Unstoppable: DDR400 vs. Rambus

[burek]

Unstoppable: DDR400 vs. Rambus

 

[Bozo Galora]

nice find, great article.

woah! 3610 sisoft memory
Kloster - you lookin' here?

 

[oldfart]

Hmm i845 and DDR400. I'm running that already. I score over 3000 in SiSoft mem benches (see link).

 

[Athlon4all]

Hot dog!!! Look at 845G!!! Why on earth would Intel give 845G DDR333 support but not 845E? If anything, they should be more concerned about giving proper validation to 845G because it will be a huge product! However, this clearly shows that not even DDR400 can stand up to PC1066 RDRAM, much less PC1200. RDRAM keeps on clearly making itself the best memory for the Pentium 4.

 

[grant2]

Ya that faster RDRAM is sure taking out the trash. Johan @ aceshardware found similar results with 533mhz RDRAM a few months ago.

 

[ElFenix]

845G with PC3200 DDR could be whipping up on PC1066 RDRAM. too bad he couldn't test that.

 

[fkloster]

Memory Bandwidth for my P4 533fsb rig...:

P4 2.4ghz 533fsb PC-1066

Ram Int Buffered----3363 mb/s
Ram Float Buffered-3355 mb/s





<< woah! 3610 sisoft memory (for Intel DDR mobo with 400mhz ddr memory) >>



Bozo where was that @? I see 2600 not 3600!!



<< 845G with PC3200 DDR could be whipping up on PC1066 RDRAM. too bad he couldn't test that. >>



WTF? Elfenix?? PC3200 IS in the test

1) PC-1600 is 100mhz DDR
2) PC-2100 is 133mhz DDR
3) PC-2700 is 166mhz DDR &
4) PC-3200 is 200mhz DDR (i.e. DDR-400)....what are you talking about they using PC3200 DDR in the test and rambus still killing it

HELLO WHAT ARE YOU GUYS TALKING ABOUT???

 

[Terrapin]

Mind if the average Joe asks a basic question?

I understand the 533 boards and 533 CPUs won't be available until the end of May. Is that correct?

I also keep hearing about the 1066 RDRAM, but have yet to see it sold anywhere. Is this also a product I can expect to find at the end of May?

Thanks,

Terrapin

 

[oldfart]

I don't get the excitement on the 845G. It allows (officially I guess) a 133 FSB, 166 x 2 = 333DDR speed right? 150 FSB overclock = DDR375. Today, with the 845D, I can run 133 FSB, 3:4 mem ratio = 355 DDR, 150 FSB, DDR400 (PC3200). Am i missing something?



<< WTF? Elfenix?? PC3200 IS in the test >>


Not on an Intel chipset, only SiS, which is certainly lower performing than the Intel chipsets.

Memory Bandwidth for my P4 533fsb rig...:

P4 2.4ghz 533fsb PC3200 (DDR400)

Ram Int Buffered----3051 mb/s
Ram Float Buffered-3045 mb/s

Not PC1066 RDRAM speed, but not bad either.

 

[dexvx]

Terrapin - Most Samsung PC800 modules run at 533 DDR (1066 Mhz) RDram clock
Oldfart - according to tom, Intel boosted the overall performance of the 845's memory controllers

From the looks of this article, it looks like CAS latency is they key. I mean, 400DDR with CL3 gets beaten by 266DDR with CL2; 200DDR CL2 beats CL2.5 DDR266, even though raising the CAS latency (bumping the speed up) gives higher Sisoft Memory benchmark results, but in real world performance it gives *less* performance (sometimes by a WIDE margin). More than another reason why more clockspeed nowadays is irrelevant in overall performance - I dont get why most i845 DDR overclockers are raising their CAS latency to ridiculously high levels just so they get a high result on Sisoft Memory bench and a high Mhz p4 when lower CAS latency gives better realworld performance.

Based on Tom's article, PC800 RDram (2 year old technology) is still a match for the latest and greatest DDR SDram offerings (DDR400).

 

[Terrapin]

With all respect, you guys are so used to overclocking, sometimes you assume everyone does. I don't.

When people refer to faster Ram, I actually think they are saying you can purchase this ram, not overclock ram to meet those speeds.

I note that the Asus P4S533 is now available for purchase, and support DDR 2700 memory.

My plan is to wait until the end of this month, when Intel releases its P4 533mz CPU, and build my new PC around that Asus board, the DDR 2700 memory, and a P4 533mz bus CPU. Please tell me if I should be doing something else; and I am not interested in overclocking.

Thank you,

Terrapin

 

[oldfart]

<< Oldfart - according to tom, Intel boosted the overall performance of the 845's memory controllers >>


Yes, Intel now officially allows 533 FSB, 333 DDR speed on the 845G. My point: Unofficially, you can already run 533 FSB and 355DDR speed on (a good) i845D board. I don't see a benefit. I didn't see any DDR mem benches that were ag good as mine.

 

[MrDudeMan]

wow rdram cleaned ddr off the table

 

[BD231]

wow rdram cleaned ddr off the table

I think you ment cleand the table with DDR, and it damn well did, and all it had to do is move up to 133mhz FSB. DDR wont go much further in it's current state, DDR 400 must cost an insane amount of money, and considering there isn't a decent chipset out there to utilize it, it's a pointless buy. Next step has to be dual channel, but even then RDRam will be the better choice due to the increase in interface from 16 to 32bit and the fact that it wont need to be run in pairs. The only thing that sucks is that we probably wont see a AMD Hammer using RDRam, we'll be stuck with ddr :disgust:

 

[Bozo Galora]

kloster:

you misunderstood:

on last page they show 533 bus rdram doing 3600+ with 850E chipset with its better architecture
what i meant was if you upgrade to this board you would get 300 more marks in SiSoft - and sleep
better at night.

 

[MadRat]

I am willing to bet HAMMER will be dependent on higher speed memory now that the FSB/64-bit L2 pathway bottleneck is eliminated. Hard to believe we'll likely see a RDRAM chipset for AMD in 2003.

 

[BFG10K]

I wonder where all of those chaps that were proclaiming the death of DDR have gotten to? Perhaps they should look at this article and re-evaluate their stance.

Either way, the Pentium4 is certainly looking more and more attractive every day.

 

[Zakule]

<< From the looks of this article, it looks like CAS latency is they key. I mean, 400DDR with CL3 gets beaten by 266DDR with CL2; 200DDR CL2 beats CL2.5 DDR266, even though raising the CAS latency (bumping the speed up) gives higher Sisoft Memory benchmark results, but in real world performance it gives *less* performance (sometimes by a WIDE margin). >>



Dexvx, you've posted the most relevent fact in this discussion. Applications only use a fraction of the memory bandwidth provided by both DDR and RDRAM. Low latency memory is what matters most for 3D graphics rendering on today's systems and it is one area where RDRAM falls short. Unless they can improve that part of the technology RDRAM is going to disappear, slowly but surely. It's a shame, because the design is really better suited to work with P4's processors than DDR memory, but the latency issue will undoubtedly diminish its performance in the next generation of apps.

 

[Neurofreeze]

<<

<< From the looks of this article, it looks like CAS latency is they key. I mean, 400DDR with CL3 gets beaten by 266DDR with CL2; 200DDR CL2 beats CL2.5 DDR266, even though raising the CAS latency (bumping the speed up) gives higher Sisoft Memory benchmark results, but in real world performance it gives *less* performance (sometimes by a WIDE margin). >>



Dexvx, you've posted the most relevent fact in this discussion. Applications only use a fraction of the memory bandwidth provided by both DDR and RDRAM. Low latency memory is what matters most for 3D graphics rendering on today's systems and it is one area where RDRAM falls short. Unless they can improve that part of the technology RDRAM is going to disappear, slowly but surely. It's a shame, because the design is really better suited to work with P4's processors than DDR memory, but the latency issue will undoubtedly diminish its performance in the next generation of apps. >>



Not true, as RDRAM and DDR SDRAM scale up, RDRAM's latency (in real-time, not the clockcycle penalty) will be lower than DDR's because of its higher scaling architecture. I can't remember the numbers off hand, but RD won't surpass DDR a while, not until PC1200 RDRAM, although PC1066 is VERY close (but that little boost combined with its already high bandwidth is enough to clean house). So really, the more time passes, the better for RDRAM. DDR's only hope is DDR-II. Well... that and vastly superior marketing and industry support (even RDRAM's big guns, Intel, is waning).

 

[Degenerate]

What excatly is preventing the release of faster RDRAM?

 

[AkumaX]

<< What excatly is preventing the release of faster RDRAM? >>



marketing

 

[joohang]

<<

<< What excatly is preventing the release of faster RDRAM? >>



marketing >>


Lawsuits?

 

[Acanthus]

samsung already has PC1200 modules out, its a matter of luck to get one though, i ordered 2 samsung 800s yesterday, hoping to get 32ns versions (32ns=pc1200, 35ns=pc1066, 45ns=pc800) I know of at least one person with 32ns chips (no i dont know tom)

 

[fkloster]

<< << What excatly is preventing the release of faster RDRAM? >> >>



Intel. thats the only reason. As soon as Intel feels the need for more speed, the appropriate rdram module is already for production.

 

[Degenerate]

Well... then, the thing that will make intel feel it is competition. Then again, why woudn't Intel increase demand for these better RAM. It will give their P4's a better performance. Perhaps, the tech is not there yet for full scale production.