Finally a stable Overclock...

[jstarman]

A big thanks to the sticky overclock guides and fine discussions on this Anandtech forum.

I've been having some difficulties with my overclock and believe I finally have a stable one. So best to describe the method I used, the results and where I'm currently at.

My set up is below in my signature. According to Zebo's A64 giude, my maxium FSB is 277 x .95 = 263.15. Maximun CPU is 265 x .95 = 251.75 . Maximun memory is 220 x .95 = 209.

To consolidate, I chose to run a 5:3 divider HTT x .83 and used 252x10 1.5V. Memtest said DDR 420 210Mhz, BIOS auto settings which used 2.5-3-3.7 and disabled 2T. Failed Memtest test 6 @ 844.7mb - 1 error. Then changed settings to 2.5-3-3-6 and failed Memtest test 6 844.6mb - 1 error. At that time I was using auto voltage for memory. So I changed Vdimm to 2.7 and memtest failed again on test 6 @ 844.7 mb after the 4th pass. Tried 250x10 5:3 divider for 208 Mhz and failed test 6 @ 884.7mb - 1 error on the 9th pass. Then tried 2.8 Vdimm ...failed. At this point, I decided to run Prime 95 and failed after 6 hours with the rounding error. Tried a bunch of different settings with similar results.

Ok... after more reading, I decided to try a 3:2 divider. 250x10. CPU/Z showed 178 Mhz CPU/14. This setup ran Prime stable for 24+ hours. Ok, now I'm making progress. U Ugggg... Memtest failed on test 4 @ 844.7mb & test 6 @ 884.7mb. 264x9.5 failed Memtest test 6 @ 605.mb. 264x9.5 5:3 divider Vdimm 2.8 2.5-3-3-7 failed Memtest test 6 884.7mb on the 11th pass. 260x 9.5 3:2 2.8v 2.5-3-3-6 failed Memtest test 6 @ 884.7 mb on pass 16. Ran Prime 95 on that setting and ran for 24+ hours no errors. Making some progress here. Used same setting but set Tras to 7. On BIOS auto, it would set that to either 6 or 7 and memory rating is 8. Failed Memtest test 7 @ 884.7mb on pass 3. Changed to 2.5-3-3-8 failed test 6 @ 605mb test 2. Then changed Vdimm to 2.7V and Memtest ran for 33 passes with no errors...OMG it passed. Ran Prime 95 again for 24+ hours with no errors. So I decided to make sure my memory was stabel, I ran Memtest 1 more time and it failed on test 3 @ 884.7mb on the 19th pass out of 33 passes. Ugggg. I'm now running memtest again and so far I have 21 passes without errors.

On Prime 95, I used the blended torture test #3 and on Memtest, I used cache on and ran all 8 tests on each pass.

Sorry for the lengthy post but I've been at this for 2+ weeks and I'm not sure if I have a stable system. There were many other combinations that I tried which I didn't list here.

So here's what's working: 260x9.5 (2470 Mhz)1.45V 2.5-3-3-8 1T 2.7V at 194 Mhz according to memtest & 190 Mhz CPS/13 according to CPU/Z.

So my concern is that I did get that error on the second run of Memtest however the 1st run was error free after 33 passes and currently memtest is on the 22 pass without errors.

I'm not familar with the errors on memtest and what they mean. The 2 reoccurring errors were at 884.7mb mostly and every once in a while at 605mb. From what I can surmise, my memory wants to run at stock settings of SPD 2.5-3-3-8 2.7V. So where do I go from here? Is my system basically stable? Or do I have an issue with my memory? I will continue to run memtest overnight which will give me at least 50+ passes.

What's most important to me on this computer setup is CPU speed as I'm using 2D financial charting software with custom algorithms that are quite CPU intensive.

Am I missing something here with my process? Any suggestions or comments are greatly appreciated. I wouldn't have gone this far without all of the fine discussions on this forum. Many thanks!

 

[jstarman]

Failed again at test 6 on the 27th pass (1 error). This address 000374cccec - 884.7mb has been the consistent failing address. Now on pass 32.

 

[Duvie]

One thing is that iin my testing that test #5 and #6 are most prone to errors....Run memtest with cache on but then run just test #5 for 100 passes (about 90 min) and then do that for test #6 (another 90 min)...

These will help save you the tims since the other test rarely fail..I run all test for 10 passes max and then do the individual test.....

Could be a flaw within that stick if it happens near the same point each time...

 

[kuester01]

My set up is very similar, same mb and cpu,but I run OCZ EL plat rev2 1gig kit. I have tried everything, I was even in contact with a great tech guy at OCZ to help me get more outa this. I can get my system to pass memtest at up to DDR 484 speeds several passes. It will run prime 95 for a few hours at a time( have not over nighted it at this speed), but when I game, I get the blue screen.

From What I understand the didvider just isnt doing the job properly, I can run at 2510 with ease (251X10) but if I get that memory running higher than 208 no matter what the timings are, Ill will get the Blue screen.

I am very tempted to get the MSI Neo2, as I feel its a shame for this ram not to be able to stretch its legs, but damn--




Kuester

 

[jstarman]

Thanks for your suggestions Duvie.

I ran test #5 for 100 passes with no errors. (1 hour, 45 mins.)

I ran test #6 for 138 passes and had 2 errors at the same address. (000374cccec - 884.7 same spot as earlier errors). 1st error was on pass 16 and 2nd error occurred somewhere between pass 28 & 112. Test #6 took about 11 hours to run.

Based on these results, what do you or anyone think?

Thanks!

 

[jstarman]

I just finished running memtest on each memory module separately in DIMM B1. Used the same settings as above.

Ran 10 passes of test 1-8 & then 50 passes of test 6 with each module.

NO ERRORS. Man I have no clue. What would cause me to get that 1 consistent error @ 000374cccec - 884.7mb when running the 2 modules together in dual channel mode?

I'm going to stick both modules back in and run test 6 overnight. Ugggggg!

 

[Duvie]

swap their places!!! put the 2nd one in the first slot and vice versa....see if the error happen in the first 512mb range then....


it is not uncommon since dual channel is harder to run .....most all single sticks can do better alone....that basically emans try to loosen up timings ever so slightly...what I mean is sometimes that slight error can be cleared by going 2.5,3,3,7 forma 2.5,3,3,6....vry slight but can make a difference

Also do the benchmark testing...you may find out like many that the better score sometimes can come from a higher last number....ie 2.5,3,3,6 was slower then my cas 2.5,3,3,8 (which was more stable as well) in superpi 32mb testing....

 

[jstarman]

I did exactly that...i swapped places... stick 1 in DIMM B1 and stick 2 in DIMM A1 and 45 passes of all 8 tests. Had an error in test 7 on pass 2 @ 00028b73314 - 651.mb. So far after 45 passes it found 2 more errors at the same spot for a total of 3 errors.

Any other suggestions. I have no clue as where to go from here. I'm assuming at this stage my system is unstable?

"see if the error happen in the first 512mb range then...." ...what did you mean after then?

 

[Duvie]

I meant see if it was related to one possible stick....It wasn't though 651mb would have put it on the second stick as well even though you swapped their places....I would say...

1) bump up vdimm a notch
2) adjust the timings slightly
3) bump up the vcore a notch...

 

[jstarman]

after i posted, i bumped up vcore to 1.475 from 1.45 and am now running memtest again. If I fail again I will adjust the timmings and or vdim as per your suggestions.

I posted my results to ram guy on corsairs website and here is what he said,

"You should really be using a 450 Watt PSU with this MB especially if you have a newer Graphics card ATI 9700/NVIDIA5900 or newer. And the errors you are getting at not consistent. If you have a failing module they would fail at the same address on every pass usually on test 5. Getting the errors like you stated 2-3 errors every several pass on test 5, 6, or 7 would suggest some other issue and the most likely cause of the type of errors you report would be a weak or noisy PSU."

What do you think?

 

[Duvie]

Hmmm..I thought you were getting consistentcy at first with error occurring at same address...

I dont necessarily agree with their PSU response....Their are many other factors that are related to power....

They are right that test #5 is the best test for testing amd systems and ram....

I had some weirdness like that until I found the sweetspot and I have a 600watt OCZ...It is always easier to pass the blame on other components....though I think the ram is likely fine you just have some volatge issues....vdimm would have been my first choice instead of vcore....

 

[jstarman]

the failing addresses have been consistent at the same address. During some of my overclocking setting adjustments, there were a few errors during other tests as i posted earlier.

The only reason i tried vcore first was because when I dropped vdimm from 2.8 to 2.7 was when I had my 1st 24 hour period of prime 95 without errors. But that could have been due to going from 5:3 to 3:2 divider. I have to check my notes.

thanks!

 

[Duvie]

It could have been yes...the divider reduced ram speed....However you are right some of these chips do not like the high vdimm is what I am finding out myself....

I actually gained stability at a certain speed by dropping the vdimm from 2.85 to 2.7v...It seems my errors were possibly heat related at certain cas timings.....

I thought maybe you were still at 2.7v so I suggested taking up a notch which would have been 2.75v...I wouldn't go over 2.8v for these TCCD chips...they may not like it!!!

 

[jstarman]

I am still at 2.7v but this A8V board only has 3 vdimm settings if I remember correctly, 2.6-2.7-2.8... oh and auto. If I get any errors on memtest during this run, I will try 2.8v again or maybe 2.6v as well. Like you said, it may be a matter of finding the sweet spot.

 

[jstarman]

I posted this to the ram guy,

"I'm still not clear why the errors occur at the same exact address. Doesn't that suggest a bad spot on the memory?"

Ram guy's response: "If you had a bad module you would be getting erors on test 4 on every pass."

 

[Duvie]

So wjhat is he suggesting?? Test #5 is not testing the module but the memory controller and somehow these other factors can randomly keep picking the same place in the modules??? I think not...Sounds like BS to me....

I had a bad stick confirmed by GSkill and it didn't error at all in test #4 but had errors in test #5 and #6 and maybe a few in test #8.....no test #4...

 

[jstarman]

Decided to change Vdimm to 2.6 instead of 2.8... Memtest ran for 28 hours with 105 passes of test 1-8...NO ERRORS! Awesome. Guess it was too much juice at 2.7v & 2.8v that caused that weird error. Now I'm going to run Prime 95 again for at least 24 hours. Hopefully it will pass with flying colors since it passed with these settings and 2.7vdimm prior.

Thanks so much for your help!

 

[CaBoOse999]

Yeah thats strange how that works. The same thing happens to me.

 

[Duvie]

Good Job...This can be a very helpful threads for others following this....

 

[jstarman]

Prime 95 ran for 25 hours NO ERRORS! Way kool.

I'm going to leave the settings as is so I can put this computer to use. However, I'm sure I can overclock some more now that the memory issue is solved.

 

[jstarman]

Had some time this week so I tweaked the settings some more... see my sig for the final results. I'm not maxing out yet but my goal was a 25% OC so I'm way happy.

Thanks to Duvie for all of your help and a special thanks goes to Zebo for his awesome OC stickies! Read them and reread them and reread them a third time.