Dual channel DDR for Pentium 4

i don;t know y anybody hasn;t notice this ..

http://www.gamepc.com/reviews/hard [...] cssid=&tp=

Although this is the first chip for workstation but looking at the result i m wandering y intel dumped RDRAM for dual channel DDR... they r equal in performance and as normal user don;t need anything above 512MB so there isn;t any limitation on the memory... and also RDRAM is almost equal in price and easy to implement in dual channel configuration and intel has already couple of experience so they could very well have made implemented dual channel RDRAM 1200 a lot better... atleast they shouldn;t have dumped it so quickly..

just my opinion

First of all, using DDR PC2100 and Dual Channeling negates the whole point of this competition. They need to up it to DDR333, in order to get 5.4GB of bandwidth to be higher than the current RDRAM at 4.2GB, which is the PC2100 Dual Channel too. Thus, until we see DDR333 benchs or DDR400, Dual Channel DDR is nothing but latency ranting.

--
Thunderbirds in wintertime, Northwoods in summertime!

Keep in mind the price difference - PC2100 DDR is about the same price as PC800 RDRAM. PC2700 DDR is significantly less expensive than PC1066 RDRAM. So, for PC800 prices a PC2100 DDR platform will perform on par (or better if you're "latency ranting" ) with a PC1066 RDRAM platform. I illustrate (prices are estimates including shipping):

All systems are dual channel hypothetical comparisons:

PC800 RDRAM (3200MBps) - 2 x 256MB = $150
PC1600 DDR (3200MBps) - 2 x 256MB = $120 (if you can find it)

PC1066 RDRAM (4267MBps) - 2 x 256MB = $300? (prices currently not announced)
PC2100 DDR (4267MBps) - 2 x 256MB = $150
PC2700 DDR (5333MBps) - 2 x 256MB = $160

Even if the the dual-channel DDR mobos are $20-40 more due to the additional traces, the advantage is still with DDR as far as price goes. Plus, higher capacity DDR DIMM are available which means that you can put more memory in your DDR based system.

If you dispute my estimates for PC1066, please provide links as to where you can purchase it for less - I would be happy to revise. If your links show higher prices it just solidifies my point.

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.

Where did you get $300 for PC1066? It seemed like a rather random figure, but you appear to be saying you'll back it up.

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft

Doesn't anyone remember the i840 fiasco? OK, the deal is, the i840 was supposed to replace the BX and GX in high end boards. But the BX and GX were able to significantly outperform the i840 simply because SDRAM has less latency than RDRAM. Period.

I changed my SDRAM Cas Latency from 3 to 2 on my board and picked up as much as 7% in my performance. RDRAM has what, in real time comparisons, Cas5?

We ALREADY see the SiS DDR333 chipset boards, using memory that has INFERIOR bandwidth to that of dual channel RDRAM, matching the current RDRAM boards in performance. Don't you think that using Dual Channel DDR, even at par to RDRAM for bandwidth, will offer even better results?

PC800 RDRAM has the same bandwidth as lowly PC1600 DDR SDRAM. And PC2100 has the same bandwidth as PC1066. Simply using dual channels for PC1600 takes it to PC3200 bandwidth. RDRAM is also at 3200MB/s using dual channels. As is the Northwood "A". So as Eden points out, the only difference is latency. And latency makes ALL the difference!


What's the frequency, Kenneth?

<blockquote><font size=1>Svar på:</font><hr><p>Who said Intel dumped RDRAM? They've got new RDRAM chipsets on the roadmap.<p><hr></blockquote><p>Maybe, but it seems that the Intel 850E chipset will only suppport PC800. I'm not saying that it can't run PC1066, as we already know that a 850 with TI DRCG's can do that, but I was hoping that 850E would make PC1066 the new starting level with PC1200 within overclock reach.

I haven't seen any PC1066 for sale yet, and what about RIMM 4200 modules built on 0.12 micron ? I've heard a lot about it, but there is no evidence that is around the corner at all.

There will only be a couple of manufacturers supporting PC1066, Samsung and Kingston, so we can only expect that PC1066 won't become mainstream and thus it will likely be expensive.

Now the only option I have is to wait for SIS to come and rescue me with a dual-channel DDR266 chipset, which will match the 533 MHz FSB of Northwood B perfectly, and still have plenty of headroom for overclock.


<i>/Copenhagen</i>

Effectively RAMBUS and Intel together has not been giving RDRAM proper attention. The 850 and PC800 is not enough for 533 MHz FSB, Intel should have been supporting PC1066 with 850E, but they're not. I mean Northwood B is just around the corner and we need official PC1066 support now.

Maybe we'll see the goods in time, but there's surely no evidence it will happen. On the other hand SIS has announced a dual-channel RDRAM PC1066 chipset called SIS658 which is also able to use RDRAM modules with 32bit bus, but that is only due in Q3 2002. Before that they will announce the a dual-channel DDR333, the SIS655. I think both chipsets ought to have been made available together with Northwood B, but I guess you're right about me being impatient.


<i>/Copenhagen</i>

Oh, I agree with you, crashman, but you have to remember that DDR-SDRAM has higher latency than SDR-SDRAM (not as high as RDRAM, but higher still). Yes, dual-channel DDR (correctly implemented) for the P4 would almost definately outperform dual-channel RDRAM. This has been my arguement since my very first post here on THGC. The point I was making is that not only would it do this, it would do it at lower cost for the forseeable future. Even when the P4 requires a 533MHz (4267), 600Mhz (4800MB/s) or even 800Mhz (6400MB/s) FSB dual-channel DDR is available TODAY and will cost less than currently unavailable competing RDRAM platforms.

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.

That was one of the points I used to push hard-back when the P4 was NEW, PC2100 was like half the price of RDRAM, AND dual channel PC2100 would have been overkill for overclockers! Plus it could have been utilized using the techonolgy of that day, so basically they've kept us waiting for what, 18 months?

What's the frequency, Kenneth?

Who said Intel dumped RDRAM? They've got new RDRAM chipsets on the roadmap

tullosh as been flush.So i will buy a I850 board and keep as long as i can.[-peep-] DDR fan it suck do you read the spec sheet.

cheap, cheap. Think cheap, and you'll always be cheap.AMD version of semi conducteur industrie

Was that directed at me? I never said anything about which was better, merely relayed a fact about the latency of ddr in relation to sdram.

The Cash Left In My Pocket,The BEST Benchmark

What's your answer to this?
<A HREF="http://freespace.virgin.net/m.warner/Roadmap_now.htm#Tehama E" target="_new">http://freespace.virgin.net/m.warner/Roadmap_now.htm#Tehama E</A>

<font color=blue> There's no such thing as hell, but you can make it if you try.</font color=blue>

In reply to:
--------------------------------------------------------------------------------

First of all, using DDR PC2100 and Dual Channeling negates the whole point of this competition. They need to up it to DDR333, in order to get 5.4GB of bandwidth to be higher than the current RDRAM at 4.2GB, which is the PC2100 Dual Channel too. Thus, until we see DDR333 benchs or DDR400, Dual Channel DDR is nothing but latency ranting.



--------------------------------------------------------------------------------

Keep in mind the price difference - PC2100 DDR is about the same price as PC800 RDRAM. PC2700 DDR is significantly less expensive than PC1066 RDRAM. So, for PC800 prices a PC2100 DDR platform will perform on par (or better if you're "latency ranting" ) with a PC1066 RDRAM platform. I illustrate (prices are estimates including shipping):

All systems are dual channel hypothetical comparisons:

PC800 RDRAM (3200MBps) - 2 x 256MB = $160
PC1600 DDR (3200MBps) - 2 x 256MB = $90 (if you can find it)

PC1066 RDRAM (4267MBps) - 2 x 256MB = $525? (prices just announced by <A HREF="http://www.theinquirer.net/22050203.htm" target="_new">Kingston</A> )
PC2100 DDR (4267MBps) - 2 x 256MB = $90
PC2700 DDR (5333MBps) - 2 x 256MB = $130

Even if the the dual-channel DDR mobos are $20-40 more due to the additional traces, the advantage is still with DDR as far as price goes. Plus, higher capacity DDR DIMM are available which means that you can put more memory in your DDR based system.

<b>If you dispute my prices for PC1066, please provide links as to where you can purchase it for less - I would be happy to revise. If your links show higher prices it just solidifies my point.</b>

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.


If the thought I thought I thought had been the thought I thought, I wouldn't have thought so much.

agreed, although I think EDO RAM is better

Intel, VIA, and SiS already have DDR chipsets that outperform PC800 RDRAM. Imagine with dual-channel DDR RAM, RDRAM will become history! Besides, DDR RAM prices are going down faster than RDRAM prices. Where I live, Samsung PC2100 DDR RAM costs $65CDN while Samsung PC800 costs $127.

<b><i>"A penny saved is a penny earned!"</i></b>

<A HREF="http://shop1.outpost.com/product/3308112/" target="_new">128MB of Mushkin PC1066 for $70</A>
<A HREF="http://www.memplus.com/Merchant2/merchant.mv?Screen=PROD&Store_Code=A&Product_Code=880538&Category_Code=MEMPC1066RAMBUS" target="_new">256MB of Samsung PC800A for $103</A>

Kingston PC1066 is on Pricewatch, but I couldn't find it at the individual stores, so I couldn't post any links.

EDIT:
Though I did see them at the same place before, and the prices corresponded with what I gave above. Maybe a little higher.

<font color=blue>Hi mom!</font color=blue><P ID="edit"><FONT SIZE=-1><EM>Edited by FatBurger on 05/22/02 09:11 AM.</EM></FONT></P>

Fatburger,

The PC1066 you listed was ony 128MB and comes from Fry's - I was just at the store yesterday and they don't actually have it in stock - notice that the site says "call for availability."

The 256MB PC800A (533MHz) Samsung you listed is 40ns. i850E requires 32ns or better (less is better, of course).

The Kingston 32ns PC1066 (RIMM2100) was just announced at prices of $263 per stick - the actual price may be lower, but not by more than $50.

If the thought I thought I thought had been the thought I thought, I wouldn't have thought so much.

Outpost is not the same thing as Fry's, they have different stock and different specials (I found that out the hard way).

EDIT:
And as for the i850E and memory, technically it only supports PC800.
From page 9 of the datasheet:

<font color=blue>Hi mom!</font color=blue><P ID="edit"><FONT SIZE=-1><EM>Edited by FatBurger on 05/22/02 12:15 PM.</EM></FONT></P>

Don't worry, I won't mention that it should be 256 instead of 254 and "dual" instead of "duel"

Dual DDR333/PC2700 would be nice. RDRAM has the potential to leave SDRAM in the dust, but Rambus seems to have a hard time dragging their fat asses around to get anything done. It's been far too long between PC800 and PC1066.

<font color=blue>Hi mom!</font color=blue>

Yeah, it's almost gonna be 2 years since the PC800.
I wonder if it's worth paying for the 1066 to get them to 1200...

--
I can't beleive Dungeon Siege has a pitchfork weapon called "Hoe"!

Intel hasn't dropped RDRAM from the desktop, the i850E is new, and there are more in store, unless they've all been axed and nobody told me.

As for dual vs. single:

DDR memory is harder to do, but a dual-channel motherboard is harder to do. I would think that a dual-channel motherboard with SDR would be harder than a single-channel motherboard with DDR, however.

Yes, that's one of the major reasons for the i850 designed the way it is. It hides RDRAM's latency.

In theory, they would be the same. But I would think that it'd be easier to make an efficient DDR memory controller than to make an efficient dual-channel memory controller. I have nothing to back that up since I've never really thought about it before, that's just off the top of my head.

<font color=blue>Hi mom!</font color=blue>

The single channel ddr 333 has less latency than the DUAL channel rdram 1066.

Thats exactly what I would expect. Watch what happens when dual channel ddr systems start shipping.

The Cash Left In My Pocket,The BEST Benchmark

I think it wasn't so much that Rambus became complacent as that Rambus was too busy putting their money into court battles than they were into R&D. Plus, they had one hell of a time actually getting memory manufacturers to produce RDRAM. (Not that I blame the memory manufacturers <i>any</i>, what with the way Rambus was behaving.)

The technology had high potential. It went mostly wasted though. Not too long from now (a year or two) I expect SDRAM and RDRAM to both be replaced by a newly emerging memory technology. It's about time for one.

That would be because (from what I've read) DC DDR is damn hard to implement with stability on a motherboard level.

Here is what understanding I have gotten through articles on the subjects:

Dual-Channel is litterally that. It requires twice as many data paths. DDR SDRAM has 64-bit data paths. DC DDR would require 128-bit data paths. RDRAM has 16-bit data paths. DC RDRAM (which is what is commonly used) would require 32-bit data paths.

Now, RDRAM also just moved to a standard 32-bit data path, which makes one single 32-bit RDRAM RIMM equivalent to a DC 16-bit RDRAM configuration. To my knowledge, there is no DC 32-bit RDRAM platform yet.

But in any event, my point is that for 16-bit RDRAM to be DC, they only needed 32-bit data paths. Where as for DDR SDRAM to go to DC, they needed 128-bit data paths. As you could imagine, a 128-bit data path is a <i>lot</i> harder to develop than a 32-bit data path. This is why DC DDR SDRAM is so rare still.

Further, to my knowledge, the problems with getting DC DDR was that the 128-bit data paths often incurred small fluctuations of electrical noise into the memory signals. This of course would make such platforms unstable, and even if error-correction was used to compensate would end up making them much slower than they should have been. So DC DDR is taking so long simply because these noise problems have to be thoroughly worked out.

Now, onto increasing the data rate. Again, to my knowledge, single-rate to double-rate required that the memory latency be reduced so that it could handle interfacing at twice the rate. So to move from double-rate to quad-rate, would again require a significant latency reduction in the memory. As you can imagine, that is no easier of a task than getting 128-bit data paths to work without noise issues.

This is why DDR SDRAM has taken so long to really improve. It's a technology that is already being pushed to it's limits. This is also where RDRAM had a <i>lot</i> of potential. A DC 32-bit RDRAM platform (64-bit data paths) would have a heck of a lot of bandwidth. And since DDR is already a 64-bit data path, it should be a relative piece of cake to develop a DC 32-bit RDRAM platform.

However, will anyone bother? No one knows...


Tech support said take a screen shot.
Putting it down with my .22 was the humane thing to do.

In theory, this will make SledgeHammer one hell of a performer, at least in regards to memory latency. The combination of dual-channel DDR SDRAM with an on-die DDR SDRAM memory controller should negate a very large amount of the memory latency. It will also give one heck of a nice bandwidth, which is needed for multiple CPUs.

However, it may also be <i>very</i> difficult to implement on the motherboard level. Noise, noise, noise! Expect any motherboards with this to be either darn expensive, or to be very unstable. Or maybe even both. The chances of them being neither is <i>very</i> low though.


Tech support said take a screen shot.
Putting it down with my .22 was the humane thing to do.

That's what DDR2 should help to aleviate. Idealy, it should also help aleviate some extra heat (albeit not too much).

Other than that, has there been any rumors about any other emerging memory technology?

"Search your feelings you know it to be true, I am your... twin sister" - Darth Vader