Stable in single-channel, unstable in dual-channel

[angra]

I have a Winchester 3000+ A64 that I am working on overclocking. Rig and components described in signature (I hope). I am having some stability issues and I am a little bit new to RAM overclocking so I'm not sure how to proceed. Basically, my CPU overclocks stable up to the mid 2500s, so I had settled on 2500 as an OC target. At 278x9, with my 2 sticks of RAM installed in dual-channel mode at 1:1 I have stability issues. Memtest and Prime run 12+ hours with no failures, but doom3 and 3dmark03 left on infinite loop eventually end up crashing. All problems go away if I cut the RAM speed to the 133 setting which, I believe yields CPU/14 rather than FSB*2/3 for RAM speed like you'd normally think on this chipset/CPU. Anyway, at that speed it is rock solid stable, and at the next setting up (227 MHz DRAM clock), the stability problems are present.

However, if I remove one stick of RAM, the system is totally stable with the RAM at full speed, all stability problems are gone. 12+ hours of prime95, memtest, and 3dmark03 with no problems at all.

So I am not sure where to go from here. If each stick is good @ 278, shouldn't the pair be good @ 278 in dual channel mode? Could it be some part other than the RAM holding me back? CPU or MB maybe? Any suggestions on things to check? If I just can't get the 2 sticks to run dual channel at full speed, am I better off trading off running a single GB chip at 278 than 2 512MB at 178?

 

[Duvie]

From what I know and have been explained, that doesn't translate into real world...It is normal to do a little worse in dual channel then best in single channel. Why I am not sure. With me and others it could have to do with power, since 2 wil draw more from the same power rail. I know it is common that 2 sicks at same speed of single stick can sometimes require up to .1v more of vdimm. So you may want to try that if you already haven't..... I can get far better with my 1 stick of gskill then both. However in my case the 1 stick wont do its rated speed and the other one is a rocket and can do 620ddr cas 3.....

 

[nealh]

have you tried to change what slot you are using..ie 1&2 ..then switch to 3&4

also I think you will find performace differences minmal with a lower ram divider if you can boost cpu speed ..see Zebo thread on this

what are the ram settings you are using??

 

[angra]

Duvie,

thanks for the response. And that makes sense.

Almost a 60MHz difference (30%) seems like a big swing to call a little worse though. I suspect something isn't quite right somewhere else, but maybe it's normal.

Specs on this RAM say rated up to 2.9 VDIMM, but my mobo BIOS only goes up to 2.85, which is what I have it at. Maybe I can find a hacked BIOS to support an increase.

Thank you for the response.

 

[angra]

nealh,

I've not tried 3&4 yet, I will do so ASAP. It takes a long time for the instability to show up sometimes so I may not be able to report back an answer until the morning. That seems as good a next test as any.

I am watching zebo's thread with great interest and have even suggested an additional test for the next round of comparison. I don't think he has single vs. dual channel as a comparator, but I might suggest that, too.

My RAM settings are VDIMM=2.85 (as high as my mobo/BIOS supports), 3-6-6-15, which is about as loose as I could set the timings, 200/166/133 max MHz (misnamed divider). aggressive timing off. That's all the settings I remember offhand. Please let me know if there are other useful settings to report.

Thanks for your response.

 

[nealh]

so have you tried all the those mem dividers...and loose settings and still ahve issues?

I thought I saw some Double sided ram in dual channel is an issue.....I wish I could recall the thread...

Also make sure to try 3x htt..again to get a better idea of stability..yuo can always go back and raise..

oh yea one other thing some people were having issue with AGP fast write(I know on myBFG 6800 OC nongt/ultrya agp it casued stuttering) set it to disable....

 

[angra]

nealh,

thanks for your response.

it works at 133 divider, which equates to about 180 MHz actual speed.

I forgot to mention that I have HTT at 3x as well already.

I have AGP fast write off, and nv/ati speedup set to "disable" (was auto for a while).

I just got to try the suggested test of slots 3&4. Stability was actually worse! Would not run stable even at stock speeds. Crashes happened very soon, not after hours like other tests. Maybe a mobo issue? There is a more recent BIOS that is on my list of things to try.

Thanks again for your input.

 

[nealh]

From my experience avoid 1.41mod bios..I had Prime95 crash with my 3200 at 10x255....but it has run for 7hr and 10hrs on Prime95 on bios 1.36b...Prime95 never crashed I stopped it..

I have decided to stick with official and beta bios from MSI..no more user mods....

I have not tried the official 1.4 bios from MSI

that is awesome ram you have..I am not sure what the issue is....have you tried very low memory settings to just get the cpu max o/c

also over at extremesystems they have an OCZ guy who may have some help ideas...

http://xtremesystems.org/forum...hread.php?t=46885...he is the O.P. in this thread

 

[angra]

nealh,

thanks again for your input.

yes I bought this RAM expecting it to perform well. I am not averse to replacing a part if I can figure out what part is holding me back. I've begun suspecting mobo, especially since the slot 3/4 test.

I have used 100 MHz settings to find the max cpu OC. In the mid 2500s (usually around 2550) I start getting hard locks rather than things like BSOD, etc. So I stick with 2500, that will be good enough.

Thanks for pointer to ES. I will check over there after I finish applying this next coat of paint to my living room...

 

[nealh]

BTW the slot 3&4 is not asign of a bd mobo..they tell in the FAQ at ES that 3&4 work for some systems in dual channel...1&2 is best

check the long thread on MSI Neo2 Plat

 

[angra]

Nealh that is one long thread over there on the K8N Neo2!

I see a couple of other people have mentioned this issue in that thread, but no resolution (yet - I'm only on post #316)

 

[nealh]

yes but excellent resource site..but PM andyocz..he works there I guess and participates in the OCZ forum about rma stuff too

 

[angra]

Thanks again, will do. Waiting on account approval currently.

I am currently testing to attempt to find the max stable FSB in dual channel mode, but that could take days. I am thinking about trying a BIOS upgrade instead and testing at full speed.

 

[nealh]

I would try bios 1.36 beta...when flashing if you use win...just start msi live update and see where the the 1.4 offical bios goes...delete and rename the 1.36 beta bios to the same name the 1.4 and it will flash it

extremesystems has links to all the bioses out there

 

[angra]

I reread all of this thread and reviewed my settings. I am currently running a stability test @ FSB 220MHz dual channel so I have been thinking out the next test. About 3 hours into the test, I will run it another hour before declaring this setting safe.

One thing I noticed was that I am actually running 1.37b mod, not 1.3 as I had thought, and I notice that you avoid mod BIOS. So I think I will try either 1.36 or 1.4 or both.

Another thing I renoticed was your comment about stuttering. I am 99% sure that I have AGP fast write off, but I will recheck that later. I have noticed some stuttering myself - specifically, in 3dmark05, during the test (I forget the name) with the baloon and the sea monster, I get what seems like a stuttering about twice a second at a regular rate. This behavior was the same with a Radeon X800 so I dont think it's a video card issue.

Thanks for your input and I will pose these questions over on ES once I am approved.

 

[nealh]

I use the official beta bios..it could be just 1.41mod bios was not very good..a poor flash who knows..all I know is when I went back to 1.36b..prime 95 was stable again for 10hrs at 10x255 3xhtt, ram 133 cas 3-3-3-10 2t......with modded bios it would go 5min to 30min and crash

As for stuttering I did see a single stutter with the ballon and monster when it first started that test(with fast writes off..but then it was silk smooth at slow frame rate)...but when i had stuttering it was constant every second or 2...at ES some suggested turning off fast writes(when i switched videocards my 9700pro had a small amount but I figured that it was the card)..I was ready to return my BFG 6800 OC

with stuttering I got 3900 3dmark05..350/870
no stuttering I got 4399 with a better o/c...370/870..remember mine is nongt/ultra

I also upped my agp vcore but this appears to be unecessary..I think

 

[angra]

nealh, after my last test completed (4 hours 3dmark03 in dual channel @ 240 MHz 1:1) with no crashes I did a reboot to check my settings again, and found that I had fast writes and AGP sideband addressing on "auto" and not "off" as I thought. Do you know what those settings default to in auto mode for my board? I turned them both to disabled and launched a 3dmark test @270 MHz dual channel 1:1, so far so good, about a half hour in. Thanks for your continued input, and eventually I will take this over to ES 🙂

 

[nealh]

I think side banding is fine, I have sued with no issues..fast writes seem to cause alot of issues

if you do flash t o1.36b...the sidebanding and some options will seem to disappear..they are hidden and require the shift + f2 and then alt+f3

I am curious what vcore is your chip doing 2500 ..that is a nice overclock on a 3000+

 

[angra]

nealh,

vcore is 1.54 according to cpuz. Another odd thing I have seen with this mobo is that the base vcore setting in bios doesn't seem to affect the delivered voltage, only the overvolt %-age setting.

My understanding was that 2500 was a fairly typical maximum for 3000+ winchesters. Is that mistaken?

Last test yielded a crash so fast writes and sideband are not the (only) answer. Will try flashing BIOS when I have a few minutes.

 

[nealh]

2500 is 700mhz over stock..I got a 3200 that will top out at 2550....

so I would say you did very well

 

[angra]

Nealh,

flashed to 1.36b BIOS, no user-mod.

Overnight test @260 1:1, super-loose timings was succesful, just launched a test @270. Beginning to be optimistic at this point.