Dual Channel DDR SDRAM vs Dual Channel RDRAM: Who wins?

[NFS4]

From what I understand, NVidia is working on a dual channel DDR chipset for Athlon/Duron systems. How do you think that this would compare to the current dual channel RDRAM solutions on the market?

 

[Noriaki]

I think it'll cost alot....
That's 96 more data traces than Dual channel Rambus needs....

All other things being equal, I suppose Dual PC1600 should perform a tad better than Rambus because it's lower latency in it's basic form.

But really do we need either for an Athlon? PC2100 gives the Athlon a full FSB...

 

[Pabster]

Dual-channel PC2100 would blow away Dual-channel RDRAM. The latency is the key, and DDR is much, much faster than RDRAM. I don't think it would help current Athlons, but future CPUs will be more suited to the extra bandwidth it offers.

 

[Degenerate]

perhaps there is not much use in the Dual DDR?

 

[fkloster]

<< Dual-channel PC2100 would blow away Dual-channel RDRAM. The latency is the key, and DDR is much, much faster than RDRAM. I don't think it would help current Athlons, but future CPUs will be more suited to the extra bandwidth it offers. >>



Once agian, our new friend Pabster is talkin' out his arse! Dual-channel PC2100 will be blowin' NOTHIN' away until AMD spends some r&amp;d on a cpu that can take advantage &amp; utilize even SINGLE channel PC2100 (...better known as PC-133 DDR)



<< The latency is the key >>



What the hell do you know about the future latency of a product that does not exist? What the hell do you know about current latency in an i850 platform that all but eliminates any latency penaltys know of rambus on its own? (wether or not i850 conceals the latency or cleverly hides it is not an issue)



<< and DDR is much, much faster than RDRAM >>



Provide me ONE memory benchmark that isolates a rambus platform (i850) in coparison with a ddr platform in memory only that shows DDR trouncing the i850?

 

[idgaf13]

Go to The Register do a serch for Rambus.
You will find all the background info you could ever want with excellent links.
Rambus is to hard to make and requires new machinery to make.
Intel gave manufacturers of memory money to Rambus.
DDR is easier to make ,is not proprietary ,does not require new tooling of factories to make.
DDR is faster in its 2100 type ,easier to make.
Support Rambus and you support an Evil Empire worse than microsoft.

 

[idgaf13]

Wow what logic ,fkloster.
you must work for someone involved in this discussion ,why get so mad ?
you accuse someone of posting inappropiate info but provide nothing in return except a bad attitude.
Do work for someone involved ?

 

[fkloster]

I'm not mad. Just frustraded with some people new to the forum...speculating and making outlandish statements about products that DO NOT EXIST bother me...

 

[shathal]

Latency isn't always the key, btw. (To Pabster)

Some things are bandwidth dependant (RDRAM wins). Others (such as for GFX-cards atm) are heavily reliant on latency.

You can't win both - not with current memory technologies at any rate. Which is a damn shame :|

 

[Adul]

The question is, can a cpu use more bandwidth then it was designed for? Take the p4 for instance, could it use lets say a dual channel PC2100 DDR SDRAM system, to its full advantage? Even though the bus can only provide 3.2 GB? I prefer to match the bandwidth to the what the bus can handle unless there is some way to squeze more efficiency or drop the latency by going with a dual channel design.

 

[ToBeMe]

{sarcasim} Oh yeah.....The Register is a great place to look for accurate information {end sarcasim} LOL!

Flockster is correct!🙂 While I do NOT care for the company Rambus..........RDRAM has a LOT of future potential much more so than DDR.😉 DDR is fine and a nice step up from SDR but future high frequency CPU's will benifit much more from RDRAM than they could from any DDR compliment!😉

 

[andreasl]

Alot of people claim that DDR chipsets have much lower latency than RDRAM chipsets. But few, or none, actually provide evidence to back this up. It's very hard to find 'benchmarks' that provide this comparison. But I remember one from Aces Hardware running cachemem on the i850 and 760 chipsets.

Memory Latency

Have a look at this graph. At first glance it looks like a clear win for DDR. BUT cachemem measures memory latency in CPU clock cycles, not nanoseconds. Since no numbers are provided we have to try to extract them from the graph:

RDRAM ~ 270 CPU cycles @ 1.5GHz = 180ns
DDR SDRAM ~ 220 CPU cycles @ 1.2GHz = 183ns
SDR SDRAM ~ 260 CPU cycles @ 1.1GHz = 236ns

P3-800 ~ 125 CPU cycles @ 800MHz = 156ns

I presume the P3 used PC133, probably on the i815 chipset.


Where does the DDR chipset 'kill' the RDR chipset when it comes to latency? To me it looks pretty even in this particular benchmark.

Just saying that DDR chipsets have lower latency than RDR chipsets is misleading. This is because latency is depending on not only the memory modules, but the FSB and chipset logic as well. A fine tuned RDR chipset can beat a poor DDR chipset (not saying 760 is poor). Just compare the early revision ALi MAGiK with the 760 for example. Or the reports of the SiS 735 chipsets compared with the other DDR chipsets.

Does anyone know of other comparisons made between these chipsets? Preferably made by people who know what they are doing...

 

[fkloster]

Pabster, you need to pay attention to this... rdram, by itself, means very little. But considering the computer world is moving in the 'serial' direction, rdram has a very bright future in PC's as long as chipsets similar to the i850 exist.

 

[Pabster]

Why can't we have a legitimate debate here without resorting to the usual fanboy vs fanboy attacks? I'm not against RDRAM, as a memory architecture. I dislike the company greatly, but that's neither here nor there. Let's stay on topic. The question asked was which dual-channel memory architecture &quot;wins&quot;. I answered DDR SDRAM, and I'll stand behind that. RDRAM *does* have extremely high latency, and when paired with CPUs other than P4 offers near zero benefit at extreme cost. There's no point in it. The P4 cannot make use of 3.2GB/s bandwidth, and current Athlons cannot make use of 2.1GB/s bandwidth. So it's pointless to start debating dual-channel memory architectures, until the CPUs are ready to support them.

Now, everyone here has a different opinion. That's fine. Just remember that everyone is entitled to their own opinion, and leave it at that. You don't have to start a fanboy attack ...

 

[fkloster]

<< RDRAM *does* have extremely high latency, and when paired with CPUs other than P4 offers near zero benefit at extreme cost. There's no point in it >>



1) What do consider extremely high latency Pabster?
2) Why pair rdram with any other chipset than i850? i850 was designed for rdram... does that bother you?

 

[Pabster]

fkloster:

Here's an interesting quote from the same article you tout as &quot;proving&quot; that i850 latency isn't &quot;worse&quot; than DDR solutions:

&quot;Choosing RDRAM (Rambus) for the Pentium 4 was, from a technical point of view, not a bad decision. Concerning bandwidth, Rambus is more efficient than DDR SDRAM: in reality it can offer 50% of its theoretical bandwidth, while DDR offers only 42% of its theoretical peak. DDR SDRAM, however, provides significantly lower latency.

I suppose that isn't valid, of course.

As for pairing RDRAM with anything but P4, I agree. When did I ever say otherwise? I laugh at the idiots purporting RDRAM should be coupled with Athlon, and I think we can agree P3+RDRAM is a joke. P4 was designed around RDRAM; so naturally it makes some use of it. Other processors see little to no performance improvement, and the cost is ridiculous.