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

[rmarango]

In conclusion I think that the move to DDR technology alleviates the memory bottleneck that, at present is the main limiting factor in system performance.
The current memory technology which is an evolution of the standard SDRAM will provide sufficient bandwidth to cover CPUs up to a processor speed of approximately 2 GHz.
Above that speed, memory latencies will become the main bottleneck in system performance.
Reducing latencies on the basis of address strobe speed appears at this point unreasonable.
However, uncoupling of the output from the DRAM array by means of an embedded SRAM row cache can hide latencies in an application dependent manner.


Greetings

 

[Dulanic]

<< 1. DDR has no real advantage over SDR at this present TIME (for a number of reasons, reasons which you don?t seem to understand or want to understand)!
2. I never said better DDR support was not obtainable.
3. I?ve stated from the START current support is lacking, and to give it time.
4. DDR is not better then RDRAM, not the way you are suggesting. Seeing Linkpack, MBP, and a few Q3 benchmarks, it is very simple to say RDRAM is better then DDR also. But to do so would be ill informed. >>



Man Ive posted all these things 10 times! Your first point is what I said before... It has no real advantage at this time with its current implmentation (Could be chipset, could be the bus, could be lots of things)

I also said its current support was lacking but you ignored me saying that everytime and thats why I kept going.

As for your 4th point this is where you need to WORD things better.... you need to say in its current AMD implementation DDR is no better than RDRAM... I would have agreed with that. But your saying DDR and that is very confusing because you could mean the DDR memory..... you could mean DDR in its current AMD implementation.... remember DDR just means Double Data Rate.... so just saying DDR can be very confusing as to what you really meant.

 

[Caitiff]

You know, it's going to be funny, but the biggest sector that is going to benefit from the DDR ram is going to be budget systems. Integrated video that using system RAM will have fairly impressive marks when using the low latency high bandwidth DDR ram. I'm fuzzy on the details, but Nvidia's new 'all in wonder' board may turn out to be a sleeper.....

 

[Bilow]

Dulanic--

Dude, nothing works &quot;in theory&quot;. It has to work in the real world. It's not like the KT133A only makes the AMD760 look bad, it makes the ALi &quot;Magik&quot; look even worse. There is something about using DDR as main system memory this is not adding up here. DDR is mostly a great thing for the DRAM industry--that's why Micron, Infineon, etc. have been so enthuiastic because it lets them use their current production base (largely). But the pin count is enormous, and there clearly is some latency (buffering?) inherent to the memory controller that isn't there on a video card.

Given enough time and a CPU designed for DDR from the ground up (the way the P-4 was designed for RDRAM from the ground up), I'm sure it would show a performance advantage. But will it be cost competitive, and will any motherboard makers still be investing in DDR chipsets if nobody wants to buy the first generation?

Intel had one enormous advantage when it introduced the i820 to ramp RDRAM production a year ahead of the P-4's launch--it had DELL willing to pre-order RDRAM and sell i820-based systems. Dell is STILL selling them. Micron just ain't Dell. I'll admit that I'm not really sure why the DDR systems available right now only perform on par with a KT133A--I'd like a technical explanation of what may be going on here.

 

[NOX]

<< you need to say in its current AMD implementation DDR is no better than RDRAM >>

Though you don?t know forsure if DDR will in fact be better with improved implementation. If you want to compare the two now, excluding chipsets, and CPU?s, RDRAM definitely out performs DDR in bandwidth, and that seems to be the issue. I believe Bilow said it best ?Look at ANY memory benchmark comparison between RDRAM on an i850 system and any DDR PC266 (aka, &quot;PC2100&quot, using STREAM, LINPACK, or Sanda Memory benchmark. If you take the marketing at face value, PC2100 should be able to perform within 65% of dual-RDRAM channels (2.1/3.2). It doesn't even come close--maybe 40%.

Will the next implementation of DDR gain 40% or even 20% on RDRAM, that?s yet to be seen. You are playing a guessing game, and as people have already stated they got burned the last time. Give it time buddy, will know soon enough.

 

[PlatinumGold]

Maybe this has been said. I've only read like 2 pages of this thread but I'll say it anyway.

There is a huge difference between RDRam a Proprietary Technology that Intel had a Financial stake in and DDR Ram an Open Technology that AMD has no financial stake in.

That is really the key difference. Both are superior technologies vs SDR Sdram. Both had the potential of being the next great memory technology. But Intel tried to force RDRam down our throats.

Does everyone else not see that? The fact that RDram underperformed was just more fuel to the fire.

So DDR Sdram has now proven to underperform, nothing unusual about that, has everyone already forgotten EDO Ram?? Memory technologies will always underperform (less than our expectations) because they are only one small component of overall performance. From the beginning the Strength of the Athlon CPU was its FSB. It wasn't as Anand stated as memory bandwidth dependent as say the P4 is. But even tho DDR Sdram has underperformed will it go away? no. within a couple of months everyone will be using it, AMD will provide us w/ CPU's that can take advantage of the bandwidth and this argument will be moot.

The main point is Proprietary w/ fincial stake by Intel BAD. VERY BAD.

Open Standard, no financial stake by AMD Good. Very GOOD.

 

[rmarango]

Good point, Platinum member.
Palomino is expected for the 1st quarter 2001 to fill this gap.

 

[Bilow]

PlatinumGold--

you wrote:

&quot;There is a huge difference between RDRam a Proprietary Technology that Intel had a Financial stake in and DDR Ram an Open Technology that AMD has no financial stake in.&quot;

The consumer doesn't know or care. There are TONS of proprietary standards in the current PC architecture. The P-6 bus (Intel). The EV6 bus (Compaq). The PCI specification (Intel). The AGP specification (Intel). Basic semiconductor function (Texas Inst.) Most manufacturers already pay a royalty of some sort to manufacture anything they make (name a component that contains no proprietary information). The thing is, quite of a few of them cross-license IP, so money may not change hands. But Rambus, Inc. is a pure IP company (a lot like Nvidia and Quaalcom, and I don't see them getting bashed for it).

Intel's financial stake is moot. They are not going to bet the future of the company on some stock play. In any case, Intel missed the milestones that would have resulted in their ownership of Rambus stock warrants. Under the original warran agreement, Intel needed to ship RDRAM memory interfaces on 75% of their shipping chipsets for four consecutive quarters. They didn't even come close. Intel's real financial stake is ensuring the success of the P-4 platform over the next 4-6 years. If that means that RDRAM makes price/performance sense (and right now it does) then you could say that Intel has a financial stake in the success of RDRAM. But Intel will shift away from RDRAM if it doesn't work for their bottom line. The warrants are not the issue anymore.

By the same token, AMD has a tremendous stake in DDR technology--they are betting that DDR-SDRAM will keep their performance edge over the P-4 as the Intel chip ramps in speed toward 4GHz. Do you think that AMD doesn't care about it's investment in the AMD760 chipset? Do you think that AMD has no stake in the success of the Sledgehammer family of CPUs and the AMD770 chipset, which are designed to use DDR (or RDRAM, if you believe some reports)?

That is really the key difference. Both are superior technologies vs SDR Sdram. Both had the potential of being the next great memory technology. But Intel tried to force RDRam down our throats.

Intel tried to force RDRAM into the market because they knew that it's success was crucial to the success of the P-4. Intel doesn't care about selling memory. It doesn't even care very much about selling chipsets. Intel is in businsess to sell CPU and other multiprocessors. Intel makes corelogic chipsets to sell more CPUs. Can you imagine how much RDRAM would cost if it did not have a 1-year ramp behind it? Instead of costing $150 for 128MB of PC800, it would cost 6-8 times that. And the P-4 would be DOA. So, did Intel try to get people to buy i820/RDRAM P-3 systems by only offering 133MHz FSB on the i820 (not the 440BX) so it could claim a performance advantage for RDRAM? You bet your ass they did--they needed to get OEMs to buy RDRAM to bring the cost down in time for the P-4.

Are AMD and Micron doing exactly the same thing right now to get a DDR-SDRAM ramp underway? You bet your ass they are. Notice that until the KT133A, if you wanted a synchronous 133MHz motherboard, you had to buy an AMD760 or ALi Magik board. And AMD (just as Intel did a year earlier) could claim that DDR made a 5-10% performance difference. Then, just as Tom Pabst showed that an overclocked 440BX board using 133MHz FSB outperformed RDRAM, Anandtech has shown than a KT133A matches the performance of a DDR system.

The real question we should be asking is: &quot;Whose side is VIA Technologies on?&quot; Launching the KT133A shows that VIA is not in the DDR camp. It's interesting to speculate that the settlement last summer of the P-6 bus patent infringement case brought by Intel against VIA may play a role. Details of the settlement have not been made public, but the two sides acknowledged that Intel reserved the right to sue VIA if it brought a new (DDR?)chipset to market for any AMD CPU. Notice how long it's taking VIA to bring it's Athlon DDR chipset to market? 6 months ago, everyone said they would be first, and VIA won't even say when to expect delivery.

Bottomline: this is war, and the industry is divided into sides. The war is over which x86 architecture will dominate the next decade--AMD's build around Hammer cores using DDR-SDRAM, or Intel's built around P-4 and McKinley cores, using RDRAM. The stuff about &quot;proprietary&quot; vs. &quot;open&quot; is a PR smokescreen. The industry is FULL of proprietary standards and IP issues. Nobody makes a semiconductor product without paying royalties--even EDO manufacturer involves an IP payment to TI.

 

[PlatinumGold]

Bilow

Ok, so you want me to remake all of the arguements that have already been made? Intel had a stake in Rambus. It didn't turn out because the technology didn't show itself cabable of sweeping us off our feet. Intel had motives other than just selling their CPU's. AMD's only incentive is to sell CPU's. they have no stake in DDR as a technology.

The consumer does care. the ones that post to this forum anyway. That's who we're talking about because &quot;Why isn't there a backlash against DDR SDRAM like there was with RDRAM?&quot; was the question posed as the title of this thread. Hence your arguement about the consumer is moot.

When RDRAM support from intel first came out there was definitely the feel that Intel had ulterior motives. Similar to when the Slot 1 came out.

&quot;Intel's financial stake is moot. They are not going to bet the future of the company on some stock play. In any case, Intel missed the milestones that would have resulted in their ownership of Rambus stock warrants. Under the original warran agreement, Intel needed to ship RDRAM memory interfaces on 75% of their shipping chipsets for four consecutive quarters.&quot;

It's not moot. You can honestly tell me that anyone knew that Intel would miss the milestones when this agreement was made?? Intel agreed to 75% because they felt they could do it.

&quot;By the same token, AMD has a tremendous stake in DDR technology--they are betting that DDR-SDRAM will keep their performance edge over the P-4 as the Intel chip ramps in speed toward 4GHz. Do you think that AMD doesn't care about it's investment in the AMD760 chipset? Do you think that AMD has no stake in the success of the Sledgehammer family of CPUs and the AMD770 chipset, which are designed to use DDR (or RDRAM, if you believe some reports)?&quot;

Not by the same token. Yes, DDR will help AMD but only in the sense of marketing. not in the sense of AMD having a STAKE in DDR Stock (no such thing).

&quot;Intel tried to force RDRAM into the market because they knew that it's success was crucial to the success of the P-4. Intel doesn't care about selling memory. It doesn't even care very much about selling chipsets. Intel is in businsess to sell CPU and other multiprocessors.&quot;

Wrong, anyone in business will tell you, any additional source of revenue adds to profitability. Why is intel still the #1 producer of chipsets for Intel Processors? why did they abandon motherboards. If what you say is true regarding just doing it to help sell CPU's why don't they still build Motherboards? because there was no profit in building motherboards and there is in building chipsets.

&quot;Are AMD and Micron doing exactly the same thing right now to get a DDR-SDRAM ramp underway? You bet your ass they are. Notice that until the KT133A, if you wanted a synchronous 133MHz motherboard, you had to buy an AMD760 or ALi Magik board.&quot;

How can you possibly compare Micron w/ Rambus? Will micron benefit from Samsung DDR Sdrams? will Rambus benefit from Samsungs RDRAM's? actually as it is right now, Rambus benefits from Samsungs DDR Sdram's.

&quot;The real question we should be asking is: &quot;Whose side is VIA Technologies on?&quot; Launching the KT133A shows that VIA is not in the DDR camp. &quot;

I'll grant you this is an INTERESTING QUESTION, but it isn't the real question. I think the real question was posted as the title of this string.

 

[Bilow]

PlatinumGold: So, you think that Intel doesn't make motherboards anymore? Care to explain this?:

http://www.intel.com/design/motherbd/index.htm?iid=prodinfo2+devmb&amp;

or the 16 pages of intel boards on pricewatch?

You seem like a bright person. Intel makes a profit on any CPU it can sell for more than about $75. It's average selling price for CPUs last quarter was just under $200. It's average selling price for corelogic chipsets was about $40. Intel makes a very marginal profit on chipsets. It makes a HUGE profit on CPUs. The reason Intel got into the chipset business in 1995 was to ensure the success of the P-5 and P-6 platforms and to make sure that the dominant technologies that defined those platforms would come from Intel. SDRAM was introduced by Intel. The PCI bus is from intel. AGP, USB, and more innovations that I could name are from Intel.

The key is defining the PLATFORM because that ensures more CPU sales. Why do you think AMD is reluctant to get into the chipset market? Because it's a marginally profitable line. Why has AMD ended up MAKING the AMD750 and AMD760 (and soon the AMD770) chipsets? Because the K-7 and K-8 PLATFORMS would be dead without them. Remember how long it took for the KX133 to be introduced by VIA? A YEAR after the Athlon launch. The Athlon is succeeding today because AMD bit the bullet and made the AMD750 chipset. If SMP Athlon succeeds, it will be because of the AMD760MP chipset. Know anyone else who is making an SMP chipset for Athlon?

Micron has a HUGE stake in DDR. It pushed the adoption by the industry since 2 years ago. It hired the marketing consultant Bert McComas (Inquist, Inc.) to generate favorable attention on DDR two years before the first chipset could be brought to market. Do you think it is an accident that the first attacks on Rambus at Tom's Hardware were written by Bert McComas?? Do you think that Jack Robertson's (ETONLINE and other CMP publications) attacks on RDRAM and Intel always quote Bert McComas and Semico's Sherry Garber??

Micron is the third largest DRAM manufacturer in the world. It's getting killed on SDRAM, selling every chip it makes for below cost; it's most profitable memory for the past two years has been EDO-SDRAM, a legacy memory. Micron desperately needs DDR-SDRAM to succeed; it's driven the JEDEC effort, and was the first company to counter-sue Rambus. Micron is betting the company on DDR-SDRAM and the lawsuits. Ironically, if Micron loses, the only DRAM it can make are EDO and RDRAM, since Micron is a Rambus licensee on RDRAM.

Intel stood to get 4 million shares of Rambus stock if it had met the production milestones. Even optimistically, assuming that the stock was worth $50 a share, it would have meant $200 million for Intel bottom line. This is peanuts. Intel's revenues this year will top $30 billion. Even five times that is peanuts if it also mean that Intel is losing market share to AMD and VIA. The whole stock thing was icing on the cake for Intel--they thought they would easily be able to achieve the targets.

You and I don't need to determine whether Rambus' claim to IP covering SDRAM and DDR are valid; courts are going to do that very soon now. But you should remember that the legal teams at Samsung, Hitachi, Toshiba, NEC, Oki, TSMC, and UMC looked hard at the patents and then signed agreements paying Rambus royalties for every SDR- and DDR-SDRAM chip they make.

As to why DDR hasn't yet generated a backlash? Simple; AMD is the underdog. Tying the name &quot;Intel&quot; to Rambus made it a target. AMD is in no position to &quot;shove&quot; DDR down anyone's throat, and neither is Micron. But both AMD and Micron has as much at stake on the success of their platforms (and DDR) as Intel ever had on RDRAM.

 

[PlatinumGold]

points conceded. I didn't do my research.

Bottom line the question was &quot;why isn't there a backlash against DDR Sdram like there was with RDRAM?&quot;

Answer is still &quot;The perception by most is that RDRAM was going to benefit Intel financially, not thru CPU's sold but thru RDRAM's sold.&quot;

The same perception doesn't exist w/ DDR Sdram and AMD.

 

[rmarango]

May I add my opinion : The KT133A should be seen like an intermediate step to the new AMD processors (supporting the FSB of 266 Mhz) that will benefit more than actual ones from DDR technology.
I shouldn't see the KT133A like an AMD760 killer, but as an evolution toward it, obviously projecting this in the near future, when the Palomino will be ready.

 

[backWERD]

Just to let you all know that all the benchmarks so far are from non retail boards all in the steps of being developed and the benches that anandtech is useing are from very early rev boards especially for the ali magik board. So as the own saying wise saying says &quot;dont count your chickens untill your eggs have hatched&quot;.

 

[NOX]

The Intel D850GB board which was used by Anand is a production board and available, the Soltek board, which was used by Anand is also available according to Solteks website. The iWill KA266, I have know idea if it?s available or not?

The iWill, and Intel D850GB board at X-bit labs

 

[backWERD]

Yes it is available I just received my Iwill ka266 tonight and I'm here to tell you all the board I have has alot better benchs than what is shown on the anandtech benchmarks (anand uses a very early board doesnt even have the multiplier jumpers) The board I have actually waxes the floor off of the AMD760 ref board. Like I said &quot;dont count your chickens till the eggs hatch&quot; words to live by.

 

[NOX]

<< Like I said &quot;dont count your chickens till the eggs hatch&quot; words to live by. >>

If you look at the link I posted that?s exactly right!

 

[backWERD]

that is correct and even those benchmarks are even lower than the retail rev that I have.

In anands benchs For this ali chipset are alot slower than any others on the net like 30 fps in quake 3 now that's quite a bit of difference the only explanation for this is these benchmarks actually come from an article ANANDTECH released on the ali magik on november 29th with the board was in its first or second rev. now its on its 5th or 6th and the speed has grown exponitialy by comparison and the board, It also wouldnt supprise me if the Linpack scores come from the same first rev board.

 

[backWERD]

Try these benches on for size

system

Iwill ka266r
1.2ghz @ 133fsb
elsa gladiac geforce 2 32mb
buffalo 256mb DDR pc2100
Sblive


sisoft sandra

memory scores for my system

594/806!!!!!

xbit labs reports
540/704

sharky extreme reports
553/699

Thats a difference of 54/102 thats quite a difference for comparable systems all due to (lower rev boards and non tweaked bios settings)

Then take a look at the diff between the the so called &quot;DDR killer&quot; and my (ali magik)

my scores (ali magik)
594/806


KT133A xbit reports
404/446

thats an astonishing difference of 190/360 thats =47%/80.7% alu/fpu


Now I have to agree the kt133a might be a amd760 killer, but I doght it will even come close to killing the ALI magik chipset.


but who knows maybe I'm &quot;counting my chickens before they hatch&quot;.

 

[NOX]

That?s definitely an improvement! I?ll post some Sandra benchmarks tomorrow from my brothers PC, right now I?m off to bed.

 

[backWERD]

me to good night

 

[rmarango]

That's great !
Please look my posts here Anand Forum on AMD760 killer

What I said is true, it's was just a matter of bios immaturity and rev. too!
Sorry Anand...