What could be causing GPU rendering on a 1080Ti to crash my machine?

[jimjamgraphics]

I decided that it might be best to start a new thread for this as I've managed to test the setup with a different PSU and have got the same results:
http://www.tomshardware.co.uk/answers/id-3761045/problem-faulty-psu.html

To summarise, I've built a new work machine, which for some reason, crashes whenever I do any GPU rendering using a 1080Ti (I've tried two different ones, both fail and my 1080 works fine).

I thought it might be a fault with the PSU, but as I mentioned above, I've just tested it with a new EVGA 1600W PSU and had exactly the same results. So I'm desperate to find out what is the cause of this problem. My next guess would be maybe a fault with the motherboard? Does anyone else have any thoughts or insight as to what it might be?

Many thanks,
Jamie

Spec:
-- CPU: i7 8700K 3.7Ghz
-- MB: Asus Prime Z370-A
-- RAM: 4x16Gb DDR4 3000MHz
-- PSU: Corsair RM1000x (1000w)
-- GPU: 1x 1080Ti 1x 1080
-- STORAGE: 1x M.2 SSD, 2x SSD 1xHDD

 

[SkyNetRising]

What manufacturer/ model name is your GTX 1080 ti? Overclocked model?
I'd try downclocking it slightly.

What crashes exactly are we talking here about? Any error messages? BSODs?

What temperatures do you get on cpu and gpus, when rendering?

 

[jimjamgraphics]

What manufacturer/ model name is your GTX 1080 ti? Overclocked model?
I'd try downclocking it slightly.

What crashes exactly are we talking here about? Any error messages? BSODs?

What temperatures do you get on cpu and gpus, when rendering?

Hi SkyNetRising, thanks for your speedy response.

The 1080Ti model in question is the EVGA GTX 1080 Ti 11GB SC2. I've tried two different ones as I originally thought the GPU was at fault and EVGA were happy to replace it. I don't have any overclock settings and neither the GPU or CPU show signs of overheating before a crash (maxing at 70°C and 60°C respectively)

I've mostly experienced the crashes whilst using Octane render and got software specific crash reports, like this:

https://imgur.com/a/wmwEVpc

However, when I try the FurMark GPU stress test, I get the BSOD..

 

[SkyNetRising]

It is overclocked model (boost clock 1670mhz).
Downclock it to Nvidia reference clocks - boost 1582mhz.

 

[jimjamgraphics]

It is overclocked model (boost clock 1670mhz).
Downclock it to Nvidia reference clocks - boost 1582mhz.

Hi and thanks again for the response. The only overclocking I’ve done was via my Asus motherboard bios. What would be your recommended way to downclock it to 1582Mhz?

Forgive me as I’m pretty new to this. In building this PC I’m also migrating from Apple Mac, so it’s all been a steep learning curve for me.

Also, if this fixes it, then what is it about my setup that means it can’t run the GPU at that clock speed? Just purely so I can have a better understanding. Would say something like a X299 Sage motherboard with an i9 series CPU handle the GPU running at the boosted speed?

Thanks again for you help on this. To finally find a solution to this problem would be amazing,
Jamie

 

[SkyNetRising]

Use MSI afterburner to dial down GPU clocks.

Factory overclocked graphics cards are not always sufficiently tested for operation in OC mode. Stability in OC mode is influenced by power delivery and cooling.

 

[jimjamgraphics]

Use MSI afterburner to dial down GPU clocks.

Factory overclocked graphics cards are not always sufficiently tested for operation in OC mode. Stability in OC mode is influenced by power delivery and cooling.

It feels like I'm certainly closing in on the solution here, I'm really grateful for all the help on this.

I used MSI Afterburner to downclock the GPU core. I first set it to -89MHz so that I got his reading on GPU-Z: https://imgur.com/a/n5rrAky I presumed that would mean that the GPU clock speed would not get boosted over 1582MHz, however when I ran the Octane Bench render, I could see on the GPU-Z sensor readings that it was exceeding that, https://imgur.com/a/juxk6ek and it crashed.

I then used Afterburner to drop the core clock as much as I could (-400MHz). I rendered again and even though the GPU core clock stayed under 1582MHz, I noticed that the GPU memory clock was still exceeding that value. Whilst this seemed more stable, it still crashed towards the end.

So I then used Afterburner to drop both the core and memory clock as much as possible and it ran the Octane Bench without crashing with these sensor readings: https://imgur.com/a/GeOJH8I

Am I correct in thinking for optimum performance, I need to tweak the settings in Afterburner so that when I run the Octane Bench render, I get 1582MHz readings for both the GPU core and memory clocks?

So my next question is where does this 1582MHz figure come from? Is there a specific reason for that amount? What determines it?

I'm a little confused about your comment about OC. If stability in OC is influenced by power delivery and cooling then why is it causing a problem here? I'm not getting high temperature readings and I've tried two very capable PSUs for power delivery.

Lastly, I noticed that when performing successful bench renders with my other 1080 card, The sensors on GPU-Z would give a reading of +1700MHz and still not crash. I know it's a different card technology so that might make a difference, I'm just trying to get as knowledgeable about this as possible, as I still intend to build a Phase 2 work machine at some point in the not too distant future, and I really need to make sure that I iron out any mistakes.

Many thanks again,
Jamie

 

[SkyNetRising]

Try reducing power limit also. If you don't do that, GPU boost tries to overclock automatically.

If reducing GPU clock changes nothing, but reducing mem clock improves things, then may be you need to change mem clock only.

 

[jimjamgraphics]

Try reducing power limit also. If you don't do that, GPU boost tries to overclock automatically.

If reducing GPU clock changes nothing, but reducing mem clock improves things, then may be you need to change mem clock only.

So when I downclock the core by -349MHz, I get a maximum GPU Core Clock of 1582MHz during the bench render. I'd still like to know the significance of that 1582 value, and whether it's set in stone or not? Anyway, it renders fine at this setting (memory clock set to default).

When you say reducing the power limit, do you mean reducing just this value? https://imgur.com/a/LIVmFGz I was a little wary of adjusting the power as I read that it could case physical damage to the hardware. Is that just the core voltage or is it a case that reducing the power values is okay, but not increasing them?

Jamie

--UPDATE--
- After getting stability with the 1080 Ti during Octane Bench render by downclocking the core by -349MHz, I decided to see what would happen if I then added the 1080.
- I used Afterburner to make sure that I had the same downclocking settings for the Ti, and left the 1080 at default settings. This caused a crash during the bench render and I noticed that the core clock on the 1080 was reaching a peak of 1595MHz. Interestingly, I also noticed that peak of the Ti core clock dropped to 1556MHz during the bench render. Is that to be expected?
- I began to use Afterburner to downclock the 1080 core to see if I could make it more stable. With no joy.
- Eventually I dropped the core and memory clock speeds on both the Ti and the 1080 to the lowest that Afterburner UI would allow me. This still crashed during the Octane Bench render.
- I decided to remove the 1080 and go back to just using the Ti with the -394MHz downclocking. Now this isn't stable during the Octane Bench render either.

This is really becoming painful for me. I decided to move from Mac to PC as I became too frustrated with Apple's lack of support for GPU rendering. I just want to be able to build a machine that can handle powerful GPU cards for my work.

--UPDATE X2--

- After doing another clean install, I eventually got the 1080 Ti rendering the Octane Bench successfully, with GPU-Z sensor scores of 1582MHz (downclocking by -339MHz)
Are they any tips as to how I should go about installing and setting up my 1080 as a 2nd GPU?