Unstoppable: DDR400 vs. Rambus

[fkloster]

lol andreasl is right It does look very much like the original author is discussing hdd access and not system memory access...

 

[Zakule]

<< Your vapor sources again, tsk tsk. Then you conveniently claim to have known before it was all over the news sites... >>



Sheesh

Tons of people knew before it was all over the news sites, you don't have to be James Bond to get your hands on that type of information. Look at one of my earlier posts (dated BEFORE the tech sites were fully advocating this position) where I clearly state that even though mainboard manufacturers might support PC1066 with the i850e that Intel would not "necessarily" support it. At the time of that posting, I was almost positive this was the case but felt it was inappropriate to make that claim until I'd confirmed it (hence the "necessarily" verbage). As it turns out, I wasn't able to confirm it until after it was posted on tech sites, so it was a moot point. Again, if all you're going to do is criticize me without offering anything useful yourself the least you can do is read my earlier posts.



<< Like the other guy said... around here at Anand's, it's all BS until you can prove it. >>



The only way to prove anything with technology is to perform the tests yourself. Mainstream tech websites hold very little weight with me due to their obvious bias and unspoken affiliations with manufacturers, and I've already stated that in my earlier posts. The fact that many people in this thread worship maximum bandwidth measurements and maximum framerates over memory latency and minimum framerates speaks volumes on how little they understand about what's important for performance in real apps vs. synthetic benchmarks.



<< Don't take this the wrong way Zakule, You say 645DX, I have just one question, are you going to then discredit Oldfart's 845-D DDR400 numbers? Or should we find somebody else to do the DDR testing to compared to Fkloster's RDRAM numbers? I wouldn't bet anything just yet, but in any case, I look forward to seeing the numbers >>



Hehe, wouldn't dream of taking it the wrong way Athlon4all. I have to say that I admire you for remaining civil inspite of all of the mud slinging going on right now. I'm not discrediting Oldfart's tests, quite the opposite. I've run those same tests before and the i850 outperformed i845D, no question. However, when I compared i850 against the SiS645 I saw that the SiS chipset not only outperfomed the i850 by a slim margin but it outperformed the i845D by a quite bit (relatively speaking). So you see it's not that I'm discrediting Old Fart's test it's just that I know what to expect and in no way do I disagree that the i850 is a better performing than the i845D. What I'm saying is that it is not DDR that is to blame it is Intel's own DDR chipset.



<< I too think that Intel would be very capable of crippling their DDR chipset to make Rambus look better. Capitalism makes strange bedfellows >>



I have very good reason to believe that they did just that. After the above mentioned test I did a little digging and my findings were.....interesting. No, I don't have a link to provide proof for this. What some of you are failing to acknowledge is that it is not always prudent to divulge one's sources. With respect to those individuals whom I receive occasional inside information from, I would not want to put them in a difficult position of having to explain why they told me something that they shouldn't have.



<< Hey madrat...you been around here a long time...for the life of me I can't remember his username, but do you remember the guy who was @ anands for a while who 'said' he ran an engineering firm in Sweden and owned sports cars and ripped on people ect. ect. ect.? >>



First off, I did not start us down this path of flaming each other, that was you Fkloster. Don't get me wrong, I enjoy a spirited debate as much as the next fellow, but trying to paint this sort of picture of me is just pathetic considering that some others on this thread have dedicated entire posts to ripping on me. I don't know about the rest of you, but I'm trying to have fun while I'm doing this. My responses are as playful as possible since I'm trying to keep my focus on RDRAM vs DDR while fending off assaults from several off you at once. Many of you are doing nothing more than trying to find something bad to say about me at this point, which is fine with me, but I would hope for your sake that you're at least aware of this before you get firmly seated on your high-horse.

Some of you are sitting there talking about how "outlandish" my statements are mainly because you don't agree with them. I could say that it is "outlandish" to put so much emphasis on 3DMark2001 since it isn't a very effect utility for benchmarking your system, but I'm not taking that posture. I'm not implying that any of you are stupid for thinking what you do; I'd say that most of you have very good reasons for believing what you do and I respect your various opinions. The continued implication that I'm just inventing what I'm saying is ridiculously narrow-minded and a feeble retort at best. Simply because I don't quote the very benchmarks that I debunk you think I'm fabricating my statements? This, to me, could also be "outlandish," let's say. The more objective of you have realized that the only way to be sure whether or not what I'm saying is true is to personally perform benchmarks yourselves. It is to these people that I will continue to debate. Those of you who are just whining and taking shots at me, I will ignore you after this post, but please continue with your barrage of adolescent slandering. If nothing else, its entertained me (and others, I'm sure) a great deal.



<< I too have read about the database file system nature of the next file system coming from MS. However a statement like this shows a big lack of understanding about most things computer related... A disk file system is the way the data or organized and ordered on the harddrive. A harddrive has an access time several orders of magnitude longer than RAM. Consider a typical harddrive with an access time of 10ms (simplified, most have a little less these days). RAM has access times of about 100ns (also simplified, and I'm including chipset latencies as well). 10 milliseconds is 10.000.000 nanoseconds. So by the time the CPU waits for the harddrive access it can access RAM about 100.000 times! The difference in access time between different RAM types is very insignificant in this sense (about 10-20ns or so). >>



You make an excellent point, and I think that what is partially lacking here is the context from which I took this quote on NT6. Even within context, the paragraph is poorly worded insofar as it trys to illustrate why RDRAM will have a negative effect on a database filesystem. As I'm sure you know, there is more to managing a file system within an OS than the harddisk itself. However, I am now relunctant to divulge the identity of somebody whom I have a great deal of respect and admiration for since a few of the people in this thread seem to have difficulty controlling themselves. Perhaps, when the fire calms down and the children have left then I will post the source of this information so that you and others can benefit from the full dissertation. There is most definitely a problem with NT6's new filesystem and RDRAM, but I agree that the blurb I posted doesn't adequately discuss why and by itself it sounds like a misrepresentation. My bad on that one.

 

[smp]

You make an excellent point, and I think that what is partially lacking here is the context from which I took this quote on NT6. Even within context, the paragraph is poorly worded insofar as it trys to illustrate why RDRAM will have negative effect on a database filesystem. As I'm sure you know, there is more to managing a file system within an OS than the harddisk itself. However, I am now relunctant to divulge the identity of somebody whom I have a great deal of respect and admiration for since a few of the people in this thread seem to have difficulty controlling themselves. Perhaps, when the fire calms down and the children have left then I will post the source of this information so that you and others can benefit from the full dissertation. There is most definitely a problem with NT6's new filesystem and RDRAM, but I agree that the blurb I posted doesn't adequately discuss why and by itself it sounds like a misrepresentation. My bad on that one.

I admire your skilled use of the english language, but I have to say that I can't buy that argument. If you could post that one paragraph, you can post the article. I'm interested in this and I'de like to know. I'm not against you, I actually find your argument interesting. There is more to the tech industry (and any business) than numbers, market share and consumer interest. Where does consumer interest come from anyways? There are ideological structures that exist that are transparent to us consumers that deserve observation and discussion (which we won't find on hardware forums unfortunately).

 

[Zakule]

<< I admire your skilled use of the english language, but I have to say that I can't buy that argument. If you could post that one paragraph, you can post the article. I'm interested in this and I'de like to know. I'm not against you, I actually find your argument interesting. There is more to the tech industry (and any business) than numbers, market share and consumer interest. Where does consumer interest come from anyways? There are ideological structures that exist that are transparent to us consumers that deserve observation and discussion (which we won't find on hardware forums unfortunately). >>



Well spoken, sir. Your skepticism is understandable given that I have not provided sources of the information that I've obtained from 3rd parties (since I am the source of most of the information I've posted in the thread in terms of performance benchmarks, I can't really back up my own position with myself, if you take my meaning).

The individual whom posted the blurb is a member of another tech forum, and in my overzealousness to bring new information to the table I.....urrr.....didn't fully consider whether or not he'd appreciate my quoting him. I've learned a lot from this man and I don't want to alienate him so I will converse with him a bit about this and find out if he minds becoming pariah number two on this thread .

What I can tell you about him:

-He was an employee of Microsoft until somewhat recently (can't remember when he left, exactly).
-He was testing i850 chipsets with RDRAM two years before their initial release.
-He was part of the Windows Millenium Edition development team (and, yes, even he admits that Windows ME sucks)
-He was part of the Windows XP development team (initially).
-Based on his experience in helping engineer Windows, I consider him an expert on the Windows Kernel
-Based on the above, if he says that he believes RDRAM will have problems with the next generation NT file system then I'm inclined to believe him even if his initial explanation of why is somewhat lacking.

 

[smp]

Bring him here!
We're all friendly

 

[smp]

Come on folks!!!!

Where's those benchys? I'm all eager and sh!t, lets go!!!!

 

[MadRat]

<<The individual whom posted the blurb is a member of another tech forum, and in my overzealousness to bring new information to the table I.....urrr.....didn't fully consider whether or not he'd appreciate my quoting him. I've learned a lot from this man and I don't want to alienate him so I will converse with him a bit about this and find out if he minds becoming pariah number two on this thread >>

If he has such brilliant credentials then he'd garner some respect. Unlike you, he'd actually know what he was saying. The only thing anyone holds against you Zakule is your prima donna personality. Get off the high horse and prove your assertions.

 

[oldfart]

Jeez! The pressure! Still waiting for the Ti4600 to get here. Fedex tracking says "Est. Delivery Date 05/10/2002".

fkloster, what benches are we using besides RTCW? Is there a bench for Max Payne?

Looking @ real time frame rates is useless. Any tests have to be actual benckmark demos. RTCW Checkpoint, Q3 Demo FOUR, UT Thunder and/or UTBench, SS Karnak, SS SE Grand Cathedral, etc. I dont have Dungeon Siege. This has to be a controlled test. Since you have yours setup already, some settings and some things I need to know:

Ti4600 Driver version?
No tweaks at all on the Ti4600! Just disable Vsync and thats it. Stock 300/650 clock speed.
Refresh rate fix installed
I assume DX 8.1
I run Win2k SP2..you??
Each game that is benchmarked, what point release?
AGP aperature size?
Test run with soundcard disabled in device manager.
The config file for each benched game must be exactly the same. You can email me your config files and I'll use them.

How about a non 3D game bench?

Thats all I can think of at the moment

Peace!

 

[smp]

 

[MadRat]

oldfart, just have fun with it. We don't need no stinki' professionalism. We're all just fanboys at heart.

 

[Oreo]

With FRAPS you can output the fps numbers to a log file which could then probably be converted into a graph of some sort which would be ideal for this comparison. You do however have to run the exact same benchmark run for it to work and be comparable.

 

[Morph]

BDSM said:
First of all: Dual channel pc1200 does NOT give 9.6 gig/sec in it's present configuration, but rather 4.8 gig/sec! You are talking about dual channel 32 bit RDRAM which is nowhere near to the horizon.

Wrong. dual-channel 32-bit RDRAM is very much on the horizon. Read this excerpt from the following article: x-bit article
"As we know so far, i850E based mainboards will be introduced only by Intel of course, ASUS, Gigabyte, QDI and maybe also MSI. By the way, ASUS is going to implement one more very interesting feature on its i850E based P4T533 mainboard: the support of 32bit RDRAM memory modules. In fact, 32bit memory modules are none other but a pair of regular 16bit memory modules combined in a single package. This way, you no longer need to use Rambus memory modules in pairs. As there are two 16bit channels implemented in a single 32bit module. Memory modules of the kind, from Kingston, for instance, have already started shipping."

And there's a picture of the ASUS board here, as it appeared at CeBit: Text Note the two 32-bit PC1066 RDRAM modules shown... they will be bundled with the motherboard.

So let's see, 32-bit PC1066 RDRAm gives us a bandwidth of 8.4 GB/s. Let's see you do that with DDR. To get that level of bandwidth with DDR you'd need dual DDR533. LOL, talk about vapor. You lose

 

[Acanthus]

right but when you put in a 32bit chip it disables the 2nd slot on the controller, so the bandwidth is actually the same.

AS FAR AS MY UNDERSTANDING GOES, please dont flame me if im wrong

to my knowledge all it does is remove the requirement for it to be installed in pairs, it doesnt give the ability to have 2 32 bit chips on seperate channels because both channels operate off of one chip.

A new memory controller would have to be developed to make 32bit dual channel work, otherwise putting in 4 rimms would give you 9.6GB/second as well, which doesnt make sense.

 

[BDSM]

Morph. I never said there wouldn't be 32bit RDRAM soon. I said there won't be dual channel 32 bit rdram any time soon!

Regarding your "challenge" to attain 8.4GB/sec with DDR. Since you are talking about technologies not yet released, I myself would go with a dual sledgehammer system. It will, coupled with plain DDR333 yield 11.4 GB/sec. And on top of that a latency that will be ultra low due to the built in 128 bit ram controller of the hammer. Also since we are talking to yet releeased stuff.. there will be DDR-2.. And qdr.. and who knows what else!

The Pentium 4 will use a 133 mhz bus (533 mhz datarate) for at least the rest of this year and most likely a good bit into the next.
Then, some time in 2003 .. or 2004 they will move to 166 mhz bus.

So.. No matter how well DDR or RDRAM might scale in that timeframe, the p4 won't be able to use more bandwidth than that.

So in the nearest 12 months or so it will boil down to who can provide enough bandwith, with the lowest possible latency and at the best price.

My bet is on both single and dual channel DDR 266, and later DDR 333. As mainstream and performance options.

 

[MadRat]

In a year Intel will come up with a chipset integrated with an L3 cache of 1T-SRAM in order to speed up the P4 when combined with either memory, so whats the big argument? My sources have told me this in confidentiality, so I cannot reveal them at this time lest they become a pariah on this website. I also heard that putting in four 32-bit PC1066 chips into one board with make it run octuple-channeled for 16.8GB/sec! Of course, this was from another more respectable website. By then 16-channel DDR2 boards will have 20.6GB/sec of bandwidth, so RDRAM is altogether going to be dropped by Intel....





Hey, I'm just kidding.

 

[smp]

Where's those benchmarks? What do you guys... have jobs or something?!! :Q

 

[oldfart]

<< you guys... have jobs or something?!! :Q >>


Yeah. Been real busy too. I'm not gonna have a lot of time to mess with this. Still waiting for my Ti4600

fkloster, could you please answer the questuons I've posted a couple times?

Thanks

 

[Zakule]

<< Looking @ real time frame rates is useless. Any tests have to be actual benckmark demos. RTCW Checkpoint, Q3 Demo FOUR, UT Thunder and/or UTBench, SS Karnak, SS SE Grand Cathedral, etc. I dont have Dungeon Siege. This has to be a controlled test. Since you have yours setup already, some settings and some things I need to know: >>



If you're implying that a utility like FRAPS (which measures real time framerates) is useless vs. a benchmark utility like UTBench then I could not possibly disagree with you more as I feel the exact opposite is true. For the record, I'm not trying to be insulting, just stating my opinion.

I've been busy, too, guys, so it may be a while before I get my NT6 dude over here. I may have to start another thread on it since I'll be swamped well into next week.

 

[oldfart]

fkloster used it


<< Max Payne was very difficult to determine average but floated around 100 to 150fps >>


"floated around 100 to 150 fps" Please explain how just looking at real time framerates in a given game can be used to accurately compare two systems. Wouldn't a demo based benchmark where the level is exactly the same and a resulting score be a better idea? Since every review site on the planet does it that way, I think I'll stick with a benchmark.

 

[Acanthus]

ill bet my rig can take ya both, rambus pc1200, ill have it within 2 weex

 

[Zakule]

<< "floated around 100 to 150 fps" Please explain how just looking at real time framerates in a given game can be used to accurately compare two systems. Wouldn't a demo based benchmark where the level is exactly the same and a resulting score be a better idea? Since every review site on the planet does it that way, I think I'll stick with a benchmark. >>



Here's my explanation:

Every review site also seems to think that how many FPS you get in Quake3 is somehow relevant to how current applications will run on your system. I don't think it is, and I disagree with the methodologys that many review sites use to conduct their benchmarks. Turn your system on, play a game and look at your FPS. Compare your results with another person who has a different system config. Do this with several games. Whoever is getting a more consistent MINIMUM framerate has the more powerful system. Whether it's due to drivers or the OS doesn't really matter. The bottom line is, which setup works best. Not in theory, but in practice with the applications that you use.

Why on earth people think that how well their system does in one or two demo-based benchmarks matters, I don't think I'll ever understand. Benchmark the applications that you actually use. It is so damed simple.

Let's say, for instance that your system scores 7680 in 3DMark2001. That's a pretty high score. Now, you go to pop in Morrowind and you notice that your FPS drops to about 23 when you're panning. Okay, that's a bummer. Now, you try Dungeon Siege and the scores are even worse. Hmmmm. On your friends system Morrowind worked just fine, but his 3DMark2001 were only 6927. Guess what, he has the better system. The example I just gave is completely fabricated but I've seen many other examples that weren't, most of them by people who kept claiming that their Radeon8500's were so much better than my GeForce3. Yeah, maybe in 3DMark2001. Maybe in one simulated environment, but what about in the games? That is an entirely different story, and guess what, it is the one that matters. I don't care if I'm getting 10,000 in 3DMark2001 if my 3D app's stutter because of poorly designed drivers. That's not to say that I didn't see some apps perform better with the Radeon, but most are just more compatible with GeForce technology. But I digress.

 

[oldfart]

<< play a game and look at your FPS >>


Sorry, WAY too subjective. The tests results will be "around 100 to 150" or "75 to 100" or so. Nowhere near accurate enough to draw any conclusion from that kind of data. Not even close to being a valid method to copare systems. If you are benching two systems, the EXACT SAME TEST must be run to compare them. Period. Just watching a real time framerate counter can not do that. Framerates will also vary TREMENDOUSLY from one are of a game to another. Doesn't make any sense to me at all.

 

[Zakule]

Too subjective? Are you going to play 3DMark2001? Oh, that's right you can't. It's a benchmark not a game. So tell me again why that number it spits out and the end matters more than how well Serious Sam works.

I'm just not following your logic, here. Why should I care more about a raw performance figured (delivered by a single graphics engine that exists only to give me a general idea of whether or not my system can handle 3D at all) than how well the games that I play work?

 

[oldfart]

Who said anything about 3Dmark? I hate that thing. I never mentioned it once. You did.

Anyway, I did a few tests. I just used the framrate counter. So here they are!!

Q3 I got ~ 265 - 400 FPS
UT, ~ 165 - 275 FPS
RTCW, ~ 125 - 225 FPS

There we have it!!

 

[Zakule]

<< Who said anything about 3Dmark? I hate that thing. I never mentioned it once. You did. >>



I was just using that as an example, I could've just as easily have used UTBench.