Unstoppable: DDR400 vs. Rambus

[Degenerate]

<< Yeah, exactley. I still hate Intel for canceling Tulloch which was going to be introing 32-bit RIMM's. Hopefully SiS and Rambus will launch a 32-bit Dual Channel RDRAM chipset. Now that would ROCK!!!! >>



Hell yeah! I really want to see RDRAM improve in speed more rapidly. Hurry everybody.

 

[BDSM]

Well now ppl.. The 850E is here. Without official pc1066 support. Told u so..

 

[Athlon4all]

Ok, well you were right. But, still board makers are saying PC1066 support, and thats all thatr matters.

 

[BDSM]

We'll just have to wait and see when the boards come out.

 

[BDSM]

Hello again Athlon4all & co!

I think it's very unlikely that board makers will go aginst Intel on this and release boards that use the chipset out of spec. Maybe Abit and a few others will but the vast majority won't. And Dell, Compaq, Hp etc. etc will most certainly be using boards running in spec!

For the overclockers it's a different thing.. But the 850E offers nothign new since this could be done with some 850 boards as well.

Further.. the 850E lacks USB 2.0! Clearly intel is going for DDR now.. And the answer to that is in my opinion Dual channel DDR 333!

Vr-Zone talks about "dual bank DDR 333" and also mentions the Springale chipset.. supporting 600 mhz fsb. Like I said before..
Dual channel DDR 333 will outperform pc1066 and pc1200 by a wide margin. It will basically kick ass with such huge bandwith ( 5.4 GB/s.. compared to 4.8 GB/s for pc1200 which I seriously doubt will be available this autumn) . And no matter what some ppl here say I still think DDR kicks RDRAM's ass latency wise.. All in all it's going to perform much better than RDRAM. And by this autumn DDR 333 will be mainstream and become really cheap.

DDR wins, rambust loses.. I am right, you're wrong

 

[Bozo Galora]

you dont have to wait: IWILL says it will.

 

[BDSM]

Hello Mr. Bozo! Well that is no sursprise at all. Given that for instance the Iwill Xp 333 athlon board supported /5 and /6 dividers some time ago. It makes it possible to run the Athlon out of spec. Like I said.. Some manufacturers will do this. But they will not constitute the buld of sales.

 

[Zakule]

I think people misunderstand my position since I happen to be playing devil's advocate (against RDRAM) in this thread. In the past, I've been a ardent RDRAM supporter and (as I stated earlier in this thread) the P4 was designed for use with RDRAM. While it suffers from both latency issues (yes, still, inspite of the benchmarks that are arguing against this) and heat issues, the idea behind the technology gives it inherently greater scalability then DDR. What I'm saying is that I don't think Rambus will get the chance to correct its issues in spite of the newfound support it is getting recently.

DDR technology (SDRAM technology, really) is going to hit a roadblock much sooner than RDRAM will, but DDR is also generating a lot more revenue for businesses than RDRAM and will continue to do so for at least another 4 months while more RDRAM chipsets are in development. There's just too many more DDR solutions than RDRAM solutions at this time. Couple that with the fact that DDR currently has lower latency and is the better performer for the majority of real world apps still in use, what conclusion would you draw?

Those of you who don't believe me take a trip over to www.fraps.com. Download the framerate utility. Now play some of the more current titles on an RDRAM system and look at your framerate consistency. Okay, now play those same games on a DDR system. Lastly, look back over all of your posts on this thread and try not to turn too dark a shade of red.

I'd love to see RDRAM still around a year from now in its RIMM 4200 form, but it just doesn't look like that's what is going to happen. If it does, I'll be the first to eat my words and I'll even enjoy doing it.

Okay, that's it for me on this thread. The ignorant vultures can now pick me apart to their heart's content.

 

[Bozo Galora]

you have good info, why do you have to ruin it with gratuitous insults?

 

[dexvx]

BDSM:

I recently built an Asus p4b-266C (i845 DDR platform). It has LOCKED AGP/PCI bus speeds. The reason why I know is because on an Asus p4t-e at 140Mhz FSB (70 Mhz AGP), my Radeon 8500LE refuses to work. However, on the Asus p4b-266C (140Mhz FSB) the Radeon WILL work. I am assuming that all the newer intel boards have locked AGP/PCI bus speeds once it gets past a certain level (which the tolerance of it seems very low).

From what it looks like you're very biased about DDR/Rambus. What you say is highly speculative. For all you know, there could be Dual Channel 32bit Rambus at 1200 Mhz, which will mean a 9.6 Gb/sec, which is almost 2x that of DDR 333 dual channel.

 

[Athlon4all]

<< I think it's very unlikely that board makers will go aginst Intel on this and release boards that use the chipset out of spec. Maybe Abit and a few others will but the vast majority won't. And Dell, Compaq, Hp etc. etc will most certainly be using boards running in spec! >>

I agree that Tier One OEM's won't use PC1066, but I guarantee you that the Falcon Northwest's and the like will, and so will board makers!!!

<< Further.. the 850E lacks USB 2.0! Clearly intel is going for DDR now.. And the answer to that is in my opinion Dual channel DDR 333! >>

True, but still, what's to keep board makers from putting the 845's ICH into 850e boards?

First off, that is the first Ive heard of 600fsb. That is cool!!!

Now about Springdale (aka Granite Bay). I will not question Dual Channel DDR's ability to be on par with a similar DC RDRAM setup (ie DC DDR266 vs. DC PC1066), However, I will not say that even DC DDR333 will "blow away" RDRAM. Here's why:

1. Latency is identical (I know what you're going to say, plz tell me where you get the claim that PC1066's latency is higher than DDR's)
2. Even though 5.4GB/ps is impressivwe, it won't do any better than PC1200 RDRAM except on nothing less but a 170MHz QDR fsb. History has shown that when the memory gets faster, it rarely gets a noticeable speed boost unless the fsb has equal bandy. Anand himself said itm the original 850's perfect fsb/mem bandwidth numbers are what made it perfect for the P4.
3. The cost of DC DDR. The pin count of DDR makes it very expensive to produce chipsets and boards for it. I feel that the cost of boards will put the overall price of the solution higher than 850e+PC1066. The cost of running the traces for 2 64-bit paths to the memory is just immense.

I do share that I doubt PC1200's hitting the market by fall, but what's to stop ocers?

Ive already asked this and I ask it again (and I ask Zakule this too). What makes you think that PC1066's latency is still high? Please give us your thoughts, no matter how crazy they are, please? I am curious to see how you come to this conclusion.
EDIT:

<< From what it looks like you're very biased about DDR/Rambus. What you say is highly speculative. For all you know, there could be Dual Channel 32bit Rambus at 1200 Mhz, which will mean a 9.6 Gb/sec, which is almost 2x that of DDR 333 dual channel. >>

True, but the reality is that it seems that it is going to be a long time before 32-bit hits the market sadly

<< While it suffers from both latency issues (yes, still, inspite of the benchmarks that are arguing against this) and heat issues >>

See above on latency, and as for the heat, I don't know what ppl make such a big deal about the heat, as long as it is kept under control, and it doesn't cost a fortune to keep it under control (and RDRAM doesn't cost a fortune).

<< DDR technology (SDRAM technology, really) is going to hit a roadblock much sooner than RDRAM will, but DDR is also generating a lot more revenue for businesses than RDRAM and will continue to do so for at least another 4 months while more RDRAM chipsets are in development. There's just too many more DDR solutions than RDRAM solutions at this time. Couple that with the fact that DDR currently has lower latency and is the better performer for the majority of real world apps still in use, what conclusion would you draw? >>

I agree totally that SDR is on its last legs, to be quite honest, I think you bring up an interesting point. I think that this is one place where RAMBUS the company needs to go. I cannot see how RDRAM is really that much more expensive to produce than DDR. Really, the RDRAM devices are similar to SDR, and the pin count is lower, so really, I come to the conclusion that its the Rambus royalties that are hurting it. That and also, the reason that there is so little support was the bad start it got off to. Will it recover, I dunno.

<< Those of you who don't believe me take a trip over to www.fraps.com. Download the framerate utility. Now play some of the more current titles on an RDRAM system and look at your framerate consistency. Okay, now play those same games on a DDR system. Lastly, look back over all of your posts on this thread and try not to turn too dark a shade of red. >>

To be honest, I dont have a P4 system to test, but that really sounds like a sythetic benchmark if you test say RTCW using its benchmark and then this and they tell 2 different things. Anyone with a P4 system care to try it?

<< I'd love to see RDRAM still around a year from now in its RIMM 4200 form, but it just doesn't look like that's what is going to happen. If it does, I'll be the first to eat my words and I'll even enjoy doing it >>

I agree, that it would be great if it hit the market, hopefully it will. We'll see.

 

[citrus3000]

I have been watching this thread for a long a time and you guys go back and forth etc. So this is the basic question I would ask myself. If u have 50 billion dollars...And u could get a 2.4 ghz chip clocked to about 3.2 ghz. with 2 gigs of ram etc. You could only hit this using rambus. Or would u rather settle for ddr? I would assume they would pick the faster system. (There might be peeps out their that say. well theirs that one program where ddr is faster. But thats one program) Then to fight rambus you say well what about ddr 2. Well what about it. I don't think I can go buy it so it has nothing to do which is better. Well then you say rambus is dead and ddr will take over. Were all smart but don't you think rambus and ddr have things that you havn't even heard of. Its like the AMD and intel fight. Amd has the hammer comming etc so intels dead. Intels not dead. Its all a politcal game. Thats it. But fact for fact. Rambus is faster then ddr with the current chipsets they use right now.

I forgot to mention the price wars. If you can afford rambus then get it. If you can't then get ddr. Its not rocket science.

 

[Chadder007]

Can't afford the RAMBUS?....well dang, its like only $20 higher in some cases for like 512 megs.

 

[BDSM]

Athlon4all.. Why I believe RAMBUS is high latency.. well.. I've seen many tests to prove this. I'll try to find them.

Try Anandtech faq first: link

try this one.. old but still : http://www.xbitlabs.com/mainboards/pentium4-chipsets/

The x-bit one only shows about 10% lower latency with DDR compared to RDRAM, however that was on a crappy px266 chipset.

I have to admit that I was probably thinking about latency comparisons of the Athlon vs. the p4 when I said RDRAM's latency is much worse. The athlon has much lower latency figures. However we don't know if this is due to the processor or the ram, maybe the 845 G will reveal this to us since it seems to have a highly optimized memory controller.

Another hint to RDRAM's high latency is how poorly it performed with the good old p3. Sdram seriously outperformed RDRAM back then.
Yes, the p3 couldn't make use of more than a max of 1.06 GB/s which was exactly what RAM could deliver. RDRAM could deliver this and more theoretically. Thus the p3 is the perfect example of RDRAM's high latency. It featured the same bandwidth, but performed much worse=much higher latency.

You may say that this may be due to a poor chipset. But there were many articles back then pointing out the high latency of RDRAM as the weak link.

Now.. there you have it.

 

[Athlon4all]

BSDM, the Anandtech FAQ is good stuff, and I am not disputing the fact that RDRAM does inheriantly have higher latency at PC800 and PC600 speeds. The benchmarks show that.

Now, I guess I did not communicate this clearly. I never once (and I doubt Fkloster did either) claimed that PC800 RDRAM has lower or similar latency to DDR266, I always said PC1066 RDRAM on the 533fsb platform has similar latency to DDR and that X-Bit Labs article only tested PC800+400fsb. Now that being said, since you clearly think that Cachemem is a vaild benchmark for latency, once again, I point you to Ace Hardware's PC1066 article from Sept. '01. He compares in Cachemem using a Williamette 1.6 and a hand ful of other benchys, PC800+400fsb (what X-Bit Labs did), PC800+533fsb (the configuration Anand used in the 2.53 533fsb review yesterday), and finally PC1066+533fsb, and for comparision he benchmarked 845 with PC133 as well. Now, I again am going to post the latency numbers:

PC600+400fsb: 307
PC800+400fsb: 270
PC800+533fsb: 247
PC1066+533fsb: 207
845+PC133+400fsb: 229

There you have it. Also notice how there was a significant drop in latency when going from PC600 to PC800, and again from PC800 to PC1066. I could dig up from Cachemem benchmarks if you need them.

<< Another hint to RDRAM's high latency is how poorly it performed with the good old p3. Sdram seriously outperformed RDRAM back then.
Yes, the p3 couldn't make use of more than a max of 1.06 GB/s which was exactly what RAM could deliver. RDRAM could deliver this and more theoretically. Thus the p3 is the perfect example of RDRAM's high latency. It featured the same bandwidth, but performed much worse=much higher latency. >>

Again, that is with PC800. Also, as the AT FAQ stated, 850 has features that hide the latency, and you gotta remember, 820 was Intel's very first RDRAM memory controller. So there's no doubt that RDRAM was not for the P3, but still, the P4's architechure was built around RDRAM and it hides some of the latency as well, plus like I said, 850 does as well.

Let me just restate something, I am not questioning the fact that RDRAM does inheriantly have in some cases higher latency that DDR, but as the clock speed increases, this latency decreases that is what I am trying to get across, that and that PC1066 has similar latency to DDR.

 

[fkloster]

<< Those of you who don't believe me take a trip over to www.fraps.com. Download the framerate utility. Now play some of the more current titles on an RDRAM system and look at your framerate consistency. Okay, now play those same games on a DDR system. Lastly, look back over all of your posts on this thread and try not to turn too dark a shade of red. >>



Ok Zakule....I'll bite...tonight I will download your ute and test the following titles and post my results tomorrow am. Could some one with P4 2.4 DDR system w/ti 4600 default clocks please do the same so we can compare ???? In addtion we must test the same three games...:

1) RTCW / all candy / 800x600x32
2) Dungeon Siege / all candy / 800x600x32
3) Max Payne / all candy / 800x600x32

What are the odds of our 20,000 member forum having the exact system and games as me but using i845 ddr?

 

[oldfart]

<< Could some one with P4 2.4 DDR system w/ti 4600 default clocks please do the same so we can compare ???? In addtion we must test the same three games...: >>


I have the P4 @ 2.4 DDR part, but wont have my 4600 till the end of the week. If nobody else runs the tests, I can do it maybe this weekend/early next week. I have RTCW and Max Payne, but not Dungeon Siege. How about Serious Sam or Serious Sam SE? I run 1.6A @ 2.4, DDR400. I'm curious to see the results.

Problem with running a 3D game related test is there are so may variables other than the CPU/mem system. To make it accurate as possible:

Disable sound card in device manager.
Run the Ti4600 @ stock speeds (we are testing CPU not GPU)
Ensure Ti4600 is configured the same (no tweaks other than Vsync)
Same OS (I run Win2K)
Same Ti4600 driver release
All games on same point release
Use same bench such as RTCW Checkpoint
Post a config file for testing (wolfconfig.cfg)

I'm up for it if you can wait a few days.

 

[fkloster]

Gotit. I post my results as soon as possible...oldfart post over weekend...then we put this ridiculous thread to rest.....funny how we have to do all the work to prove what should rest squarly on Zakules shoulders to prove... Oh I can't wait till Monday.....hope Zakule is still around

 

[oldfart]

I see you have 1024 meg of ram. Can you go to 512 instead?

 

[fkloster]

No. I won't remove any rimms. In support of Zakule's perverse logic, my 1 GB of rambus, because of its serial nature & high latency coupled with extreme heat issues, becomes even slower & would most likely 'hinder' my frame-rates more so than would 512mb so let the test continue.

 

[oldfart]

Hence the problem with such a test. Too many variables. For a like to like comparison, the systems should be configured as closely as possible. 1024 of ram is way above the norm, and twice the memory I have available for testing. How is this test supposed to compare 2 otherwise closely configured systems? Very few people run 1 Gig of ram. 512 meg is pretty common. I have no beef against anyone here, and fully expect the RDRAM system to come out on top, but unless it is a test with matched configurations, I'm not interested.

 

[fkloster]

*fkloster contemplates amputation of memory only in his zealous to bury Zakule and his non-sense...*

 

[Athlon4all]

I have to agree with oldfart about taking your Memory down to 512MB Fkloster, however if you wish, you could always yourself do a seperate test, one with 512MB and the other with 1GB, howz that sound?

EDIT:

<< *fkloster contemplates amputation of memory only in his zealous to bury Zakule and his non-sense...*
>>

Fkloster, did you even read the PM I sent you? Please

 

[fkloster]

*sigh* ...agreed... I will do two seperate tests.... one with 512mb / one with 1 GB

 

[Zakule]

<< you have good info, why do you have to ruin it with gratuitous insults? >>



"Gratiutous" implys that it is unjustified. Take a look at what Fkloster has posted. THOSE are gratuitous insults which warrant a like response, in my opinion.



<< To be honest, I dont have a P4 system to test, but that really sounds like a sythetic benchmark if you test say RTCW using its benchmark and then this and they tell 2 different things. Anyone with a P4 system care to try it? >>



FRAPS is not a synthetic benchmark. It takes reads on frames per second while in 3D apps and displays them in the upper left-hand corner. Not 100% accurate in all cases, but fairly reliable. Results here mean much more than 3DMark2001 since it allows you to test a variety of Directx8.0 engines, whereas 3DMark2001 is uses only ONE directx8 engine. It's a simple, but effective utility that even the greenest computer tech can use.



<< Gotit. I post my results as soon as possible...oldfart post over weekend...then we put this ridiculous thread to rest.....funny how we have to do all the work to prove what should rest squarly on Zakules shoulders to prove... >>



Man, where does this sense of entitlement come from? I have nothing to prove to anyone on this thread, least of all you, Fkloster. I'm offering information just as everyone else is. Whether or not it is received as credible or not doesn't matter to me and why should it? I've already figured this stuff out.

Okay, NOW I'm outta this thread (so hard to remove myself from a good debate).