Intel 845 DDR Chipset Review

[Wombat]

You'd need the sdram ddr modules themselves run at least at 250 mhz(500 mhz DDR) in a dual channel 64 bit architecture to match the performance of a dual channel 32 bit rambus memory running at 533 mhz(pc 1066).

Yeah? And? You think only RDRAM tech has been evolving for the last two years? Now start piling in more and more sticks of memory, and which bus is likely to handle the load better?

And besides,regardless of it's disadvantages,it's still one the fastest types or memory around for the p4 and only now,over a year since the 850 has been introduced,do we see DDR chipsets getting close to it's performance in real world apps.

You are very confused. Here's what happened, and some of how:
1. Intel designs a processor around a high-bandwidth, high-latency technology, hoping to use it to push Intel's competitors into at least using technology from a company Intel holds stake in.

2. Said technology (Rambus) happens to suck in the practical world, with cost, yield, and reliablilty issues. Intel has to put out the MTB, and still uses SDRAM in its server-class-ish offerings.

3. Intel switches to DDR solution ASAP, as it gets its legal bindings to Rambus loosened. DDR "reaches Rambus performance" because it is used with a CPU designed around bandwidth.
P3s and AMD offerings were designed with SDRAM bandwidth in mind. Putting DDR with them was/is nice, but what good is a fatter fuel line if the engine can't burn gas that fast?


Also, let's not forget that Rambus, Inc. is a bunch of complete ****oles. They fraudulently patented technology, and tried to use said patents to push people to their technology when it couldn't succeed on its own merits. No matter what Rambus sells, that's reason enough for me not to buy it.

[superfly]

Look,you know as well as i do that the main selling point of DDR sdram is that it had a much better price to performance ratio than Rambus did,but today that's no longer the case and that yeild issues are still a concern is no longer true either since you don't see pc 600 and pc 700 rdram on sale anymore because 12~18 months ago they had trouble hitting it's intended speed.


In fact yeilds are so good that the vast majority of Rambus modules sold right now can easily operate at even higher operating speeds with no trouble at all, that's one of the reasons why it looks like intel will push the release for processors running at 533 mhz even earlier than previously though,if we follow the rumors floating about.



Like i mentioned in my previous post,the price difference is now about 30$ in my area between 256 megs of Rdram and the same amount of DDR sdram pc 2100,with pc 2700 costing more,as early as six months ago,no one could have predicted that memory price would drop so low(yes even intel) that it makes the release of an intel DDR sdram chipset have much less impact on p4 sales right now,compared to six months ago where DDR sdram still had a significant price advantage over Rdram.


You're right in saying that DDR sdram will handle more sticks of memory,but in this day and age where 256 meg sticks are quite cheap,it's a non issue with 95% of users out there.

By the time that it becomes common where we see apps that most users use actually need a gig of memory,any current p4 or athlon system will be obsolete anyways,so it's a moot point.

Only those who run either extremely demanding apps or run servers that actually need multi gigabyte memory equipped machines,then DDR sdram is better suited in those situations because of it's latency advantage.


The main reason why p3's didn't make use of the extra bandwith that DDR sdram provides isn't because the processor couldn't use it,it's mostly because it's FSB wasn't fast enough to fully use it,limited at a maximum of 1 gig sec.


For instance the L2 cache in the p3 has about 20 gigs of bandwith and you can be sure that the cpu uses every bit of it,so it sin't going to be a memroy standard that runs at least 10 times slower that's going to satisfy it,that's for sure.


I know that Rambus(the company) are a bunch of lying,cheating ****oles,but what matters to me is performance and RDram is delivers and is supported by a stable,compatible chipset,something that companies like SIS and ALI are far from proving.

It takes more than a good memory controler to make a good chipset.

[Wombat]

Look,you know as well as i do that the main selling point of DDR sdram is that it had a much better price to performance ratio than Rambus did,

Actually, you think you know it, and I know you're wrong. SDRAM wins on latency, reliability, cost, and expandability.

The main reason why p3's didn't make use of the extra bandwith that DDR sdram provides isn't because the processor couldn't use it,it's mostly because it's FSB wasn't fast enough to fully use it,limited at a maximum of 1 gig sec.

The CPU <B>is</B> the FSB limiter. There's RAM out there that will run a good bit faster than the current generation of CPUs is really to take. But CPUs are designed to expect a certain amount of performance from their bus interface, and giving them more than they expect produces sub-optimal gains. If a processor's FSB interface could handle the higher speeds, then just program in a lower multiplier, and unleash that P3 1166/166 on the market. Boom! More performance with minimal changes to the design. So why doesn't that happen?

For instance the L2 cache in the p3 has about 20 gigs of bandwith and you can be sure that the cpu uses every bit of it,so it sin't going to be a memroy standard that runs at least 10 times slower that's going to satisfy it,that's for sure.

Have you ever actually spent any time studying cache and/or queueing theory? It certainly doesn't look like it. 98+% (and more realistically 99.9x%) of the memory requests from the L1 to the L2 will be serviced directly from what the L2 already has in store. Some fraction of a percent of those requests will miss the L2, and will have to go and hit the RAM for some small amount of vital information, and need it as soon as possible. Here again, latency is more important than bandwidth, because the CPU is basically languishing through every clock that it waits on the data. Loosely speaking, a memory standard could be 50(!) times slower than the L2 bandwidth, and still meet the needs.

Not only are you wrong, but you gave another example of why SDRAM is more desirable than RDRAM. Thanks.

[superfly]

I think there's no more point in arguing this really,you're too set in your ways as i am in mine and there's tons of proof out there proves my point that when Rambus is used with with the p4 and the 850 chipset, it's still the fastest setup out there,regardless of Rdram's latency issues,wich i agree with you are inherently very high.


If it's due to either a desing characteristic of the p4 cpu or the 850 chipset itself or even both,i don't know,but regardless,it works extremely well and cost isn't an issue anymore like you point out yet again,at least for home users.


Personally i haven't seen any case of Rambus memory not being as reliable as any other type of memory currently available,so i'm not sure what your trying to prove when you say that sdram is more reliable than Rdram.


Again i agreed that Sdram is inherently more expandable than rambus and has better latency as well,but that only applies to servers that absolutely need huge amounts of memory or other extremely demanding apps that 95% of users don't use and by the time they will,even todays high end systems will be obsolete anyhow.

Will the vast majority of users out there need more than a gig of memory anytime soon,even if we never see Rdram modules bigger than 256 megs(which we will anyhow).

Idon't think so and you know it as well as i do....


And again my point still stands that by the time that happends,any high end system availble today is obsolete regardless of what type of memory it's currently using.


I believe that intel didn't release a 1166 mhz p3 with a 166 mhz fsb may have been one of timing for the most part....

We all know that the0.18 micron version was on it's last legs and when intel tried to release a 1133 mhz version it got quickly recalled because it wasn't operating reliably at that speed and by the time it was possible to build it at 0.13 micron which allows it to work reliably at ghz+ speeds,the P4 was already on the market.

[Paulr]

Well, it just appears to me that it's suddenly fashionable to bash Rambus along with MS.
This board seems to have a high number of people who just love to bash products even though the results are sitting there staring people in the face.

Lets see, MS Windows Sux, M$ Windoze, M$ Shit to name but a few of the names given to the software giant.
No constructive posts about how competitors products are better or how there are plenty of alternative products for the home user and what they offer over the MS products.
It's a fashion, bash MS, it's cool doing that.

Then we move onto Rambus.
Various review sites have already had off the shelf Samsung RIMM's running on a 133Mhz FSB (PC1066 specs) without a single problem.
The benchmarks show that Rambus is still the better performer.
Rambus prices haven't suddenly taken an increase in price so the difference between RIMM's & DDR is hardly noticeable.
The next generation RIMM's run faster and are actually cheaper to manufacture, yet we still get the bashers.
RAMBUS Sux, RAMBU$, yawn.

Do you honestly think when I was looking to upgrade to a P4 I didn't look at the reviews and the various options?
Do you think that I didn't know full well that within two week of my purchase the DDR solution would be available?
I knew all of this, however the RIMM solution offered me better performance at not much more cost.
I'm talking real world here, I'm talking about firing up my PC and using it for applications/browsing/games etc.
So what if I'm running on a higher latency etc.

Instead of the constant fashionable bashing of products from various manufacturers why don't some of you actually tell us why the alternatives are so much better rather than quoting the odd bit of technical jargon from in a lot of cases less then reputable sources.

[Pace]

Ok, lets see...here's a little price-up of some components...

I priced the AMD system, then a P4 for the same price roughly.

Athlon XP 1900+ (1.6GHz) - £230
Abit KG7 RAID (AMD 761) - £110
Kingston 256MB PC2100 - £50

Pentium 4 1.8GHz S423 - £190
Abit TH7 RAID (i850/423) - £130
2 x Kingston 128MB RIMM PC800 - £75

Athlon 1900+ Total: £390
Pentium 4 1.8 Total: £395

What would you choose?

I didn't put much thought into this, but I'd choose the P4. That's based on buying "today". Not last month, not speculating on next month, but today.

I wouldn't buy that exact config either, but it's just a comparison.

Now then, anybody have AthlonXP DDR vs Pentium4 RAMBUS benchmarks?

Just curious

P.

[Novdid]

Those benches are everywhere. An AthlonXP 1900+ is much closer to a P4 2,2 GHz than to a 1,8 GHz. Try the price of an AthlonXP 1600+ if you want the same performance. A KT266A board is also a bit cheaper than the 760 board and faster, but it will need some tweaking to reach the same stability, wich I'm personally willing to do...

I wouldn't buy Kingston RAM either, it usually costs more than other brands, that goes for both the RDRAM and the DDR.

And if I go the AMD route I can later next year get a Barton CPU and use the same MB, but that sort of upgrading path is more or less ruled out with the socket 423 board.

BTW the Athlon is a much more balanced CPU than the P4, It performs good in everything you throw at it and it doesn't need SSE2 optimizations to compensate for a lacking FPU...

These are only some of my personal considerations I would make if I was going to buy parts like that.

Note that I'm not bashing anyone or any company in this post.

[superfly]

While it's true that the athlon does offer better performance for less money than the current P4,we still have to consider that the p4 northwood is nearly here and it seems that it's about 15% faster than the existing p4 willamette,even when both are running at the same clock speed.


Here's an early review of the northwood running on an asus p4b 266(845 DDR chipset).

https://www.hardware.fr/html/articles/lire.php3?article=403&page=1

It's in french,but you can see neat bar graphs that explain everything quite nicely and you'll notice that they managed to overclock that engineering sample to 2.53 ghz without even breaking a sweat.


It's also a given that there will be a fsb speed increase for the p4 platform in the near future which will further give another 15~20% overall speed increase and i checked with asus support and it seems that both the p4t-e and the p4b 266 will support 533 mhz fsb with just a simple bios update.


On the AMD side of things,we'll soon have the xp 2000 chip which will aparently be the last version that uses the palomino core,but what a little distressing is that it's replacement,even though it will be using a 0.13 micron process,won't have any architecture enhancements over the palomino(no 512k cache like the new p4 will have) and AMD has already officially confirmed that info.


Then add that there are also no fsb increases planned for the athlon,and i don't see how they'll be able to keep up with intel,at least until the sledgehammer series shows up,hopefully by the end of the year.

[Novdid]

Well AMD will probably switch to a 333MHz bus before Barton arrives, sometime in the 3rd quarter I think (I'm not sure about this, but it's the wisest thing AMD can do).

And it's not a difficult task for AMD to do, but it's important that the chipset makers release boards that supports the new FSB frequency fast.

And let's not forget that the dieshrink will make Athlons very very cheap. The die of the northwood will still be quite large due to the additional 256K cache, and I can bet all my money that Athlons will provide the best bang for the buck even in the future.

[superfly]

They'll provide the best bang for the buck,since AMD is willing to cut it's profit margins quite alot to keep it that way,but i'm not so sure anymore if they provide the best bang for the buck AND the best overall performance over the coming year,like it happened this year and the previous one as well....


And since intel has it's mind set in releasing p4's running at 3 ghz well before the end of the year,which the northwood core is quite capable of doing and with both a 533 mhz fsb and PC1066 rdram, it's going to be an uphill battle for AMD to compete in,at least until the sledgehamer/clawhammer series are ready.which it looks like they won't show up before late Q4/02 or even Q1/03.

[Novdid]

Well I wouldn't underestimate the Athlon's ability to scale in clocks either. Exactly one year ago people thought that AMD was in big trouble because the TB wasn't supposed to be able to clock higher than 1,3 - 1,4 GHz, and everybody thought the P4 would be a speed demon when they read the specs. Even If they didn't release a higher clocked thunderbird they would be able to scale to 1.7GHz (?), but they would need some serious cooling at those frequencies though (overclockers know).

Now that thoughrobread(sp?) won't feature 512kb cache, they will probably be very cheap as I said before. And they should be able to scale to atleast match the performance of future northwood CPU's, or else AMD will be in a serious hurry to release Barton which I have read will feature SOI tech, to continue to keep the performance crown, which they have had for 2 years.

And then the Hammer will show up sooner or later (probably later), and then probably Intel will have to pull something out of it's sleeves to keep up...now I'm getting carried away. No one can tell the future but you can always guess.

[isochar]

Originally posted by Novdid
Exactly one year ago people thought that AMD was in big trouble because the TB wasn't supposed to be able to clock higher than 1,3 - 1,4 GHz,

It didn't reach past 1.4ghz. Palomino is what pushed it past.

Now that thoughrobread(sp?) won't feature 512kb cache, they will probably be very cheap as I said before. And they should be able to scale to atleast match the performance of future northwood CPU's, or else AMD will be in a serious hurry to release Barton which I have read will feature SOI tech, to continue to keep the performance crown, which they have had for 2 years.

Those are some pretty big assumptions. Palomino's are already using .13m gates and some SOI. They've already reaped most of the benefits of the die transition...

Thus, Barton won't be a big deal. AMD is going to be playing catch-up until Hammer debuts. With its twelve stage pipeline it'll scale a little better than the K7.

AMD has not had the performance crown for two years. There have been interim periods with Intel holding the lead. It would have been more accurate to say they were the performance leader for the past two years.

And then the Hammer will show up sooner or later (probably later), and then probably Intel will have to pull something out of it's sleeves to keep up...now I'm getting carried away. No one can tell the future but you can always guess.

The P4 is poised to handle hammer quite well. It'll still scale better due to its 20 stage pipeline. Not to mention the benefits of hyperthreading will reach consumer PCs. Couple that with its 533fsb and use of dual-channel PC1066, it'll be an interesting time indeed.

[Novdid]

I'm trying to be objective here, isochar. I haven't been biased in my posts except the one I posted some day ago in this thread about Rambus. If Intel can provide a better solution than AMD I will buy Intel and vise versa. Today I feel that AMD is the best choice, considering price offcourse. Performance vise the offerings from both companies are quite equal.

AMD has not had the performance crown for two years. There have been interim periods with Intel holding the lead. It would have been more accurate to say they were the performance leader for the past two years.

I know that, and everybody knew what I was saying. Personally I think that you are very very biased in your posts and do not see the whole picture and you spit out some ridiculous uneducated comments that gives your posts a bad tone. Try posting like superfly has done.

[isochar]

I know that, and everybody knew what I was saying. Personally I think that you are very very biased in your posts and do not see the whole picture and you spit out some ridiculous uneducated comments that gives your posts a bad tone. Try posting like superfly has done. [/B]

Uneducated comments that gives your post a bad tone... hmmm ....

And they should be able to scale to atleast match the performance of future northwood CPU's, or else AMD will be in a serious hurry to release Barton which I have read will feature SOI tech, to continue to keep the performance crown, which they have had for 2 years.

Speak for yourself.

Here are my current system specs if you were wondering:

Dual Athlon XP 1.6ghz
Tyan TigerMP
1gb Crucial PC2100
Matrox G400Max
Adaptec 29160N
Creative Labs Audigy Platinum Ex
Hauppauge WinTV-PCI
3COM 3CR990-TX-97 NIC
Seagate X15-36LP 18gb
Western Digital 1000bb 100gb
Plextor Ultraplex 40TSI
Plextor Plexwriter 24x10x40
Toshiba DVD-303S
Mitsubishi LS-120
Areasys C990 "Genie" Case
PC Power & Cooling 400W Silencer
Sony CPD-G400
Microsoft Intellimouse Explorer 3.0
Microsoft Natural Keyboard Elite

[Novdid]

And they should be able to scale to atleast match the performance of future northwood CPU's, or else AMD will be in a serious hurry to release Barton which I have read will feature SOI tech, to continue to keep the performance crown, which they have had for 2 years.

Well maybe I should have changed that "should" to "have to".

Either way that was just my opinion on your post, I didn't mean to flame or attack you in any way and if I did I do ask for your apology.

BTW nice setup!!!

Merry Christmas to you all.