Windows Error - The system failed to flush data to the transaction log.

[Karac]

Hi all,

I'm trying to use Realbench (v2.4) Stress Test for my overclock, but everytime the test fails (instability detected), it appears this event on the Windows Event Log:

Event 57: The system failed to flush data to the transaction log. Corruption may occur

without a system crash/BSOD.

After that, I'm forced to run a scandisk to fix my file system, or otherwise my PC will completely freeze and/or crash after a couple of seconds from the boot up.

What am I doing wrong? Is there a way to prevent that?

System Specs:
CPU - i5 3570K
MB - ASUS P8Z77-V LX
RAM - Kingston HyperX DDR3 1600 8GB
OS - Windows 7 Professional

I'm running the test from my secondary HDD, a storage-only drive (I don't know if it's better to run RealBench from my primary SSD one) using up to 8GB of RAM and letting the system manage the pagefile.

Thanks!

[MeanMachine]

Hi Karac and Welcome to ROG.:)

Karac wrote:
Hi all,


I'm running the test from my secondary HDD, a storage-only drive (I don't know if it's better to run RealBench from my primary SSD one) using up to 8GB of RAM and letting the system manage the pagefile.

Thanks!

Herein lies your problem, Realbench should be run from the Root Directory of your C: drive.
Uninstall Realbench and reinstall to its default location.

MM

[Karac]

Thank you very much! I run it from the secondary drive because I've read elsewhere that the x264 test could "damage" the SSD...

In any case, do you know if I need to format and reinstall Windows after those errors (they didn't appear anymore after the scandisk)?

How much should the stress test be run to be sure enough to have a stable system?

EDIT - UPDATE: Unfortunately, I got the same error and the same problem even running RealBench from the root directory of my C: drive.
I downloaded the latest version and copied the folder on C:\. I don't know if there is a version with an installer.

[MeanMachine]

Hi Karac. I am assuming your windows 7 is 64bit?

Running Realbench will not damage your SSD. You will need a stable system to run the Benchmark.
Test for stability using Prime95 (small FFT for 10 minutes whilst constantly monitoring your temperatures)

I have run Realbench several times from my SSD and have no problems, so the root cause lies elsewhere.
You may have corrupted windows files, during installation,

Try running "sfc /scannow" (in a command prompt as administrator, without the quotation marks). Windows will fix corrupted files or tell you it can't, in which case a reinstall or repair of Windows will be necessary.

Also check your registry and clean up windows using CCleaner.

Please complete these tasks and report back.

MM

[Karac]

Sorry for the late reply, I did a ton of tests yesterday 🙂

I had indeed some corrupted Windows file that have been fixed by sfc /SCANNOW.
So far I haven't had any other problem, nor error reports on the Event Viewer.

I have some other questions if someone is so kind to enlighten me :

- is it fine to run Realbench on offset voltage mode (I have a Ivy Bridge processor)? I've read elsewhere that you should stress test only with a Manual/Fixed VCore.

- how much should the test be run?

- is there a way to quickly rule out glaring instabilities before a long stress test? So far I had more luck using the Heavy Multitasking Benchmark rather than the stress test itself.

MeanMachine wrote:
Hi Karac. I am assuming your windows 7 is 64bit?

Test for stability using Prime95 (small FFT for 10 minutes whilst constantly monitoring your temperatures)

Yes, I have Windows 7 64 bit.

I really don't like Prime95. My machine passed a full 24 hour custom blend test with 90% of available RAM (version 27.9) and yet it crashed while playing League of Legends... On top of that, it raises the temperatures way higher than Realbench.
Actually I stumbled upon Realbench while searching for an alternative to Prime :).

Thank you very very much for the support MM.

[MeanMachine]

Hi Karac

Yes I understand what your saying about Prime95 and IB, however your CPU would throttle well before any damage is done. I only run it for 10mins to test for initial stability as the algorithms increase after each pass. I would not recommend long runs without constantly monitoring temps.

Prime95 does not stress test your GPU, so if a game crashes, it may be a driver or game crash or the GPU itself.
There are other testers Aida64 (evaluation is free) and Intel has one too. HWbot is a good site to get software for testing and benching.

I agree that fixed core is better than offset mode.

Realbench is the best real world tester there is, IMHO if your system is stable for an hour of stress testing then you should be OK. Some fanatics think its great to thrash the CPU overnight, unattended to claim ultimate stability and wonder why their CPU has degraded, within a short period of time.

No glaring instabilities, Just make sure your system is well cooled for the OC you want.

MM

[Karac]

MeanMachine wrote:

Prime95 does not stress test your GPU, so if a game crashes, it may be a driver or game crash or the GPU itself.

I'm 100% sure the culprit was an unstable overclock since the BSOD code was 0x124 (IIRC) and after I lowered the multiplier it has never happened again once.

MeanMachine wrote:

I agree that fixed core is better than offset mode.

So should I never stress test in offset mode, or never use offset mode at all?

[MeanMachine]

Hi Karac

These articles should answer your queries re; CPU Core Voltage offset:

http://rog.asus.com/51092012/overclocking/overclocking-using-offset-mode-for-cpu-core-voltage/
http://rog.asus.com/forum/showthread.php?2162-Overclocking-Using-Offset-Mode-for-CPU-Core-Voltage

for further reading on the topic, go to: http://www.overclock.net/t/1219588/updated-part-ii-offset-mode-overclocking-starter-guide-and-thread
Lots of differing opinions.

MM

[Karac]

Yeah, I've already read all 3 articles that you posted and I'm still not sure if it's fine to stress test on Offset mode 🙂
What I've done is finding a "fixed" VCore that was stable, than switching to Offset, loading up RealBench for a couple of minutes and then trying to match that VCore increasing or decreasing the offset voltage adjustment.