Jump to content

Dual Channel Is Killing Me!


Recommended Posts

Hey all,

 

I've got an Abit AN8 mobo and 2 x 512 sticks of Kensington PC3200 RAM. I am trying to get dual channel to work.

 

Each stick of memory, if installed alone, passes Memtest86+ at speeds greater than spec.

 

When two sticks of memory are installed in slots 1 and 3, dual channel is disabled and Memtest86+ will pass.

 

When two sticks of memory are installed in slots (1&2) or in (3&4) then dual channel is enabled and Memtest86+ FAILS.

 

The failures fail even when I restore my machine to stock clock.

 

Any help greatly appreciated!

 

Sqopt

Share this post


Link to post
Share on other sites

failing at stock is obviously BAD! :O but the general concensus around here is that single channel is almost always more stable oc'd than dual channel is. single channel will also allow for greater overclocks most of the time. there are exceptions, but they are few and far between.

 

edit: dual does however make up for it in raw performance. you will not be able to oc single channel enough to make up for the lost bandwidth you get with dual.

Edited by upok

Share this post


Link to post
Share on other sites

have you tried setting the memory timings and voltage manually, quite often the auto settings are completely wrong and this may mean that the settings chosen by the motherboard won't allow your ram to run in dual channel...

Share this post


Link to post
Share on other sites

have you tried setting the memory timings and voltage manually, quite often the auto settings are completely wrong and this may mean that the settings chosen by the motherboard won't allow your ram to run in dual channel...

569782[/snapback]

 

The memory is Kingston Value RAM, and has a CAS of 3, and timing of 3-3-3. I candidly admit that I'm not fully clueful what those values mean.

 

With timings I have messed around practically not at all. I have the timings all auto at the moment, save for to tell the mobo to assume my RAM is DDR400 (it assumed DDR333 automatically for some reason). I have undone all overclocking for now. All voltages are stock.

 

SINGLE CHANNEL EXPERIMENT

------------------------------------------

With DIMMs in slots 1&3, it claims single channel, and when I run Memtest86+ it tells me that my RAM is running at 200 MHz, and 2.5-3-3-8. Memtest PASSES.

 

DUAL CHANNEL EXPERIMENT

--------------------------------------

With DIMMs in slots 1&2, it claims dual channel, and when I run Memtest86+ it tells me that my RAM is running at 200 MHz and 3-3-3-8. Memtest FAILS.

[Edit: the failures in Memtest seem to start happening at a particular address (remember each DIMM memtests just fine singly).]

 

So, what to do next? Should I inch Vdimm upward first? Should I mess with timing first? And if I must mess with timing, which parameter should be first changed, and how?

 

Sqopt

(who's just a bio/law geek trying to fold faster, darnit)

Share this post


Link to post
Share on other sites

Have you tried running the computer normally? Memtest86 is not infallable, it is possible that there is really nothing wrong. The program may not be reading the memory through dual-channel settings correctly, especially if both sticks pass individually.

Share this post


Link to post
Share on other sites

Have you tried running the computer normally?  Memtest86 is not infallable, it is possible that there is really nothing wrong.  The program may not be reading the memory through dual-channel settings correctly, especially if both sticks pass individually.

569798[/snapback]

 

I had indeed tried running the computer normally. The OS died an ugly segfaulted death shortly after the kernel loaded.

 

Update: I told my BIOS to set up the RAM according to "SPD" and not "AUTO" and it has passed 1 iteration of memtest. Upping Vdimm, by the way, had no effect.

 

To the previous poster: yes, the sticks of RAM are totally identical.

 

Man, this is a real puzzler, but I think it's starting to unravel. If you guys are interested in what the MOBO's by-spd settings for my RAM are, I can enter 'em.

 

Procrastinating from NYC,

 

Sqopt :foldon:

Share this post


Link to post
Share on other sites

Memtest is not infailable, it won't always detect real errors, but I have never heard reliable reports of it giving false positives. If memtest can produce the error, some other program, in real world use, can produce the error and be harmed by it.

 

My suggestions:

 

Try different dimm slots if possible.

 

Use more voltage.

 

Increase read preamble and max asych latency by 1ns each (or to 6ns and 8ns).

 

Change memory driving value.

Edited by oralpain

Share this post


Link to post
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
 Share

×
×
  • Create New...