Z77A-GD65 + i5 2500K Overclocking - How to start?

[profJim]

I only use it as GPlops test and quick stability test.

How do you enable a GPlops test?

[Zach_]

Actually it doesn't get too hot, the softwares just won't work when LinX is used, however, that doesn't happen with IBT and I am able to monitor all the temps.

At 4.6GHz I'm idling @ 29-35C and 69-74C at full load with 6637MB used in IBT (Ambient around 24C, have the Air-con working hard :))

Just finished 20 passes of stress testing at Very High with IBT and then another 20 passes with Maximum Stress setting of 6637MB, I deem 4.6GHz stable atm, will run P95 overnight just to make sure however... :)

Screenshot:


My voltages currently are:

Vcore - 1.37 (Droop to 1.352 at full load)
S/A - 0.965
I/O - 1.07
PLL - 1.85

Bernhard from the MSI forums stated that:

"As far as I remember, the 1,89V should be the absolute max for the PLL. SA absolute max is 0,971V, I/O is listed as 1,05V +3% as absolute max."

Seems like my S/A and I/O are quite close to the max, I'd like to try again with

Vcore - 1.37
S/A - 0.945
I/O - 1.05
PLL - 1.84

What do you guys think?

[Chike]

I've seen people using SA over 1.1v, haven't tracked their CPU condition over time.

Have you tried a lower PLL? Not sure about SB but some new processors work better with under default PLL.

I'd keep pushing it without IBT get it stable Prime95 small FFT for several hours instead.

[profJim]

There are some absolutes when it comes to bios voltage settings.
Some settings can be somewhat higher than Intel's design specs, other settings should never be exceeded.
People can tell you from their experience what generally works.
The goal is to use the lowest stable voltage settings.

Go ahead and test with lower voltages, but only change one setting at a time and re-test.
Changing more than one variable at a time makes it impossible to know which settings caused the failure.

Make sure that you are using the IBT and Linx versions that support AVX testing.
I suggest limiting IBT or Linx testing to 5 loops using all available memory.
These two programs are fine for preliminary stress testing only.
Using Prime95 and/or HCI Memtest in windows is a far more rigorous and reliable stress test.
Your GPlops** scores are excellent and comparable with Ivy Bridge results.
Your maximum vcore temperatures are almost 20^o (C) lower than with Ivy Bridge processors.

** sorry, I can't get my mind out of the gutter

[Chike]

It is obvious from his plops that he is using AVX version of IBT.

[profJim]

It is obvious from his plops that he is using AVX version of IBT.

Good point, you're right.

[Zach_]

We're in business, I tested 4.6 with IBT on Maximum Stress with 6800MB, top temp was 70C (Zalman CNPS10X is a champ :)), 4.6 was fully stable for 10 passes of IBT on the following voltages:

Vcore - 1.37
S/A - 0.945
I/O - 1.05
PLL - 1.84
PLL Overvolt - Enabled
Digital Compensation Level - High
EIST, C1E, C-State, Overspeed Protection, Tubo, EuP 2013, Phase Control - All Disabled.

I peaked 124 GFlops in 134 seconds, highest I ever got :)

As I suspected, it was just the Vcore that needed a little boost, however, when I tried PLL at 1.82 I got a BSOD after 5 IBT passes, so I left it at 1.84.

This is a 100% pure raw OC, just as I like it :)

Going for P95 all night long now, hopefully when I wake up tomorrow morning all threads will still be running :) BTW, what's the most effective way to test Prime? I usually just use the default Blend test, without increasing RAM, right now it's using 2.5 gigs of RAM, I tried to test it once with all the available RAM in Custom Blend but it wouldn't stress the CPU at all, only the RAM usage was high...

Thanks

[profJim]

With Prime95, are you going to run each of the three tests, small FFT, large FFT, and Blend for at least 6 - 8 hours each?
Lsdme feels that HCI Memtest is as good as Prime95 and it doesn't take nearly as long to run.
You can run several instances at one time for several hours until the memory coverage has completed at least 100% - 200% coverage.
HCI MemTest: RAM reliability tester for Windows
Check out the Manual and Support + FAQ pages for details.
Download the free version now and you can optionally buy the Pro version for $5 or the Deluxe version for $14.
When you buy, email delivery is cheapest for international orders.

Zach_
Please add a detailed hardware list to your forum signature.
Note: as of now, the PC Specs feature still hasn't been fixed.

[Zach_]

Actually I was just thinking of running Blend for 10-14 hours and deem the system stable, as it tests both Large and Smal FFTs, isn't that enough?

After running Blend for 10 hours I have no problem running another couple of hours of HCI, but if IBT and P95 is enough, then I'm fine with it, just let me know if I am testing correctly on P95.

Thanks

[profJim]

Actually I was just thinking of running Blend for 10-14 hours and deem the system stable, as it tests both Large and Smal FFTs, isn't that enough?

After running Blend for 10 hours I have no problem running another couple of hours of HCI, but if IBT and P95 is enough, then I'm fine with it, just let me know if I am testing correctly on P95.

Thanks

What price for stability?
My preference with Prime95 testing is to run each test for at least 10 - 12 hours or longer.
I've found that Prime95 will sometimes run error free for over 10 hours for one or two of the tests, but fail the third test, sometimes only after 8 - 12 hours. If your system isn't stable, you run the risk of file corruption. A 14 hour Blend test is similar to running the small and large FFT tests for 7 hours each.

• The Small FFT test primarily stresses the cpu with very little memory testing.
• The Large FFT test heavily stresses the memory sub-system.

I'm comfortable with running Prime95 Blend for 12 - 14 hours, followed by HCI Memtest for at least 8 -12 hours.
As stated earlier, IBT or Linx are good for short preliminary stress testing only, and their GFlops scores are a good indication that your system performance is in the ballpark.

Click on the Prime95 tutorial link in my signature for more information.

just my 2 cents