Upgraded my RAM and now struggling to keep OC.

[Martsmac1982]

So I'm definately having JUST Realbench related issues as I can rin other benchmarks for a long time without anybissies whatsoever, but every time Realbench ends it keeps on throwing up driver related errors or blue screens pertaining to the nvlkmmd I think it is driver (this is running with the latest version)

I have run Aida for hours without fail whilst stressing BOTH the MEMORY and CPU along with all 4 of my Overclocked TITAN X's,( yes I've already tried testing with three cards disabled and both with all 4 at stock speeds and just 1 enabled at stock speeds)

Seems to me like nvidia have a Windows 10/Realbench/TITAN X in SLI issue going on with Realbench.

I'm going to try an older version of the program but this is weird.

[Avenger411]

Hey,

Overlay programs such as Evga Precision X and MSI Afterburner make REALBENCH go boom badadum tis! Shut them down before running RB next time, it might be the culprit. I for one, can't run more than precisely 2 passes on RB when afterburner (rivatuner overlay) is enabled. Programs crashes at more or less 8mins. Shutting down overlay and i can bench all day

Max

[Martsmac1982]

So I have good news and bad news.....

The good is I found the issue that kept crashing real bench (all three versions) the bad news is due to finding this issue I now have to start my overclocking exploits from scratch...

Drum roll please.
SLI

Yep SLI causes Realbench to crash the video driver when halting the test either manually or due to it being the end of the test,

I decided to get somewhat scientific about this so I installed a fresh install of Windows 10 x64 on a spare SSD I had laying around and kept it completey minmal as an install.

Installed only Nvidia drivers (no GFE or 3DVISION) and Physx. I then decided to use the brilliant feature that Asus built onto this board of the PCIExpress slot jumpers to make sure that only one card was active at a time, then proceeded to start and Manually halt the tests at varying times 4 separate times with different card enabled each time. Then repeat in SLI.....

Guess what EVERY time I ran in SLI the video driver crashed upon halting. Every time I ran in single card mode it succeeded.

So now to find my max stable OC for my RAM and CPU I Have to run each test without SLI,

Either that or use a different test program.

I hope this gets addressed at some point by ASUS/Nvidia/Microsoft.

[Arne Saknussemm]

I run SLI and have no problems OCing the cards with PrecisionX and running RB or halting anything....at least on 362.00 driver...

Are you on latest Nvidia driver? the last two drivers are system killers....

[Martsmac1982]

Yep I use the same driver as you (precisely because I read about the issues with the newer ones). TBH my experience with Windows 10 driver side has been awful, if I wasn't interested in directx12 ID have reverted long before now back to 7.

Nope csn reproduce this EVERY time though

Stock CPU mem/overclocked CPU/mem

stock GPUS mem/overclocked GPU's mem

Makes no difference fails EVERY time (on closing/ending test, actual test runs fine) , the fact that other stress tests (particularly GPU based ones) show no errors leads me to believe it's an issue with ROG Realbench/Nvidia SLI conflicting with it.

Like I said if I disable SLI whether that be through software or using the board jumpers the issue disappears.

For now I guess I'll have to OC and bench my CPU in nom SLI mode then turn it on for gaming!

I should note where I said "fails" the test NEVER fails. It just crashes the Nvidia driver in SLI, I'm going to try in windows 7 this weekend to see if it's a Windows 10 issue (I'm running Pro x64 version)

[Arne Saknussemm]

leads me to believe it's an issue with ROG Realbench/Nvidia SLI conflicting with it.

Hmmm, thing is if this were the case I'd have problems too but I can't repeat this here...just been trying

[Martsmac1982]

Yep it's certainly weird happens on all three versions of RB too 2.42, 2.43, and the HWBOT version. I have also tried running it from the C: drive (and where I usually "install" it to my Google Drive folder on my Downloads drive).

Same results each time. Tonight I'm gonna try this out on my daily driver hard drive (gaming setup), and if the same results occur then try it under windows 7

The driver also NEVER crashes under aida64 when I choose to stress the GPU's and EVERYTHING else along with it with SLI on OR off.

[Martsmac1982]

I also noticed Arne that our specs differ slightly what with me having 4 TITAN X's and 64Gb ddr4 , could possibly be more of an issue than your 2 in SLI? will also try to run 2 and 3 cards in SLI tonight.

[cekim]

I also noticed Arne that our specs differ slightly what with me having 4 TITAN X's and 64Gb ddr4 , could possibly be more of an issue than your 2 in SLI? will also try to run 2 and 3 cards in SLI tonight.

I would definitely try that experiment. heavily loaded PCIe + 64G + instability not under load, but as it the CPU ratchets down to lower load all smells like my current pickle in linux...

I was able to get "more" stable at higher clock rates with more SA voltage, but while my cache will run steady at 4.5 @ 1.3v (that's the cache speed - cpu at 4.5->4.8) under 100% load, it will snap off like a twig as the machine idles down in linux.

I disabled speed-step/C-states in the BIOS and am in the middle of my longest run so far (16 hours). Linux is still adjusting the clock, but it twicchy and tends to ramp it up fast, so it is idling at 33C vs 22-25 (the top-end is great because voltage is low).

Short version - I am beginning to think there is something marginal even when things work and even at stock speeds that is exposed under heavy DMI/PCIe load related to power management. What got me wondering about it is my 10GbE adapter. That thing runs HOT (which means it uses a lot of juice).

[Martsmac1982]

So I just ran the first of this weekend's tests, with my settings at stock I passed 15 mins of Realbench WITHOUT the driver issue striking once (with SLI disabled). However when running it with SLI enabled , it crashed at the very end of the test. Has to be an SLI issue.

Running it for 4 hours now with SLI disabled

This was on my day to day gaming rig using version 2.43 of REALBENCH.