Quest for prime stable (24+h)!

[Gak]

Hi!

As all of you guys can see im new around here!

Furthermore im Portuguese! So if by any way my english is rather lousy, all that i can say is that it will improve! Right now it is REALLY rusty!



So, on in the subject!

I have a system with the config in the signature. I can provide some more details has it is using a Be-Quiet 470W P5 PSU, 2 HDD's (2x80GB Sata II Hitachi's in RAID 0 and 1 160GB IDE HITACHI) a Nec 3540 and an Audigy 2 ZS. All of wich are in an Aerocool Baydream. The CPU runs with a Zalman CNSP-7000B Al-CU.

I've always overclocked my rig's however i've never dedicated so much time to them as with this one. I would just do the regular things as "upping" the fsb, the vcore if unstable, tweaking memory timings and voltage etc's.

If it could run a bench of any 3dmark and work/play for sometime with no significant problems (reboots or "locks") that was fine for me.

However this time i made a decision to be more thorough after some quite annoying lock ups @ a certain frequency and voltage when i thought that i had the computer stable.

I started with Prime95 however after hearing about SP2004 i quickly changed to it since it is a bit more "friendly". I use it with priority @ 10.

At the beggining it would fail prime (sp2004) around 2 minutes after starting... Excelent huh?

I should by now have said that im aiming, and i was also running the system @ 2600Mhz with 1.482vcore with the mems @ 185mhz @1.5-2-2-5 with 2.6 (using obviously dividers to exclude any interference that they could cause). The 2.6Ghz were made by a 10x multi and a 260HTT with a 3x LDT.

So as all the other things were correct, so they seemed, the only solution was to up the vcore.

From 1.482 to 1.485 things got a little better however randomly between 10 minutes to 5 hours the thing would fail.

So, up with the vcore again. 1.485 to 1.507 things improved by now i could do up to 8 hours prime stable. But, it would always fail around that time. So, again, not stable.

And again up with the vcore. 1.507 to 1.508. And don't know what happened, and this is one of my questions, with this vcore, prime fails randomly but always in a short period of time. No more that 15 minutes. WEIRD! Someone could point out why?

Once more, up with vcore to 1.512 (i must say that the board seems to do some undervoltage so the "real vcore by this time is around 1.47). Priming again. This time i made up to 9 hours stable. However, and this has happened this morning i really really needed to see my mailbox, so patiently i made my way through Outlook.

However after beeing reading some e-mails for a while prime failed.

So here's the other doubt. Has the soon of a b**** failed because of me. Because i requested cpu usage and that influenced the calculation (that i thought to be like F@H, when the CPU was requested to other activities the program would idle) or it has failed just because it had to fail due to still insufficient vcore?

I sould also mention that running prime @ 1.485 with priority set to 1, it was more frequent to fail sooner. Like after ten minutes whereas running it @ priority ten it could go up to 5 hours.

What the hell is going on here!!!

Much help appreciated!!!!!

Edit:

I do not think that temps can influence much here.

I'm even running with the case open. The CPU configured this way runs @ full load @ 37/38º degrees max! And the HSF isn't even @ full rpm's since the bios is configured to do so just if CPU gets up to 42º. Therefore the HSF sit's around 1500rpm's. The room temp i do now know, but with no heating has is it now i have to use a coat to be in here.

The computer @ stock voltage idles @ 20º so imagine...

 

[nealh]

first off the dfi mobo actually overvolts underload..winxp apps report vcore too low

anyway..set ram to cas 2..why are you using cas1.5??

rerun prime 95...ram timings are not that important since the memory controller is on the cpu...

the best way to test cpu max speed is set ram to loose timing and ram 100/133 divider
so set ram 3-3-3-8-1t w/100 or 133 divider

then determine max cpu speed by running small fft on Prime95..then once you determine cpu speed go back and tinker to max. ram setup

max vcore I would use is 1.55v -1.6v

also on the dfi mobo you need to set pcie freq to 105 for ati videocards....

great overclocking inof at dfi streets

http://www.dfi-street.com/forum/forumdisplay.php?f=9

forum is down ...but check this website

 

[Shimmishim]

at first it sounded like a heat problem...

upping the vcore increases heat and usually running prime for about 2 minutes is an indication of heat problems...

but it's strange you were able to make it 9 hours or so after increasing it even more...

i agree with nealh that you should just release your timings to 3-3-3-8 or 3-4-4-8 as to completely eliminate the mem as the problem.... 1.5-2-2 is pretty tight even if it is running at only 185 mhz.

 

[Gak]

I've never had a voltage reader (multimeter is the name isnt it?) but i was thinking on getting one (for years... )

I'm using MBM with a config that is avaliable @ DFI Street for this mobo. So that's were my reports come.

And i'm not trying to max out the CPU. That i have done already 2951mhz with CPU-Z verification and 3003Ghz however unable to save the screenie since it rebooted. But to do that i've almost had to go up to 1.8V and that isnt a really good, cool or whatsoever thing to do, therefore im not repeating the joke so soon...

What i want for this particular rig is a good (at least in my opinion) co-relation between speed/temps/voltage. So 2600Mhz seems to me to be the sweetspot (im not yet trully convinced. After beeing able to prime24h i'll do it all over again, however @ 250 or even lower HTT with tighter ram timings. Just to compare frequency/timings.

These are TCCD rated to work @ 200Mhz stock 1.5-2-2-5 (even that A64's don't use anything under 2) so that's why i left them @ that speed.
And i've not understood very well what you said (neahl) about those ram dividers... At 100/133 the ram would be going even slower that 185 as for now i'm using a 150 divider. What's the point then? Isn't this slow enough?

I will let him prime this night @ stock settings. CPU, and mems just to exclude any software related problems.

Tomorrow i'll post some info about it.

I am almost a daily reader of DFI street. Since i own one! And believe me i've never came across that info about ATI cards! However as soon as possible i will certainly read the thread. This could be also due to some lack of interest on the card. I''ve not been so lucky with it. And when others can get up to 600mhz on core this one completely "maxes" out @ 517mhz. With a VF-700. So for quite a while im thinking on getting a 6800GS. However just yesterday i finished my mid-term exams so hadn't got some free time to think about it...

 

[nealh]

the purpose of the divider is to remove the memory from the equation...please try it at 133 divider..do not use an odd divider like 150, the cpu may not like it..100 and 133 are traditional dividers

also setting ram to cas 1.5 is not always going to work well with all bios and mobos..even though it is rated at cas 1.5 this is"odd" setting and again issues can arise

please try my suggestions

also I am not talking abouit getting a max cpu speed for screenies but prime 95 stable max, by lowering ram you remove this variable from the equation

also set the LDT voltage to 1.3 and chipset to 1.6v..also recommend by Angrygames(mod/employee of DFI) at DFI street to help make things more stable for overclocking

all you want right is to find max stable cpu speed for prime95 and it may not be 2600!

DI street is down..but once I set ATI freq to 105 my X800XL overclocked better without artifacts in ATI Tray tools

I wish I could remember the thread that was recommended

 

[Gak]

About those voltages... I already was aware of that suggestions however has i said, i like to keep them whenever possible @ stock...

And it never occured me about the "odd" setting with the 1.5... Nice point!

I'll perform your suggestions tomorrow, all of them including the voltages. As i said overnight i leave it priming @ stock. Just to be shure!

And a big thanks for the help!!!

[[]]



See you tomorrow. Now BED!

Im almost dead!

 

[nealh]

Originally posted by: Gak
About those voltages... I already was aware of that suggestions however has i said, i like to keep them whenever possible @ stock...

And it never occured me about the "odd" setting with the 1.5... Nice point!

I'll perform your suggestions tomorrow, all of them including the voltages. As i said overnight i leave it priming @ stock. Just to be shure!

And a big thanks for the help!!!

[[]]





See you tomorrow. Now BED!

Im almost dead!

I agree I like stock voltages too

Angrygames state the LDT voltage often at stock is less than 1.2v..mine reads like 1.17...so he recommends the small bump to make sure you get 1.2v
as for the chipset it does little to temps 1C or so but supposedly helps with the higher htt

 

[Gak]

Well, just "finished" 13 hours priming @ stock. Not a single hitch.

As my max at 2600mhz where 9hours it seems evident that the problems arise from the configuration.

Tonight i'll then try with the config you have suggested.

However, wich vcore should i use now? Start all over again @ 1.482? I guess it is the best since all the other "tests" seem to be compromised... Right?

And you have suggested early to use smal fft's. I have always been using small fft's. Precisely to stress the max out of the cpu leaving the mems aside.

 

[nealh]

try a vcore that works..you can try lower later

 

[Gak]

Well...

Something strange happened.

When i got to the computer a while ago windows was at is logon screen. I figured, the light went out meanwhile in the night. However when i entered windows i was welcomed with the "the system has recovered from a serious error" screen.

This is what i have in the sp2004 logs:

Self-test 24K passed!
Test 1, 270000 Lucas-Lehmer iterations of M376833 using 28K FFT length.
Test 2, 270000 Lucas-Lehmer iterations of M372735 using 28K FFT length.
Test 3, 270000 Lucas-Lehmer iterations of M368641 using 28K FFT length.

Self-test 20K passed! [Tue Feb 28 10:33:30 2006] Self-test 24K passed!

I'm not quoting the entire log obviously... Just where it ends.

I've followed all neahl sugestiongs. Only except the ram where it was configured to a 166. Wich i guess to be also a tradiotional divider, right? At 100/133 i wans't unable to boot to windows with a blue screen with PFN LIST corruption error.

What the heck may have happened here???

 

[nealh]

Prime 95 crashed and your system is set to reboot on this error

you are going to need to lower the overclock or increase the vcore

are you saying the system will not boot with the ram set to 100 or 133 divider? what mhz is the ram at with 166...

You may need t ostart all over...also possible the xp install is borked by the crashes ..do you have a backup image prior to all the testing

I would restart...set vcore to auto and set cpu to default..boot and system up..

try maybe 2400 at stock and prime 95 for maybe 30min to 1hr and see hwat happens..then move slowly up

3500 is 2.2 ghz at default? you cpu may not be able to do 2600

 

[Gak]

At 166 its @ 216Mhz.

I gave it 2.7v just be shure that they were receiving more than enough juice.

And yes i have a Ghost image.

2400 at stock?

Stock it's 2200...

What a hell of a crap of a cpu i got here, han?



I'll try a new install again...

I'll be back.

 

[nealh]

I am hoping it will oc at stock to at least 2400...but you will need to go slow and I would recheck ram timings maybe the mobo is setting it wrong??

OCing can be a real pain in the rear at times

 

[Gak]

Once again...



Fresh install, because the old image i had was for IDE. I forgot i am using RAID 0 now... Upgraded a couple days back.

The system is running at 2.4Ghz stock voltage.

10xMulti, 4xLDT, 1.4V. Ram with 166 divider running at 200Mhz with 2.6v and 3.3.3.8 timings. All other (possible) definitions have been set to auto hoping to get the maximum stability.

I'm going for a walk now and i'll leave it priming for one hour, our and half, pherhaps longer if i take more time...

C ya!

 

[Gak]

Here i am again!

Went for a walk, bla bla bla



SP2004 ran for 2 hours with no errors.

What would you suggest i should do now?

I was thinking in pumping up the HTT till it doesnt boot anymore. Then giving it more vcore and testing prime till it fails in the first 15 minutes. Afterwars i was thinking in using that formula that is in the XS foruns (if im not in mistake) to "find" a max stable HTT (HTT*095) and after getting that value, then tweaking it to the max to get the 2600Mhz.

What do you guys think?

 

[nealh]

nope..I would try 2500 at stock and prime again for 30min to 1 hour....if it fails give it a small vcore bmp..if not try 2550 not 2600

I expect that you may fail prime in a few min...just small vcore bump and re-prime

 

[Gak]

Hi again!

@ 2500Mhz stable for one hour.

Moving to 2550Mhz.

Still at stock voltages and with 166 divider.

 

[nealh]

great..you may want to run prime 95 for 4-5 hrs this time to see how stable you are

doing well

 

[Gak]

Not stable at all.

Fails after 1 to 3 seconds.

I'm guessing that if i left him @ 2500mhz priming for more that 1 hour it would also fail.

 

[Gak]

Well,... Ill leave him priming overnight with stock voltage @ 2500Mhz. Just to see what happens...

 

[Gak]

Primed for 8 hours 2500mhz. Stock voltage. Seems stable.

 

[Gak]

New bios flashed.

704-2BTA

Seems more stable @ 255HTT.

Still fails within first minutes but no longer @ first seconds.

Bumped the voltage to 1.425. Ill let him prime for a couple of hours now. If he manages to...

The rest of the settings maintain the same config.

 

[nealh]

Good things are looking better

FYI: the DFI mobo will overvolt underload so just note the vcore is higher than bios setting at load...seems to run for most 0.025 or so