Why do the majority of people here like DDR better than RDRAM??

[Pabster]

SteelCityFan wrote:

"PC1066 is going to cost more since it is a lot faster than DDR333... and it is very new to the market. If you want faster Ram, you pay more money. Why is this surprising to you?"

Got to call ya out on that one, bud. PC1066 isn't a "lot" faster than DDR333; in fact, it falls behind in several areas, and squeaks out only a point or two in very specific (extreme memory bandwidth dependency) applications. Considering that DDR333 costs less, runs cooler, and comes in a wider array of module sizes, I'm having trouble understanding why anyone would bother with RDRAM. Keep in mind that we're comparing a dual-channel architecture to a single-channel one as well, which makes DDR333 look even more spectacular. Once dual-channel DDR chipsets for the P4 arrive (particularly from Intel ), even PC1200 RDRAM (assuming it ever materializes) won't be competitive. Even with a 120mm fan mounted on each RIMM to keep cool

 

[Soccerman]

Soccerman, you know damn well there is a big difference between constant (average) thoughput and peak (highest) attainable throughput. Why are playing devil's advocate

fkloster, average throughput is what matters more, becuase average happens more often than peak, correct? if this is the case, then in theory the time which DDR SDRAM is using the same amount of bandwidth (or slightly lower), is larger than when RDRAM is achieving it's peak. if RDRAM indeed does have longer times before data is transmitted than DDR SDRAM (which I can neither confirm nor deny), than it will not achieve it's peak bandwidth as often as DDR SDRAM right?

 

[cmdrdredd]

Sure RDRAM is best for a P4 but DDR is more standard blah blah...

it comes down to what you do...if you run SiSoft Sandra memory bandwidth tests all day then RDRAm is for you. If you actually do some work or games for that matter it won't mean diddly poop if you video is crap. Add to that the difference in playability between 140fps and 180fps is almost NOTHING!

 

[Athlon4all]

I think that the difference between PC1066 and DDR333 on 845G and P4X333 is not as big as it has been in the past I will admit that, but I still think PC1066 is the best.

 

[cmdrdredd]

well...let me say this to help the guy deciding between DDr or RDRAM. Do you have an Athlon rig with DDR now? If s just buy a DDr mobo and P4 northwood now and pop in your already in posession memory. Then when you feel you want or need to drop a stick of DDR333 or DDR400 in there and clock the FSB higher. The memory really has little to do with the overclockability of the CPU. I've seen people hit 3Ghz with a DDR based P4 system. This is of course running from a 2.26 Northwood CPU.

 

[SexyK]

Originally posted by: busmaster11

Originally posted by: SexyK
Anyway, I also beg to differ on the "intelligent and efficient design" point. I think any intelligent Anandtech user could see that the fact that the P4 will be able to run at 3GHz+ on a .13u process is a fairly "efficient" and "intelligent" design. It isn't worse engineering just because it takes a different approach. AMD would love to have a chip that could scale to 10GHz in the freseeable future, but it looks like they are having trouble breaking the 1GHz mark with the Hammers right now on the same process Intel is getting 2.53GHz (many OC's to 3GHz+ TODAY!) with. So while AMD is bumping their "superior" architecture 66MHz at a time, Intel will continue to roll out chips will 100-200MHz jumps and pull away from AMD. We'll see what hammer brings to the table. but to call the athlon more "intelligent" than the P4 design is pretty stupid, IMO.
Kramer

You're comparing pre-production 64 bit Hammer clock speeds right now with the latest P4? Exactly whose intelligence are we supposed to be judging?
:disgust:

you know damn well that that wasn't the point of the post, so please stop trying to shape anything anyone says to fit your wants and desires, its really pointless to argue with someone like you, since you're clearly unable to possibly listen to, understand, and appriciate any perspective but your own. Enjoy your Athlon/DDR and I'll enjoy my P4/RAMBUS. End of arguement.

Kramer

 

[apoppin]

Originally posted by: cmdrdredd
well...let me say this to help the guy deciding between DDr or RDRAM. Do you have an Athlon rig with DDR now? If s just buy a DDr mobo and P4 northwood now and pop in your already in posession memory. Then when you feel you want or need to drop a stick of DDR333 or DDR400 in there and clock the FSB higher. The memory really has little to do with the overclockability of the CPU. I've seen people hit 3Ghz with a DDR based P4 system. This is of course running from a 2.26 Northwood CPU.

After all this, I am really confused. I guess I will go with DDR SDRAM and keep my $50 towards a better videocard. I don't need a couple of extra points on memory scores.

No, I don't have an Athlon rig - it's an older BX system with a 1.2Ghz Tualatin Celeron and SDRAM. I must buy completely new memory amd will definitely go with a 1.6A P4 (for starters).

 

[busmaster11]

Originally posted by: SexyK

Originally posted by: busmaster11

Originally posted by: SexyK
Anyway, I also beg to differ on the "intelligent and efficient design" point. I think any intelligent Anandtech user could see that the fact that the P4 will be able to run at 3GHz+ on a .13u process is a fairly "efficient" and "intelligent" design. It isn't worse engineering just because it takes a different approach. AMD would love to have a chip that could scale to 10GHz in the freseeable future, but it looks like they are having trouble breaking the 1GHz mark with the Hammers right now on the same process Intel is getting 2.53GHz (many OC's to 3GHz+ TODAY!) with. So while AMD is bumping their "superior" architecture 66MHz at a time, Intel will continue to roll out chips will 100-200MHz jumps and pull away from AMD. We'll see what hammer brings to the table. but to call the athlon more "intelligent" than the P4 design is pretty stupid, IMO.
Kramer

You're comparing pre-production 64 bit Hammer clock speeds right now with the latest P4? Exactly whose intelligence are we supposed to be judging?
:disgust:

you know damn well that that wasn't the point of the post, so please stop trying to shape anything anyone says to fit your wants and desires, its really pointless to argue with someone like you, since you're clearly unable to possibly listen to, understand, and appriciate any perspective but your own. Enjoy your Athlon/DDR and I'll enjoy my P4/RAMBUS. End of arguement.

Kramer

LOL. I'm supposed to read your mind. That's almost as articulate as josphII's one liner...

Please explain what exactly is your point in involving Hammer in this discussion then, when naming XP would have been more relevant otherwise; if not we will be forced to believe you're merely throwing names out for the heck of it...

Of course, anyone would agree that AMD would like to have thoroughbred scale the way the P4 has, but it doesn't. Kudos to the P4 for that. But if the tables were reversed and the Athlon had the highest performing CPU right now (as it did six months ago) you would be argueing the opposite of this. And no, I do not believe the Athlon is the more intelligent design, merely that in terms of clock speed, the P4 is relatively "lazy".

 

[imgod2u]

I'm gonna have to pick at this one.

Originally posted by: busmaster11
I've noticed that AMD has generally preferred the more conservative route. While Intel goes for brute force architectures that enable higher clock speed, ie P4 / RDRAM, AMD has gone for more intelligent, yet evolutionary designs.

SDRAM is far from revolutionary. DDR technology was in RDRAM long before it was in DDR SDRAM.
AMD's K7 design is merely a brute force way of shoving more execution units in a chip based on an old 10-stage design with very little life left (well, at the initial K7 launch, it had some life left but we're seeing it reach its end now). Think Pentium 3 (P6 core) with a few more execution units and fetch capabilities and you get the K7. It's a great short-term, uber-performance per clock design, but in no way revolutionary or even innovative.
The P7 core is a high-scalable design that will last years. The P4 is a hacked-down, compromised design that relies on brute clockspeed to gain speed advantages and make up for its lack of fetch abilities and FP power. Same concept of stressing one feature to make up for the shortcommings of another part of the design.

All things being equal, increasing performance through higher clock speeds will generate instability more often than increasing performnce by making a more intelligent chip, which is what AMD has done. While DDR and QDR, if and when it arrives, will always be able to trace its roots back to SDRAM, RDRAM is more revolutionary. But its key negative, that it does so little per clock cycle, bugs me. That plus the fact that its been years since its introduction, and its prices are still double that of DDR.

Work per cycle is nothing more than a design concept. It's relation to efficiency, or "intelligence" as you call it is the same as clockspeed's relation to performance. In other words, it has none. The market is moving towards narrow-band, high per-pin frequency designs. Would you say Hypertransport "bugged" you because it is only 8-bits wide? It's simply a different route and currently, it's the better route.

Intel has seen it, and its no coincidence that they immediately surfaced with i845DDR immediately after their contract with rambust expired. I hope the bastards at rambust are forced to pay back every penny they conned out of the Jedec organizations.

RDRAM had a lot of potential. However, I think even Intel was a bit frightened by Rambus's boldness. If you ask me, they had a sweet deal with Intel. They really didn't need to pull all of that crap. But hey, Granite Bay's looking great.

the Athlon IS as bandwidth hungry as the P4. that explains why you see increases in performance as the fsb and RAM increase in speed.

http://www.anandtech.com/cpu/showdoc.html?i=1595&p=12 seems to indicate differently.
The Athlon is not memory-starved, not at 1.67ish GHz anyway. Memory bandwidth requirements are directly releated to how many instructions are fetched per second and how many instructions the CPU works on at any given time. For the Athlon, that's 10 integer instructions at any given time (once the integer pipeline is filled up) and x86's fundamental limitation of 3 instruction fetches per clock. At 1.67 GHz, that's not a lot of strain on the memory bandwidth.
The P4 fetches 3 x86 instructions per cycle as well and works on 20 integer instructions at any given time (assuming the pipeline is filled). At 2.0 GHz or 2.2 GHz, that's a big strain on memory. So in other words, it is much more memory starved.

 

[busmaster11]

Originally posted by: SteelCityFanI also find it amusing that they (HardOCP) only benchmarked games using 640x480 which does not take much memory bandwidth. Of course they cover this by saying that they don't want the Video card affecting the results. However, since they are using the same card on each system, their statement defies logic, and is one of the stupidest things I have ever heard.

I don't get it. How come? Same video card, so the video card won't be the variable. Low resolutions... So video card fill rates won't be much of an issue. That leases - ram/chipset as the variable to benchmark, right? Am I missing something?

I think cmdrdredd is right, after looking at the benchies again. Get RDRAM if you just run Sandra all day.

 

[busmaster11]

Originally posted by: imgod2uSDRAM is far from revolutionary. DDR technology was in RDRAM long before it was in DDR SDRAM.

I SWEAR I've never said SDRAM was revolutionary. VX chipset days perhaps...

As for DDR technology being in RDRAM, sure, I believe it, back when DDR standards were in the works and Rambus was hiding info from Jedec. I guess it depends on how you look at it. I've read articles on rambus white papers but there's no way I believe that beneath SDRAM and DDR SDRAM are lgally patented rambus underpinnings...

Perhaps that fact alone is the root cause of my hatred of rdram...

 

[apoppin]

According to TomsHardware, Rambus is the ONLY way to maximize performance from a P4

This review is todays' . . . whaddaya think?


His conclusion:

Still, the actual Rambus technology leaves no room for complaint: RDRAM offers a large bandwidth of up to 4.2 GB/s and offers the best performance, particularly when used together with the Intel Pentium 4. Intel's only chance to keep up with the performance level of PC1066 memory (533 MHz) is a new memory interface (chipset). Internal documents from Intel reveal that at the beginning of next year, the successor to the P4 will be launched. Codenamed "Prescott," it will integrate a dual-channel DDR interface with DDR333.

Here, you should keep in mind that the Intel Pentium 4 has a maximum bandwidth of 4.2 GB/s. In the near future, this will reach well over the 3 GHz limit. Only Rambus memory in the form of PC4200 (533 MHz) is capable of taking full advantage of this bandwidth. By using DDR SDRAM, such as DDR266 or DDR333, the bandwidth remains restricted to 2.1 GB/s and 2.7 GB/s, respectively

So, now I am not so sure.

 

[imgod2u]

Originally posted by: Soccerman
Soccerman, you know damn well there is a big difference between constant (average) thoughput and peak (highest) attainable throughput. Why are playing devil's advocate

fkloster, average throughput is what matters more, becuase average happens more often than peak, correct? if this is the case, then in theory the time which DDR SDRAM is using the same amount of bandwidth (or slightly lower), is larger than when RDRAM is achieving it's peak. if RDRAM indeed does have longer times before data is transmitted than DDR SDRAM (which I can neither confirm nor deny), than it will not achieve it's peak bandwidth as often as DDR SDRAM right?

Not quite, let's use the following analogy:

2 guys are in a library. They have to read a certain set of books. One person walks slow and can't find books as fast, but once he finds the book he can read it very fast. The other person runs like Forrest Gump and can find any book pretty fast, but once he finds it, he looks at the pictures first and then tries to read it. If these two were sent to read an encyclopedia set of 20 books. Even though Forrest can find it first, he's stuck there reading it slowly word by word while brainiac there will finish reading the set in record time. In this case, brainiac would finish reading first. If, on the other hand, they were sent to read 50 different, completely unrelated pages in 30 different, completely unrelated books. Forrest would finish first.
The speed at which one reads books would be the bandwidth. The MB/sec. The speed at which one finds a book would be the latency. A lower latency does not make a faster bandwidth, but rather, a faster completion of a job if the job was just reading small bits of data. A higher latency, but high bandwidth solution would be able to read and write huge amounts of information in no time, but if it were to only write small bits of data at different times (write one bit, wait a while, read one bit, read a while) and the overall job will be done slowly.
Rambus would be the brainiac, SDRAM would be Forrest. This is where prefetch comes in. Say you knew you had to read bits of data in short sessions later on while you were working on something. Instead of fetching them from memory when needed, you load the neccessary ones onto the cache all at once. This would mean you're not searching, opening and transfering a small bit of data but a large piece of data (or rather, the total combined of all the small bits). This would make a job go much faster.
Bandwidth simply means the speed at which data can be transfered once it was found. Bandwidth benchmarks are based on transfering streams of data. Latency is another aspect which affects performance and is separate from bandwidth.

 

[busmaster11]

Well, the bottom line is that RDRAM will outperform DDR by a negligible amount, or by a substantial amount, depending on who you believe. The first thing that came to mind after reading this is that Tom shoulda had some DDR333 board as a reference board to comapre to, otherwise those numbers mean very little. Ultimately, the article was a motherboard shootout, and not one that will decide for you DDR or RDRAM.

I do maintain that there is a substantial price difference, and I'm not quite ready to believe that RDRAM is worth it if you're on a limited budget, especially if its future is in doubt as I've always believed.

I also believe that Rambust hurt Intel by not allowing them to release a DDR chipset before the end of last year and I am still incredulous about a whole lot of RDRAM support beyond GRanite Bay, which may well be just Intel's way of hedging a bet if DDR performance doesn't pan out.

 

[apoppin]

Well, Tom is saying it is substantial - 4.2 GB/s for the RDRAM vs. 2.7 GB/s for the DDR 333.

Or am I missing something?

 

[Dug]

Well, Tom is saying it is substantial - 4.2 GB/s for the RDRAM vs. 2.7 GB/s for the DDR 333.
Or am I missing something?

It is if you have a motherboard that can't do anything, or if your not overclocking.
Fortunately there are other options such as the 845g that have 3:4 and 4:5 memory settings
and allow for AGP and PCI speeds at ANY SPEED. And yes that means you can get DDR 400 speeds.

As far as getting a new rig, I wouldn't get anything but DDR.
The boards and the memory are so much more flexible in what you can do with overclocking.

Recent 1.6's are easily hitting over the 160fsb mark now. Even the old one's did 150fsb pretty easily.

I don't know about rdram hitting those high of speeds.
I'd rather run a 1.6@160fsb-170fsb on DDR than a 1.6@133fsb with rdram

If rdram can hit those high fsb without effecting the agp and pci then I'd go for it in a heartbeat.

Maybe someone can enlighten us on how high really good rdram can go.

 

[BFG10K]

RDRAM is faster than DDR RAM but I think most people like DDR because it's cheaper, doesn't have to be used in pairs and uses less power and heat, plus the fact that it's compatible with both Intel and and AMD platforms, making future upgrade paths to either platform an option.

 

[alphatarget1]

isn't RDRAM:

more expensive?
less overclockable compared to DDR-SDRAM?

RDRAM is superior compared to DDR in terms of bandwidth and speed. DDR might be my choice for the computer but they might not be the best. Intel CPUs are designed to be used with RDRAM and AMD CPUs aren't, so there aren't much to compare anyway.

 

[NinjaGnome]

Well I went with the samsung true pc2700 ddr for my p4 northwood 1.6ghz im running the proc at a 166mhz fsb which makes it 2.65 ghz which is fast enough for me without doing any tweaking. When i set my ram in the ratio (which i havent done yet but know the ram can easily handle) my setup could beat a lot of things out there for a lot less money than what most paid. The 850 chipset motherboards cost on an average of 10 dollars more than 845's and pc1066 rd ram is hard to come by and you pay a huge premium for it. If i could afford it i would get the i850 and 512megs of pc1066.
my mobo proc ram combo
1.6a northwood (got from newegg) 137
256megs true pc2700 samsung DDR (bought off a friend) 65
Asus p4b266c (bought here) 90

Id say 292 dollars for a 2.65 ghz system isnt bad.

 

[fkloster]

hmmm, apparently many of you are running your 1.6ghz P4's well over 2.4 ghz....are you also using the stock cooling to save money? and you are achieving 90% stability or better?

 

[oldfart]

I have an "average" 1.6A. Not the best, not the worst. I run mine @ 600 FSB, 2.4 GHz, 1.65 Vcore with the stock HS/Fan. I run Samsung PC2700 using a 3:4 ratio for DDR400 3.2 Gb/s bandwidth. 100% stable setup. People that get the "really good" CPU's can get them up to 2.6+ GHz.

 

[apoppin]

It is if you have a motherboard that can't do anything, or if your not overclocking.
Fortunately there are other options such as the 845g that have 3:4 and 4:5 memory settings
and allow for AGP and PCI speeds at ANY SPEED. And yes that means you can get DDR 400 speeds.

As far as getting a new rig, I wouldn't get anything but DDR.
The boards and the memory are so much more flexible in what you can do with overclocking.

Recent 1.6's are easily hitting over the 160fsb mark now. Even the old one's did 150fsb pretty easily.

I don't know about rdram hitting those high of speeds.
I'd rather run a 1.6@160fsb-170fsb on DDR than a 1.6@133fsb with rdram

OK, now we are getting somewhere . . .

What kind of memory bandwith are you getting with "DDR 400" with the FSB at 160(plus)? How does it compare with the 2.7GB/s of DDR333 (and RDRAM's theoretical 4.2 GB/s)? What practical difference is there?

(yeah, I also posted this question in O/C'ing as this topic was getting pretty hot here)

Edit (again, sigh . . .). I see oldfart is getting 3.2 GB/s from his "average" 1.6A o/c. Is RDRAM's (+1.0 GB/s) advantage purely 'theoretical'?

 

[JellyBaby]

apoppin, only memory benchmarks can really see the difference aside from the odd memory bandwidth-intensive game.

For instance, if you play 3D shooters using the Quake 3 engine, you may benefit slightly from RDRAM or OCed DDR. Otherwise the difference in performance is only theoretical. We're talking percentage increases in the single digits here and only in software that accesses memory a lot.

 

[sean2002]

Athon is old technology, there's no way around it.

True the Athlon is an older technology, but clock foir clock it still spanks the P$, now that said my next system will be some 533FSB P4

 

[apoppin]

Originally posted by: JellyBaby
apoppin, only memory benchmarks can really see the difference aside from the odd memory bandwidth-intensive game.

For instance, if you play 3D shooters using the Quake 3 engine, you may benefit slightly from RDRAM or OCed DDR. Otherwise the difference in performance is only theoretical. We're talking percentage increases in the single digits here and only in software that accesses memory a lot.

Well, it looks like the RDRAM supporters have no case to make on "bang for the buck" IF you overclock. AND with the future of RDRAM in serious doubt, it seems (at least for a Ferengi like me) DDR is the ONLY way to.

Thanks for helping to lay this to rest.