DDR vs. RDRAM

G

Guest

Guest
Alright so whats the deal with these two types of RAM??? I mean which one is better Im considering upgrading to a new comp soon and any input on this would be nice.
 

Arbee

Distinguished
Jun 4, 2001
305
0
18,780
DDRAM has a much better price/performance ratio tham RDRAM.

But the memory you choose depends upon the platform you pick. If you want a Intel Pentium 4 - then, for the time being, you'll have to buy RDRAM. If you want a Pentium III you can buy all three kinds of memory, although SDRAM (the cheapest) will suffice.

If you side with AMD a DDR supporting platform is your best choice.

How terrible is wisdom when it brings no profit to the wise
 

Raystonn

Distinguished
Apr 12, 2001
2,273
0
19,780
Using equal amounts of RDRAM and DDR-SDRAM, the RDRAM will outperform. The issue most argue is a matter of pricing. RDRAM costs more than DDR-SDRAM. IF you want the best performance, go with an RDRAM platform. If you want a better price, you might look at other options.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
 

Arbee

Distinguished
Jun 4, 2001
305
0
18,780
I'm afraid RDRAM doesn't outperform DDRAM.
DDRAM has superior bandwith and lower latency, hence better performance.

Of course that it is easier to implement a dual-channel RDRAM solution (although nForce will be a dual-channel DDRAM chipset), therefore one *may* say a RDRAM based system has better performance than a DDRAM based system, but not that RDRAM has better performance than DDRAM, at least for the time being.



How terrible is wisdom when it brings no profit to the wise
 

Raystonn

Distinguished
Apr 12, 2001
2,273
0
19,780
"I'm afraid RDRAM doesn't outperform DDRAM.
DDRAM has superior bandwith and lower latency, hence better performance.

Of course that it is easier to implement a dual-channel RDRAM solution (although nForce will be a dual-channel DDRAM chipset), therefore one *may* say a RDRAM based system has better performance than a DDRAM based system, but not that RDRAM has better performance than DDRAM, at least for the time being."


I will avoid the issue of your latency comparisons, which are actually incorrect. I've covered that in detail in prior posts.

Current RDRAM system implementations all use a dual-channel memory subsystem, plus the channel [implemented as an AGP bus] to the memory on the video card (thus three actual memory channels for the total system.) Current DDR-SDRAM system implementations all use a single channel memory subsystem, plus the channel [implemented as an AGP bus] to the memory on the video card (thus two actual memory channels for the total system.) The original poster is asking about getting a new system soon, thus we can eliminate any theoretical (read vaporware) systems and technologies not yet released. (The nForce chipset, by the way, still offers the same two memory channels counting the video card memory. It's just the first integrated chipset to offer the same number of memory channels as a comparable system with an add-on card would.)

Looking at present-day implementations for the sake of the poster's request of getting a new system 'soon', all RDRAM implementations outperform all DDR-SDRAM implementations currently in existence.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
 

Arbee

Distinguished
Jun 4, 2001
305
0
18,780
I will avoid the issue of your latency comparisons, which are actually incorrect. I've covered that in detail in prior posts.
Could you give me a link? - I'm really interested on that info. TIA
Current RDRAM system implementations all use a dual-channel memory subsystem
If one forgets the single channel i820 (some mobos still have them)...

plus the channel [implemented as an AGP bus] to the memory on the video card (thus three actual memory channels for the total system.)
Actually thats not all - if you want to include AGP and PCI (or even USB) devices the number of memory channels will be quite superior - most devices have memory, such as HDDs, CDRs and even NICs.

As I've said in the my original post, one cannot choose between RDRAM and DDRAM at our will, since there is no AMD based RDRAM solutions nor DDRAM Pentium4 solutions (I'm counting PIII as "dead"). Therefore the issue is between choosing an Athlon (DDRAM) or a Pentium4 (RDRAM).

Regarding nForce, it has a dual channel DDR controler and, AFAIK, there is no "graphic memory slot" - but if you have other info...
But it still is unavailable (hence still vaporware)


How terrible is wisdom when it brings no profit to the wise
 

FatBurger

Illustrious
Here's my take on RDRAM latency, as was explained to me by a friend who's read up on this more than I have.

RDRAM has high latency for the first read, but it lowers for each successive read. Basically, the memory has to be placed in RAM in the correct order, so it can be retrieved easily. This means programmers must give better algorithms so that data is placed in RAM in the correct order. This will allow faster retrieval with RDRAM. But I digress.

SDRAM will maintain close to the same latency through an entire extended read. RDRAM will become quicker as the reads progress, until it is asked to provide something that is out of order. So it'll be:
<font color=red>request.........read 1.....read 2...read 3.read 4.read 5.read 6.........read9</font color=red>
Make sense?

Ok, bandwidth. DDR-SDRAM has a clock speed of 266 MHz (more or less), and a datapath of 64 bits.
RDRAM has a clock speed of 800 MHz and a datapath of 16 bits.
Therefore, PC2100 has bandwidth of approx. 2.1GB/s, whereas RDRAM has bandwidth of approx. 1.6 GB/s.

Last factor: price. Prices from PriceWatch.
256meg of PC2100 - $36
256meg of PC800 - $102

-----------------
Whoever thinks up a good sig for me gets a prize :wink:
 

Arbee

Distinguished
Jun 4, 2001
305
0
18,780
And the winner is?... :wink:

BTW AFAIK programmers can't change the way memory is placed along the RAM modules - that is the job of the memory controller.

Also, I'm not so sure if your friend was correct. For a very good Rambus latency article check <A HREF="http://www.realworldtech.com/page.cfm?ArticleID=RWT110799000000" target="_new">http://www.realworldtech.com/page.cfm?ArticleID=RWT110799000000</A>


How terrible is wisdom when it brings no profit to the wise
 

Raystonn

Distinguished
Apr 12, 2001
2,273
0
19,780
"Could you give me a link? - I'm really interested on that info."

Hmm, search this forum for "FAO". It might pop up. It was probably 2 months ago.


"Actually thats not all - if you want to include AGP and PCI (or even USB) devices the number of memory channels will be quite superior - most devices have memory, such as HDDs, CDRs and even NICs."

No, all these devices go through the 133MB/sec PCI bus. The AGP bus is separate and provides 1.064GB/sec of bandwidth.


"Regarding nForce, it has a dual channel DDR controler and, AFAIK, there is no 'graphic memory slot' - but if you have other info..."

I do. I've responded to a post (possibly one of yours?) in another thread over <A HREF="http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&p=18217#18217" target="_new">here</A>. The nForce will offer the same 2.1GB/sec of bandwidth to the CPU that other PC2100 DDR motherboards do.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
 

FatBurger

Illustrious
Bandwidth is calculated by the speed (in MHz), multiplied by the datapath width (in bits), and then any other factors, like dual-channel.

DDR:
PC2100 - 266MHz x 64-bit = 17024 /8 (bits to bytes) = 2128, or 2.1GB/s, thus the PC2100.

RDRAM:
PC800 - 800MHz x 16-bit = 12800 /8 (bits to bytes) = 1600 x 2 (dual-channel) = 3200, or 3.2GB/s.


BTW AFAIK programmers can't change the way memory is placed along the RAM modules - that is the job of the memory controller.

Can someone go a bit more in depth on this? I'm just quoting what I heard someone say, but I'd like to know for sure, one way or the other.

And what does AFAIK mean? I can't remember.

-----------------
Whoever thinks up a good sig for me gets a prize :wink:
 

Arbee

Distinguished
Jun 4, 2001
305
0
18,780
A small precision (that we debated previously and I know you are quite aware of it :wink: ):

Bandwith is measured in data per a certain amount of time, usually bytes per second. In order to calculate it you multiply the amount of data send, <font color=blue>the number of times it is sent per cycle</font color=blue>, the number of cycles per unit of time (second) - AKA frequency (usualy measured in Megahertz - or million of times per second). This gives the bandwith of a single channel - if you have multiple channels and want to measuure the total bandwith then multiply by the number of channels (if all equal and non exclusive).

So, DDR (PC2100) is:
64 x 2 x 133,333,333 = 17066666667 bits per second
(note - usually it is used 133Mhz (or 133,000,000) - the clock generator is, in theory, generating 133.33(3) Mhz. In practice clock generators are never that acurate and the actual value can be slightly below (or above) 133Mhz - therefore it really doesn't make much difference)
To convert to MB/s divide per 8,000,000 (8 - for the bit to byte conversion - and 1,000,000 - for the byte to Megabyte conversion) to obtain 2133 MB/s = 2.1 GB/s

AFAIK is As Far As I Know :wink:

And I'm pretty sure no program can change the way memory is actually placed. Remember that nowadays memory management is left to Windows. Decent versions of it (2k) don't let individual programs to act on their own.


How terrible is wisdom when it brings no profit to the wise
 

FatBurger

Illustrious
Yeah, yeah, I know. Couldn't remember exactly how the conversation went before. Oh well, green grapes or red grapes, they both make wine :)

-----------------
Whoever thinks up a good sig for me gets a prize :wink:
 
G

Guest

Guest
You can manipulate where and how data goes into memory IF you're programming in lower level languages, assembly or machine code, but for applications we use that are written in higher level ,C, which is less tedious to write, limits the programmer and cannot manipulate any hardware, so the job is then passed to the OS. This is my take.
 

Ncogneto

Distinguished
Dec 31, 2007
2,355
53
19,870
If you want to get particular I believe the question was a comparison of RDRAM to DDRam. This would exclude chipsets in the comparison. Or if you like we can add the I850's memory controller ( the only real reason RDRAM actually performs as well as it does) Which I admit is probably at this moment one of the most advanced memory controllers out there ( possibly until nforce we will see). Fact of the matter is RDRAM does have less bandwith compared stick for stick with
DDRAM. For proof compare a RDRAM system with only one stick of RDRAM to a system with only one stick of DDR RAM and then see which RAM comes out on top. It would be interesting to see a chipset such as the I850 designed to run with DDRAM. Perhaps this is why Intell does not want to license the p4 bus to Nvidia?
No, all these devices go through the 133MB/sec PCI bus.
not on my board SIS 735
I do. I've responded to a post (possibly one of yours?) in another thread over here. The nForce will offer the same 2.1GB/sec of bandwidth to the CPU that other PC2100 DDR motherboards do.
Yup true. But again misleading. Is bandwith the only thing that affects performance? Hardly! On paper the nforce's memory controller is even more advanced then the I850's. Bandwith of 2.1 gig to the CPU is correct however. But with a hardware prefect and reduced latencys will/should bring quite a bit of a performance increase to the table.

Fact: All aplications are not bandwith limited. Many applications are affected more by latencey.





Video editing?? Ha, I don't even own a camera!
 

Ncogneto

Distinguished
Dec 31, 2007
2,355
53
19,870
Current RDRAM system implementations all use a dual-channel memory subsystem, plus the channel [implemented as an AGP bus] to the memory on the video card (thus three actual memory channels for the total system.)
please explain to me where the difference lie between the nforce's memory controller and the I850's. It is my understanding the the I850 is not in itself a true dual channel memory chipset either as the p4 still only uses a 64 bit path from the cpu to the memory controller itself. The only difference being that the bus is 100 mhz quad pumped resulting in a cpu to memory contoller bandwith of 3.2 gigabytes where as the athlon uses the same 64 bit path at 133 double pumped resulting in a cpu to memory bandwith of 2.1 gigabytes. So, if my understanding of this is correct neither the I850 or the nforce is a true dual channel memory system like the alpha's bus which uses a 128 bit memory to cpu path.

Answer me this. Why can't intel design a chipset just like the I850 useing DDR instead of RDRAM? They want us to beleive do to the fact it would not perform as well as the RDRAM solution. Via is starting to shed a little doubt on this theory with there single channel p4 bus,imagine what it could do if it were dual. Why is intel limiting the I845 to pc133 at the beginning? This is ludicrous! Don't they want the best for there customers? why the delay before they will alough motherboard makers to use the chipset with DDR? You trying to tell me there isnothing fishy with this?

Video editing?? Ha, I don't even own a camera!