RDRAM vs. DDR Article

Toggle sidebar Toggle sidebar
A

ath0mps0

Distinguished
Feb 16, 2002
579
0
18,980
I can't believe it!!!!!

Does Frank Völkel actually believe that the political issues with RAMBUS are over? The issue with P4 and DDR is not that RDRAM is faster. A stick of DDR-266 already is 33% faster with lower latency than a stick of PC-800 RDRAM. DDR-333 is faster still. The true issue is that RDRAM has the benefit of the requirement of Dual-channel. The i850 chipset REQUIRES it. The i845D does not offer it. The problem is that the i845 does not allow this speed to be doubled as the i850 requires - RDRAM on the i850: 2 * 1600MBps = 3200MBps while DDR on the i845D: 1 * 2133MBps = 2133MBps.

I wonder if this problem wasn't actually one of the limitations RAMBUS imposed on Intel during renegotiations to allow Intel to introduce DDR with the P4.

Until the chipset manufacturers actually create dual-channel DDR chipsets to complement the RDRAM equivalents, the P4 will always be subject to memory bandwidth limitations with DDR. If this isn't a political issue, what is?

Alan Thompson
 
Your right, a dual channel DDR memory system should leave the dual channel RDRAM solution in the dust.
Click to expand...
Sure, but a dual channel DDR memory system is harder to implement, and will be harder still as the memory clock speeds go up.

"Ignorance is bliss, but I tend to get screwed over."
 
Ah, but if nVidia could do it, Intel could. The only reason nVidia's results were sub par is that they were using it on a system bus that can't take advantage of the aditional bandwidth.

What's the frequency, Kenneth?
 
It has more to do with the total width of the memory system, rather than the number of channels. A single DDR channel is 64 bits wide. A dual channel RDRAM system is 32 bits wide (16 X 2). Therefore, we can conclude that RDRAM is hindered by the narrower bus. It's actually DDR that has the "benefit" here since its operating on a bus that's twice as wide. Interestingly, Intel is moving away from the dual-channel design later this year with a soon-to-be-available 32-bit RIMM.

A 128-bit DDR system is possible, but not practical. The whole premise of DDR was to provide a cost-effective alternative to RDRAM. Going to 128 bits would defeat that purpose by adding cost to the mainboard. It would elevate the cost of a DDR system above the cost of an RDRAM system. Just look at the price of the Nforce boards compared to the others.
 
Actually, dural, you are only half right. While RDRAM is only 16bits wide, it is running at 800MHz. Like a long stroke 2 cylinder, superhigh rev'ing engine. This is one of the reasons that it was so hard to get PC800 RIMMs early on. There were huge yeild problems; most of it came out as PC600 or the less used PC700. Early on there was even PC 400. If you do the math, 800MHz X 2Bytes (16 bit) = 1600MBps. The speed ramping issues are exactly why Intel is being forced to re-widen the bus to 32 bits.

DDR on the other hand runs at comparatively low RPMs with higher actually speed; like a short stroke 8 cylinder engine that rev's "normally." 266Mhz * 8Bytes (64 bits) = 2133MBps. Because of this wider, shorter bus, DDR doesn't have the same latency issues that RDRAM has.

RDRAM added huge initial costs to the motherboard as well as RAM costs; the first available P4 mobos were well over US$200. If you look at comparative pricing for the nForce chipset and, say, the VIA KT266A, you will find that the motherboard manufacturers are price gouging, the nForche chipset costs very little more and requires even less circuitry. The additional cost can only be negligibly attributed to dual-channel memory; it must be attributed to GeForce 2MX Video, 6 channel Audio, LAN, etc.

Stop spreading the FUD. On the "difficulty of building" scale, I'd say they're about equal. DDR has always been as fast or faster than the fastest RDRAM (yes even DDR-200/PC1600 - it is the same speed as PC800 RDRAM). Even the not-yet-released PC1066 RDRAM is only as fast as DDR-266 (2133MBps). The limitation is political only; Intel doesn't want DDR P4 systems to compete on an level playing field with RDRAM.
 
Speaking of dual channel...since nFoce is dual channel doesn't it prove the concept , DDR can indeed be dual channel? So it's all a matter of the chipsetmakers designing dual channel DDR chipsets as someone mentioned.

Mark-

When all else fails, throw your computer out the window!!!
 
One other point here...this article is almost diametrically opposed to a THG article from May of 2000 according to the Inquirer.

Mark-

When all else fails, throw your computer out the window!!!
 
That's what I just said. Some people used the small size of the performance increase on the nForce to try to proove that Dual-Channnel DDR doesn't work like it's supposed to, but have proven nothing, because the memory bus can only handle as much information as the FSB can pass through it. So the P4 would be the only processor right this minute that could proove the effectiveness of dual channel DDR. Makes you wonder why Intel and nVidia haven't come to an agreement?

What's the frequency, Kenneth?
 
You stated your reason why you believe Intel hasn't released a dual-channel DDR chipset. What about VIA, SiS, Ati and ALi?

If DDR scales as easily as RDRAM, why do we need DDR2? DDR2 is expected to be released next year; RDRAM won't have a successor until about 2007.
 
I would venture a guess that Insmell :wink: has signed a revised contract with RAMBUTT :tongue: that while allowing them to produce DDR chipsets for the P4 after 2001, only allowed SDR before 2002 and doesn't allow dual-channel (until...at all?). This again promotes the false performance "lead" of RDRAM. I would imagine that these contracts also flow downhill to the Intel licensees (SiS, ATI, ALi, etc.). ATI - the only DDR dual-channel chipset manu - won't even state whether or not they have a P4 chipset in the works. This only leaves VIA. How many people have (respectively) purchased VIA P4 based mobos? With all of the legal FUD that Intel is spraying, VIA is already having a hard enough time selling its P4 chipsets to go investing additional design money in dual-channel. If VIA were to introduce a stable, high-performance P4 chipset with dual-channel DDR, I would be one of the first in line to purchase such a mobo.

I didn't say that DDR scales as well at all; I simply stated the fact that DDR is already performing at 33% (DDR-266) and 66% (DDR-333) higher than the current highest bandwidth RDRAM (PC800). I also stated that RDRAM manus are having difficulty getting decent yields out of the upcoming PC1066.

DDR manus are being forced to push the speed of DDR for high-bandwidth video purposes. I don't see RDRAM in this space since the outdated N64 - 1997?. This seems to be a piece of direct evidence that DDR outperforms RDRAM. DDR2 seems to be the next logical step since all of the available DDR mobos for the P4 use single-channel asynchronous buses. Not only does the asynchronous conversion eat up significant bandwidth, but the single-channel speed kills it.
 
I cant believe it either ..

this is a 180, with skid burns to prove it...
as allready stated
-nVidia has gotten 4.2G/sec out of dual chan PC2100 allready..(twin bank, sue me..) and imagine a P4 in one of those babies.. that TODAYS technology, not rambus of tomorrow.
-Via's PX4 is more or less on par with i850 .. and thats with 2.1 div 3.2 the bandwidth
-Intel is working on a dual channel version of the Brookdale.(and yea, we KNOW it can be done)

Articles like this makes me wonder who signs the paycheck here... or maybe we just ran out of exciting hardware news for the moment, so we had to spin this to get a story of some sort...

its sad.

and to top it all of, the stunt rambus played on jedec and the rest of the industry(or tried to) should get em decapitated right there... they're not playing with a full deck of cards... in more than one aspect.
 
You are so right in every aspect man (although 66% more than PC800MHZ, that I have a hard time beleiving. I'd say 33% is more fair) but the comparisons you are doing in those RAM is the same dilemma we have with the P4's high speeds low performance and the AXP's low speed high perf! When you get down to it it's the same story and since you favor DDR, AMD is also to favor here logically speaking. But then many people tend to ignore MHZ and see what is the strongest regardless of price performance, like the NW 1.6A OCed to 2.5GHZ. Ya it's big numbers and it does better than AXP 2000, and pretty much many ignore that big number and focus on the raw performance it gives. I personally still look at numbers and at 2.5GHZ it is not impressive as so. Try telling that to the CPU forum, that should really be deep!

--
For the first time, Hookers are hooked on Phonics!!
 
although 66% more than PC800MHZ, that I have a hard time beleiving. I'd say 33% is more fair
Click to expand...
Do the math. It's a fact. DDR333 (333MHz * 8 bytes per cycle = 2666MBps) vs. PC800 (800MHz * 2 bytes per cycle = 1600MBps) 2666 / 1600 = 1.66 or 66% greater. Yes, these are theoretical limits, but the benchmarks show that the both maintain about the same percentage of the theoretical maximum.

<A HREF="http://www6.tomshardware.com/cpu/02q1/020128/octitans-14.html" target="_new">http://www6.tomshardware.com/cpu/02q1/020128/octitans-14.html</A>

Look at the memory benchmarks.

2952/3777MBps, synch dual-channel, RDRAM, FSB@473, P4@2600MHz = 78% synchronous usage

2088/2900MBps, asynch single-channel, DDR, FSB@545, O4@3000MHz = 72% asnychronous usage (asynch losing up to 10% in the chipset).

like the NW 1.6A OCed to 2.5GHZ. Ya it's big numbers and it does better than AXP 2000, and pretty much many ignore that big number and focus on the raw performance it gives
Click to expand...
I always go for the actual total performance of the system, not just the highest clock. I have to agree that the 1.6A is tempting price/performance, but I am stocked with equipment for the time being. Keep in mind, though, that when the Tbred comes out this advantage will most likely disappear and people that purchased this combo will again end up behind the curve (like with everything).

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.
 
Okay, now I am scared! Toms Hardware is endorsing RDRAM? Seems like they've conveniently forgotten all the "dirty pool" that Rambus employed. Honestly, it sounds to me like someone's been bought. I mean we know that Intel would never hamsting a chipset to make Rambus look good, so we should trust that they made every effort to make the 845D chipset run at it's prime. Heh, yeah right!


-AF_Sting
 
Dural, you are exactly right. The important thing isn't channels, but the real estate used on a motherboard. A dual-channel RDRAM platform uses exactly half the space on the motherboard as a single-channel DDR-SDRAM platform. They would need to add more layers to the PCB to implement a dual-channel DDR-SDRAM platform, which would drive up costs.

So let me ask you: for a $250 motherboard, would you rather have (true) dual-channel DDR, or oct-channel RDRAM? At current clockspeeds (assume PC800 and PC2100), the PC2100 would provide 3.2GB/s of bandwidth. The PC800, however, would (theoretically, of course), give you 12.8GB/s of bandwidth. That's two different memory technologies using the same amount of available space on the motherboard.

Or would you rather have motherboards that are twice the size of todays, to cut down on costs?

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
 
While you are, of course, right about the real estate issue, nForce proves that it can be done for under $200 and not increase the size of the board (and add decent video, awesome sound, LAN, et. al.).

Hey, check it out: from stranger to journeyman in 4 days!!! I've got to get a life! This was my first post to Tom's (though I've been an avid reader for years).

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.
 
Of course, my example was merely to point out what you could have using the same space on the motherboard.

The nForce does pack quite a bit for the money, but the video is nowhere near good enough for most enthusiasts. That's the problem. The other problem is that the bandwidth numbers are not that much higher than KT266a or SiS735 numbers. If the nForce was released in June like it was supposed to be, it would've been awesome. As it stands, it's been outshadowed by the KT266a, which more people are comfortable buying (and you can get for slightly cheaper).

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
 
I just recently spent much time discussing this issue in a private technology conference. We agree with your points, you have obviously researched this issue in detail.

Could I perhaps Steal you away from your current employment. :tongue:

Enoch

there are absolutely no absolutes
 
Yeah? What conference was this, and who do you work for? (Just curious)

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft
 
Both the new XEON chipsets (iE7500 and SeverWorks GC) use dual channel DDR.

The ServerWorks chipset goes a step farther and allows you to set up your memory as you want. You can mirror the two channels and hot swap failed modules, you can stripe them for maximum performance, or you can "JBOD" them.

The Nforce isn't really dual channel DDR, it uses the standard 64bit channel for the processor, and provides another 64bit channel for the AGP. While this keeps AGP data transfers from bottlenecking the system, it isn't really dual channel.

- JW
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom