has the 4 dimms @ 333MHz Athlon64 problem been solved with the AM2 generation?

[R3MF]

i have a s939 board whose memory i cannot double by sticking in two more dimms because the FSB speed will drop from 400MHz to 333MHz.

does this problem persist with AM2, or can I get 2x 1GB DDR2-800 dimms safe in the knowledge that i can stick a further two in the board later on with no silly performance penalties?

regards

[edit] please not the revised thread title [/edit]

 

[Peter]

It's not a problem, it's laws of physics at work. Bus speed will always /have/ to decrease as load increases.

 

[betasub]

Many Athlon64 motherboard BIOS will default to 333MHz when using 4DIMMs because that is the design specification based around full compatability with the early Athlon64 revisions. Because the memory controller is on the processor, not the motherboard, it isn't a limitation of the motherboard. In fact, many Athlon64 processors (particularly Rev.E) can handle 4DIMMs at 400MHz.

Are you able to set memory speed manually in BIOS to 400MHz with 4DIMMs? Have you confirmed whether the 4DIMMs limitation is a result of your mobo/BIOS or a limitation of your X2? And just how big a performance penalty are you afraid of? Memory bandwidth is rarely a limiting factor in Athlon64 performance.. can you run with 1T and low latency settings?

 

[R3MF]

sorry if i was unclear, i was really asking if this limitation has persisted into the AM2 generation?

 

[Peter]

Sorry if I was unclear. I was really saying that these laws of physics have persisted into the AM2 generation.

 

[R3MF]

shame, i object to having to use equipment at a degraded performance, in the same way i avoid video cards with disabled pipes, it sits wrong with me.

 

[R3MF]

so, is it confirmed that using four DDR2-800 dimms in a nVidia nForce5 chipset when using dual-core A64 results in a FSB speed drop to DDR2-667 speeds?

 

[yacoub]

Originally posted by: Peter
Sorry if I was unclear. I was really saying that these laws of physics have persisted into the AM2 generation.

oh come off it already. they can build a board that allows four DIMMs to run at 400MHz. If they haven't decided to allow that for the public, that's one thing. Claiming there's some sort of physics law that prevents it is sheer fanboyism.

 

[R3MF]

this is my point, i realise that some m/b makers have got round the problem on s939, but most have not.

i merely want to know if the problem persists into the AM2 generation.........................?

this question is proving to be a real struggle!

 

[BassBomb]

Originally posted by: R3MF
this is my point, i realise that some m/b makers have got round the problem on s939, but most have not.

i merely want to know if the problem persists into the AM2 generation.........................?

this question is proving to be a real struggle!

not exactly

revE as stated above can handle 400mhz x 4 dimms.... any other wont be able to do it, even if the motherboard can

 

[R3MF]

(on the presumption that you are still talking about s939, and in the knowledge that my K8NGM-FID will not run 4x dimms at DDR400 speeds)

I am talking about AM2, not s939, i couldn't for the purpose of this question give a toss about s939.

 

[BassBomb]

well do you expect amd to revert back to things pre rev E in their technology ?
that would be smart...

im gonna guess and say that it doesnt happen on am2, depending on the board

 

[R3MF]

this is a first-gen DDR2 memory controller for AMD, so anything is possible.

i am merely trying to confirm from users of AM2 systems and those otherwise knowlegable on the matter as to whether the problem exists in AM2 systems.

i do not know, therefore i ask, and while i appreciate your logical surmise' they do not really answer my question definitively.

 

[Steve]

Originally posted by: yacoub

Originally posted by: Peter
Sorry if I was unclear. I was really saying that these laws of physics have persisted into the AM2 generation.

oh come off it already. they can build a board that allows four DIMMs to run at 400MHz. If they haven't decided to allow that for the public, that's one thing. Claiming there's some sort of physics law that prevents it is sheer fanboyism.

LOL! No it isn't. Go to www.jedec.org and do some reading. Bottom line: getting it to 'work' is out of spec.

 

[Peter]

DDR2 does have some more headroom in signal integrity, but the DDR2-800 speed grade is pushing it already.

Like it or not, signal integrity on a bussed signal IS a function of frequency and load - raising either makes it worse, lowering makes it better. Load is chip count, connector count, and trace length.

Rev. E socket-939 processors could run /some/ DDR400 setups with four DIMMs, but only at degraded timings. Laws of physics still apply.

Mainboard makers have to come up with something that works for everyone, with DIMMs that /just/so/ fulfill the JEDEC specifications, for the entire range of temperature and voltage fluctuations. If you, in your individual setup, choose to use better material, a narrower temperature range, and maybe a raised voltage, then you /might/ get better results. You're free to try, but this is definitely YMMV territory. Just like it was with DDR1 technology - the specification /promises/ DDR400 operation with one DIMM (per channel), DDR333 for two, DDR200 for three or more.

 

[betasub]

Originally posted by: Peter
Mainboard makers have to come up with something that works for everyone, with DIMMs that /just/so/ fulfill the JEDEC specifications, for the entire range of temperature and voltage fluctuations. If you, in your individual setup, choose to use better material, a narrower temperature range, and maybe a raised voltage, then you /might/ get better results. You're free to try, but this is definitely YMMV territory. Just like it was with DDR1 technology - the specification /promises/ DDR400 operation with one DIMM (per channel), DDR333 for two, DDR200 for three or more.

QFT. But to return to my question to the OP, what is preventing you from running your current set-up with 4DIMMs at 400MHz? Is there a hardware limitation, or are you unwilling to try an "unsupported" or "out-of-spec" BIOS setting?

 

[R3MF]

ok, i'll try this a different way:

has it been stated by AMD, or discovered by anybody, that running 4 dimms of DDR2-800 will result in the bus speed dropping to DDR2-667 speeds?

or for that matter; that using 4 dimms would cause said dimms to run at CAS5 instead of their normal CAS4 rated performance?

and from here on in let there be no more discussion of DDR400 on the s939 platform!

 

[REMF]

you tell them!

 

[customcoms]

This is what I know about the AM2 memory limitations:

A single core AM2 processor can only use up to DDR2-667; this has been a problem for people who bought DDR2-800 and a single core AM2 processor and then it wouldn't boot; G. Skill took DDR2-800 back and reprogrammed the SPD's to show the ram as DDR2-667, even though the ram is guarenteed to hit DDR2-800. Overclocking works around this limitation.

A dual core AM2 processor is rated up to DDR2-800 with the possibility of DDR-1066 memory dividers in the near future (similar to rev. E processors and their ability to "overclock" the ram independent of HTT speeds by running a divider higher than 1:1 (>DDR400).

Based on this information, it would appear that the dual core processors have a superior memory controller, or AMD is just limiting the single core processors because they feel that they don't need the extra bandwidth afforded by DDR2-800 and above. However, how this limitation is implemented is not very clear. Most AM2 boards are still in their early stages of developement, and as most of you know, DFI has yet to release their AM2 boards (well, they have one infinity available at ewiz today); this means their is still the possibility of motherboards+bios to overcome this single chip limitation.

As for the four dimm limitation, I doubt it will be as limiting as the S939 limitation (which if you have a good board (i.e DFI Lanparty or Asus A8N), is not really a big deal, it just forces you to use a command rate of 2T when using four dimms, they still operate at DDR400 and the same timings as with just 2 dimms), because this is a fourth generation (S754>S939>S939 Rev. E>AM2) memory controller and AMD has learned from its other 3 integrated memory controllers!

Why has no one (including myself) answered the question directly? Because of two reasons:

1) Hardly anyone has an AM2 setup, because their was ZERO real world performance gain over S939; this will change as the platform matures, but people who have decent S939 setups have had no reason to switch to AM2, especially with the release of conroe.

2) Those who have AM2 setups don't have four dimms of DDR2 lying around to test this. Why? They more than likely bought 2x1gb sticks when they bought their rig. Their is no real need for more ram atm (unless your pixar, but then you have quad dual core opterons with 32gb of ram).

In short, the only people that can test this are hardware sites who have been too busy running conroe benchmarks and X2 EE cpu tests to bother with this question. Anand has stated that their new memory platform is conroe!

 

[R3MF]

many thanks, best answer by far.