Why isn't there a backlash against DDR SDRAM like there was with RDRAM?...RE: VIA KT133A

[hungrypete]

some of us were thinking it would do for the whole computer what ddr ram has done for graphics cards. It, apparently, does not. Good thing Im too poor to buy this stuff when it first comes out

 

[Noriaki]

I can't give you a picture of my specific case as I don't own a camera.
My old 250watt power supply choked when I tried to clock my 550 .25um Athlon to 600. So I bought a new one. And I went a bit overboard.

They my brother had need of a case with a decent power supply so I bought an InWin Q500N with a 300watt AMD approved supply and gave him that new supply and my old case. Using the nice new Q500N and my 450 watt power supply.

It's all good fun and it pisses off my roommate...apperntly my power supply fan is ungodly loud...I don't notice it hehehe...I'm to used to all my fans.

Anyhow like I said I can't show you a picture of my personal case...but I can show you some reviews of the case with pictures inside and out if you want...I should really get a camera..

 

[Dulanic]

See your ripping on DDR... when its not DDR that is at fault. We don't know if it is the chipset, or that the Athlon wasnt designed with that much memory bandwith in mind. However the facts stand....

DDR less cost than RDRAM (Not this second because there is so little supply and so little demand, which will change once the DDR boards hit the market)

Dual Channel DDR (Nvidia Crush chipset) = 4.2GB/sec
Dual Channel RDRAM 800 =3.2 GB/sec

Rambus has higher latency than DDR.


So the fact stands DDR is better than RDRAM.... it may just be the implementation of DDR that is holding back the Athlon with DDR.

We can make a better statement once we see a DDR P4 platform, since the P4 is designed with the higher bandwith in mind. I guarantee dual channel DDR on a P4 platform will kill a dual channel Rambus setup, and in a few months be cheaper too.

 

[JohnGalt]

because without DDR, Nvidia sucks?
doh!

 

[Sohcan]

It looks like Nvidia's Crush has been postponed for a while...

"Most Taiwanese mainboard manufacturers, including ASUS, were going to announce their mainboards based on NVIDIA Crush in H1 2001. However, NVIDIA told its customers that they wouldn?t get any Crush samples in 2000. Now the first Crush samples are expected to become available only in Q1 2001. Therefore, the first mainboards based on it may appear in the market only in Q2 at the earliest. Besides, NVIDIA is going to sell its first chipsets to OEMs mostly and not to the mainboard-makers, which affected the roadmaps of some mainboard manufacturers"

 

[Shuxclams]

Well DDR is priced competitivly w/SDRAM and does show some, albiet not what was expected, improvement which RDRAM has not shown. Pretty clear to me, cost to performance makes DDR a better choice.




SHUX

 

[Imaginer]

Well then.

I am sure glad that I got the KT133 based K7T Pro 2A. Sure this stuff is slightly faster and the DDR is a 10% increase but I am sure I would not mind...

...with that 1 Ghz T-bird of mine. Perhaps I should get one more stick of 128mb along with it.

 

[Noriaki]

Dulanic: You make a good case, but you forget one thing:

Dual Channel DDR (Nvidia Crush chipset) = 4.2GB/sec = 128 data pins (plus power, ground, and control)
Dual Channel RDRAM 800 =3.2 GB/sec = 32 data pins (plus power, ground, and control)

I'm reasonably sure that dual channel rambus doens't need 96 pins more than dual channel DDR for power, ground and control, so Rambus probably needs far fewer pins which can concievably save on motherboard cost. Though that might not have any effect to..

It will be interesting to see though.

 

[Dulanic]

Sohcan my point wasnt really related to Crush.... My point was directly related to Dual Channel DDR vs. Dual Channel RDRAM... And the winner is clear.

 

[Dulanic]

Noriaki... That may be true but remeber what NVidia is after.... the OEM market, thats why they are going with a all in one MB.... NIC, Sound, Video everything.... so if it was THAT much more to produce they wouldnt do it. Id expect maybe a $10 premium on the Dual Channel Crush over the Single Channel. Even with that premium it will probably still be cheaper with the DDR and Dual Channel then it would be with RDRAM. Top that off with an extra 1GB of bandwith.

 

[Noriaki]

Fair point.
The OEM market is definately cost sensitive and nVidia wouldn't be targetting that market with a $300 mobo.

 

[NOX]

<< See your ripping on DDR... when its not DDR that is at fault. We don't know if it is the chipset, or that the Athlon wasnt designed with that much memory bandwith in mind. However the facts stand.... >>

I hope you understand what you have just said, because what you have said supports the entire point of this thread.

We didn?t know back then if it was the i820 or RDRAM etc, which was performing poorly, and lacking the support needed to show its full potential. As you take a look at it now with more recent benchmarks the Intel 850, RDRAM and P4 perform very well in memory intense applications almost doubling DDR in Linpack, and more then double DDR in MBP. Though we have only experienced this after the Intel 850, why we didn?t see this from the i820 is simple, lacked support.

<< So the fact stands DDR is better than RDRAM.... it may just be the implementation of DDR that is holding back the Athlon with DDR. >>

Doesn?t this also apply to RDRAM, implementation at this point sucks, you're creating a double standard IMO.

And you shouldn't guarantee anything, or state anything as fact, unless you can prove it without uncertainty of future activities.

[EDIT]

Anyone visit Ace's lately?

 

[sandorski]

Well, look at it this way: When RDram was introduced as a new standard, it was 10x the cost of the previos standard(SDram).

Now DDRram is being introduced as a standard, it costs what, maybe 2x. Add to that that the cost of DDR is about the same as SDram before it got rediculously cheap and really there is no reason to equate the the introduction of RDram and DDRSDram.

 

[Dulanic]

NOX not at all..... I posted solid facts for their top situation. I will repeat it again since you totally took what I said out of context.

This is the BEST both can do...

Dual Channel RDRAM 800= 3.2GB/sec
Dual Channel DDR PC2100=4.2GB/sec

Add in DDR is lower latency.... so once again my facts are solid... DDR beats RDRAM.
What you just did was try to compare chipsets and CPUs... Im TOTALLY dismissing the chipsets and CPUs and showing what both types of RAM are capable of.... DDR beats RDRAM.

 

[NOX]

<< there is no reason to equate the the introduction of RDram and DDRSDram. >>

No there really isn?t, but there is a reason to equate the lack support from both Intel and AMD.

 

[NFS4]

256MB PC2100 DDR SDRAM = $379
256MB PC133 CAS2 SDRAM (Crucial) = $179
256MB PC133 CAS3 SDRAM = $100

-from current Anandtech Price Guide.

OOOOOOOOOOOOOOOOOOOOOOUUUUUUUUUUUUUUUUCH!!!!!!!!!!!!!!

I doubt that we will actually see any real performance benifits of DDR SDRAM until wayyyyyyy after Palomino. And I think that it is plausible to compare it to RDRAM. Even now, RDRAM is still only showing better performance in synthetic benchmarks.

Until applications start taking advantage of all of this extra bandwidth, DDR is a waster of $$$ IMH0 and I will stick with PC133 until the performance differential is much greater.

 

[backWERD]

NSF4: I feel like I got sucked in too.:|


Good thing I got my DDR sdram at a good price $113/128mb sticks


I still spent quite a bit on the board though $169

I wonder how much the new turbo boards will be compared to their older alternates

 

[NOX]

<< NOX not at all..... I posted solid facts for their top situation. I will repeat it again since you totally took what I said out of context.

This is the BEST both can do...

Dual Channel RDRAM 800= 3.2GB/sec
Dual Channel DDR PC2100=4.2GB/sec

Add in DDR is lower latency.... so once again my facts are solid... DDR beats RDRAM.
What you just did was try to compare chipsets and CPUs... Im TOTALLY dismissing the chipsets and CPUs and showing what both types of RAM are capable of.... DDR beats RDRAM. >>

You are doing nothing more but showing numbers, which are not evident of current performance. You want me to post a review for you to look at, and then you tell me that DDR is for a fact better? I?m not suggesting RDRAM is better either, but you seem to think that 4.2GB/sec is the actual performance currently experienced by the DDRAM platform, and that is way it's better. That is totally not the case or not relevant as we?ve seen SDRAM perform as well or on par with current DDRAM. In memory intense apps, RDRAM doubles that of DDR, and latency seems to not be an issue.

Dulanic, you said:

<< when its not DDR that is at fault. We don't know if it is the chipset, or that the Athlon wasnt designed with that much memory bandwith in mind. However the facts stand.... >>

So much for ?dismissing the chipsets and CPU?s! Like I said, that?s the point of this thread really, and Anands review.

You can speculate, or give your opinion all you want, but neither you or I know what?s really problem. I think it?s the lack of support, but I don?t know that for a ?fact?.

 

[IBMer]

Dual Channel RDRAM 800= 3.2GB/sec
Dual Channel DDR PC2100=4.2GB/sec


You seem to forget that Dual Channel DDR needs way over 3times the traces of the a Dual Rambus needs.

This is very expensive.

That is why you don't see it being used. I seriously doubt NVidia can make it cost effective.

 

[Noriaki]

<<
This is the BEST both can do...

Dual Channel RDRAM 800= 3.2GB/sec
Dual Channel DDR PC2100=4.2GB/sec
>>



I have to aruge that...you are comparing nVidia's Dual Channel DDR Crush chipset to the i850 which is current. The i820 had the i840 what's to say intel isn't developing an &quot;i870&quot; that uses quad channel Rambus? Then you'd get 6.4GB/s sure you'd need to put in RIMMs in quadruples but still it could be done. And it will still only need half the data pins of dual channel DDR.

I'm not saying your wrong DDR might kick arse, especially for the P4 which seems to be designed around high bandwidth, all I'm saying is you should wait and see how it performs before you say which will be better.

Several people (myself included, by implication if not saying it directly) have declared that the AMD760 would propel the Athlon to bury both the KT133 and anything at all from Intel...we are obviously eating our words now yes? I'm just saying you should wait until you see the performance of the nVidia Crush before you decide that it will work over the i850.

 

[zippy]

Having only read NFS4's post, here is my response:

There is a distinct difference, for one, AMD isn't promoting DDR SDRAM every chance it can get. In fact, I doubt the average user has ever heard of DDR SDRAM (except for maybe their GeForce or Radeon). Another difference is the price; 128MB PC2100 goes for about $150 on average (this is NOT the lowest price on pricewatch or anything), 128MB of RDRAM debuted at what, $1000? LOL! One final difference is that RDRAM never had ANY lead over PC133, yet the AMD 760 has a 5% lead in general usage- it's not much, but it is more than 0% and under!

Now, I hope you all don't think that I am very pleased with the AMD 760, because I'm NOT! I was quite dissappointed to say the least. It isn't all it should be, but really, that isn't AMD's fault, it's just that it has got ahead of it's time due to the competition between AMD and Intel. If you look at benchmarks from applications that are very taxing on the bandwidth, the AMD 760 does take a noticeable lead- however, the general public won't be using these very taxing programs day in and day out for a year or two.

Right now, the smart money is on a Thunderbird, Via KT133A, and 256MB of high quality CAS2 PC133.

 

[Windogg]

I shot my mouth off way too much during the &quot;First Memory Crusades&quot; and decide to sit back and lick my wounds. I now say this about DDR, the same thing I said about RDRAM, prove to me the performance benefit exists is cost effective manner.

In a previous thread I was ripped new on because I suggested that in the inital months, DDR would cost as much as it's equivilant in RDRAM. Many people suggested that the premium on release would be less than 25% with some people going as low as 5%. Have to do some digging but I know it's there somewhere.

Windogg

 

[Vegito]

I just think that there isn't backlash on ddr because the new mobo still can use SDRAM and the price isn't gonna cost you a arm and leg....

 

[NOX]

<< I just think that there isn't backlash on ddr because the new mobo still can use SDRAM >>

Not unless it has the 184 pin, and why would anyone do that.

 

[OneEng]

For todays programs the KT133A and CAS2 PC133 along with an Athlon processor are the best money can buy in the majority of applications. NFS4, I agree that the hype did not unfold into reality; however, there are some signifigant differences in DDR and RDRAM.

RDRAM is fundamentally inferior to parallel RAM of any kind in nearly every respect and in nearly every application. Before I go into the details of why this is true, I must first point out that it is the i850 chipset and the P4 cache arangement and memory pre-fetch that is making up for most of the deficiencies in RDRAM and thus covering up what would be simply dismal performance from the memory.

First
CPU's get the majority of their information out of cache running on die at processor speeds with very low latency in clock cycles (4-20) from the L1 and L2 respectively. Only when there is a miss in the cache system is it necessary for the system memory to be hit. These kinds of memory subsystem access represent relatively small amounts of data transfer, but at a moderately high rate of requests. Since in modern processors the micro runs at 1GHz + on a 32 bit bus while the memory runs much much slower rate, the latency of those hits are enormous. RAMBUS RDRAM has insane latency since it is a serial protocol RAM. It will never be the equal of parallel RAM for this reason alone.

[Second]
RDRAM must run at much much higher clock frequencies since it transfers so little information per clock. Higher clock frequency parts will ALWAYS be more expensive to produce, run hotter, and require more stringent main board designs in order to work than lower frequency parts.

[Third]
RDRAM is proprietary. I guarantee we would all be paying through the nose for RAMBUS memory if they could find any angle to make us do so. It is in everones best interest for the industry to adopt non-proprietary standards. Intel and RAMBUS would have us tied to a proprietary standard to prevent competition so that they could make better margins without the threat of better performing systems from other manufacturers.

DDR is not yet that much an advantage over PC133. This much is absoultely true. Given the choice between RAMBUS and DDR/QDR/etc, I would take DDR any day.

As an aside, P4 may not perform well with DDR since it has been designed to minimize the impact of high latency memory access. DDR is currently less capable of providing higher bandwidth than multi-channel RAMBUS. Since P4 was designed to shine under high bandwidth situations, I suspect that P4 + DDR is not much of a winning combination although it might curtail some of the poor benchmarks the P4 has shown to some degree.

The big advantage of the P4 is its bandwidth and high latency memory access work-arounds. RAMBUS RDRAM has been, is currently, and always will be an inferior memory methodology. Only in applications where high bandwidth and large amounts of continous memory streams are needed does RDRAM shine in the PC industry. In other hardware markets there are some applications that benefit from RDRAM's low pin count.

I wouldn't bet any money on RDRAM making any come-backs. DDR will likely be the next standard with higher and higher frequencies with higher and higher mult-pumping as time goes on to raise the bandwidth and lower the latency.