can wipe out your hard drive at any time.

i might look into dc ddr in about a year. by then it should be running fairly smoothly and we'll see what happens with rdram in that time period.

 

[AZN]

this thread is full of crack heads.

its full of crackheads that failed in math.

$87 GOOGLEGEAR.com RDRAM 256MB PC800 NonECC
(Kingston Technology)

$99 GOOGLEGEAR.com DDR 256MB PC3200 Cas2.5
(corsair)

$103.5 GOOGLEGEAR.com RDRAM 256MB PC1066 Original Rambus 32Bit
(samsung)

$116.5 GOOGLEGEAR.com DDR PC3500 DDR 256MB 3500C2 (corsair)

that price was taken from PW on NOV 24-2002

Math was my strongest subject. the DDR cost more then the RD ram and even the PC1066 RD.

when we say RD cost more then DDR were we comparing it to PC2100??? and if u want to get corsair CL2 then u will be paying even more.

AZN

 

[goldenlight]

The way Rambus stock is going, IT will be no more.

 

[Maxvla]

The way Rambus stock is going, IT will be no more.

does that make the ram any slower?

no.

thank you drive through.