Stress Testing

[james1701]

When you stress test your system after you change your settings, do you surf the net, or watch movies or anything while you have Prime95 or OCCT running? Or do you go do something else or have another computer running to do these other things to pass the time?

 

[nyker96]

LinX.

 

[mindless1]

The way you've worded the question it seems to imply you are changing important parameters on a primary use system. I generally build it, test it, and after it's been tested only then would it be considered as a replacement for my primary use system.

If you just mean misc. upgrades to your already used primary system where it was already known stable except for one subsystem changing like memory, let Memtest86+ run overnight at 5% higher o'c than my target. Video, let 3Dmark and ATITool run while I'm away from the system doing something else. CPU, I don't usually upgrade CPUs anymore on the same motherboard, but again I'd start the test when I was ready to get up from using the system and about to go do something non-computer.

Basically it's not as though you could use the system for something else in cases like memtest86+ or 3Dmark looping, and in other testing you don't really want to use the system for something else because if there's task switching going on you may be causing the system to spend more time on a less stressful task which decreases the stressfulness of a stress test.

 

[james1701]

The question is geared more to testing your oc. I can start Prime95, then open up a web page to read, or watch youtube or something while it runs. Many people are like me, they find a stable setting, then tweak it many times just for giggles. And when that happens, its gets boring watching 10k test passed and so on all night.

 

[mindless1]

Lots of people have at least a 2nd set of spare parts, so they aren't needing to test an o'c on a system they'd be using as their primary one.

There are lots of reasons for this, including that o'c can easily corrupt the filesystem making it all the more of a PITA if it's the system you regularly use.

By watching utube or whatever else you'd do, you are not stress testing nearly as much. Even an empty loop of program code that is merely keeping the system from using power management, will take time away from the stress test so it is not actually testing what you had wanted to - a true compute intensive full load instead of just busy time. It's like with a person, you can be very busy doing nothing and it's not as stressful as lifting the most weight you can at one time over and over till you can't anymore.

 

[JonW]

I start with memtest86+ 20 passes of test4/5 then boot to windows and prime95 blend. I re-check the ram with HCI Memtest in windows then PCMark04/05. Reboot the systems about a dozen of times (sometimes low vtt/vfsb can make the system to not behave properly eventhough it passes all the stress test) and finally games. When I game and it freezes, it generally points to low nb voltage. A bump of voltage on the nb cures it. The most thing I worry about if i'm oc'ing is nb/ram, these 2 can corrupt files etc. Once the O/S is corrupt, it's very hard to pinpoint whats causing the instability.

By the way James, what are your temps on that U120E with your Q6600@3.4Ghz?

 

[james1701]

I get core 0 and 1 at 65C and core 3 and 4 at 55C. and TMPIN0 goes to 50C and TMPIN1 is 23C after 15 hours of prime95 and 10 hours of OCCT. I have to run 1.4 on my cpu voltage to get stable, with a vid of 1.275.

 

[JonW]

Originally posted by: james1701
I get core 0 and 1 at 65C and core 3 and 4 at 55C. and TMPIN0 goes to 50C and TMPIN1 is 23C after 15 hours of prime95 and 10 hours of OCCT. I have to run 1.4 on my cpu voltage to get stable, with a vid of 1.275.

How much does your mb vdroop on load? My vid is 1.2875, a little higher than yours. At 3.4Ghz, it needs about 1.320v loaded 1.36v set in bios to fully stable. Temps shoot up to 71c-75c with my crappy cooler That's why I run my Q to 3.25Ghz atm. I'm waiting for the copper version of that U120E Those HDT coolers would work but it's been awhile since I used a thermalight product, last one I had was the SP94. Was a great cooler on it's time.

 

[james1701]

I go from 1.408 to 1.392 in cpuz.

 

[JonW]

Originally posted by: james1701
I go from 1.408 to 1.392 in cpuz.

How is your gtl ref set up through the bios? That's a little high imo for 3.4ghz, if your vcore only droops from 1.4v to 1.39v that is. My mb set the gtlref core1-.63x and core2-.67x.

My ram sucks, it only goes to 515@2.30v. I'm focusing on high fsb and tighten all the ram subtimings as much as I can. P45 is very inefficient. It's a little colder today but ussually hits 65c on this current config.
http://img529.imageshack.us/im...96/46x7ramtweakak8.jpg

 

[james1701]

I leave it at auto. I set the SB to 1.3, and VTT to 1.3, and the Vcore. I have not touched any other settings.

 

[JonW]

Originally posted by: james1701
I leave it at auto. I set the SB to 1.3, and VTT to 1.3, and the Vcore. I have not touched any other settings.

Try tweaking your gtlref voltage, probably you can lower your vcore a bit. Gtlref is 2/3 (67%) of your vtt. Looks like your mb is using an identical scheme as my biostar mb which is in multi.

 

[james1701]

I will give it a try tonight. I will save my current profile in my bios, and play with it from there. I always drop core 3 at any lower voltages. Its never any other core. This would be great if it helps. Thanks for the heads up.

 

[james1701]

Well, I decided to go ahead and mess with it. I tried lowering my cpu volts down to 1.36 and messed with gtlref 0 and 1 but it still flopped on core 3 in about a minute. Do you think this was too much of a voltage drop to start with.

 

[JonW]

Originally posted by: james1701
Well, I decided to go ahead and mess with it. I tried lowering my cpu volts down to 1.36 and messed with gtlref 0 and 1 but it still flopped on core 3 in about a minute. Do you think this was too much of a voltage drop to start with.

How are you trying to achieve 3.4Ghz James? Are you somehow using 425x8 or 377x9. You said your vcore droop was 1.40 (idle) -1.39 (load) which is actually pretty good. Can you jot down you full bios settings here and see what could be the culprit. At which fft test in p95 where it fails on core 3?

Edit:
Try Core0 - 0.756 (63% of vtt)
Core1 - 0.790 (67% of vtt)

or

Core0 - 0.790 (67% of vtt)
Core1 - 0.790 (67% of vtt)

 

[james1701]

Here are my settings

377*9
Cpu voltage 1.4075
Gtl ref 0 auto
Gtl ref 1 auto
Dram voltage 1.8
DDR ref A auto
DDR ref B auto
MCH voltage 1.352
MCH glt ref auto
Vtt FSB 1.35
ICH voltage auto

memory is @
754 mhz
5-6-6-18

not the greatest memory, but it works.

When I try to adjust Gtl voltage, it goes from.83 to .88 and 50 numbers inbetween.

I am on bios 1.0

 

[JonW]

With 1.35vtt set both core0/core1 to the highest.

Core0 .67x1.35vtt=0.9045 (choose the closest value)
Core1 .67x1.35vtt=0.9045 (same as above)

If the highest value you can select is .88v, select this vaule for Core0/Core1. Now you have 67% of GTL ref applied to your vtt voltage. So if you change your vtt voltage higher or lower, the gtl ref will also scale with percentage you set through the bios which is 67%.

 

[james1701]

Nope no luck. I even set the VTT up to 1.4 so I could get .9175 and upped the Core to 1.375 and it still bombed on the 3rd core on me.

 

[james1701]

I just read that MSI locks their boards to 65% gtl ref.

 

[JonW]

Originally posted by: james1701
Nope no luck. I even set the VTT up to 1.4 so I could get .9175 and upped the Core to 1.375 and it still bombed on the 3rd core on me.

Set Core0 to 63% of VTT = 0.8505 and Core1 to 67% of VTT = 0.9045 with 1.35VTT. I'm assuming that Core0 is for data bus and Core1 is for address bus. My board likes 63/67 setting better than 67/67. Try testing the cpu with small fft on prime95. It's weird that your cpu needs that much voltage for 3.4ghz and at 377fsb, your vtt should be around 1.25v-1.30v. Your nb voltage is acceptable as you're running 4gb of ram.

Btw, once you have set your gtl ref to 63%/67% of your VTT. You don't have to fiddle with them as they scale with whatever vtt voltage you set in bios.

It's still too early to say that your Q6600 1.275v is a dud on that last core that keeps on failing. Do you mind trying 425x8. Some boards have fsb holes.

 

[JonW]

I was reading this review of your mb and that's how I came up with the formula to calculate the percentage of gtl ref. MSI probably uses 65% if it was set to auto. The range is between 63%-67%.
http://www.tweaktown.com/revie...motherboard/index.html
That review states that the default gtl ref range was 0.7614-0.7906

Default VTT for 65nm cpu is 1.20v
.63x1.20v=0.756v
.67x1.20v=0.804v
.65x1.20v=0.780v

 

[james1701]

Thats a no go on the 425*8. Exact same thing happened. Here is a link that talks about gtl ref on this board. See if you can get more than I do from it. Read the whole post and not the first thread to get more information about it.

http://forum.msi.com.tw/index....topic=119345.msg910487

So far, I dropped my Vcore to 1.395 and set VTT to 1.39 and raised gtl ref 1 to the highest. It looks like it made me loose 1C in my temps.

 

[JonW]

You might also want to read this thread, http://www.xtremesystems.org/f...howthread.php?t=202292 especially the posts by Mike. So you can get a good feel at how to tweak this setting. Also check the article here "DFI Lanparty UT Tweakers Rejoice" take a good look at the settings used at that particular fsb and how the gtl ref was implemented during the process. Your mb is fairly straight forward as it co-relates to the actual gtl ref voltage at a given vtt voltage that's being applied to your particular setting. You just need to jot it down to closest of the percentage you want.

Are you running the ram at 1:1 or with higher divider. I'd like you to remove this out of the equation. Because your running a very low fsb for a p45 chipset. At 400fsb, I only need 1.30v VTT with my mb. Where does exactly prime95 errors out on your quad core. e.g. 1024K-8K-768K etc.....

 

[james1701]

Yes I am running a 1:1 divider. I keep running into the same problem, my bios will not let me set the gtl ref voltage by 67% only by 65%, meaning while everyone else is taking VTT*.67=GTL ref 1, I can only use VTT*.65=GTL ref 0. Every article I have read says you have to use 67%, no more, no less, except for gtl ref 0 which is set to 63% or VTT*.63=GTL Ref 0.

When it errors, its always on 8K. I have never had it pass of 8K when it errors.

 

[JonW]

When 8K errors out, it's mostly cpu related. Did you try both Core0/Core1 to both 63% yet? Here's another setting we can try, 400x8@3200Mhz set vcore to your default vid. If your cpu can pass this frequency even though your vcore will dip to 1.24v-1.26v under heavy load. We can pinpoint that DrMos is the culprit here. Set VTT@1.30v/NB@1.35v/GTL Ref Core0-63%/Core1-67%. Set your ram timing to auto and bump your vddr voltage to 2.0v-2.1v (or its rated voltage). Also try 333x9@3000Mhz@1.16v loaded it should also run stable. I"ll see if I can find any info on DrMos....