BSOD with RTX 2070

[JoeW]

I'm new to Luxcore (with Blender 2.83-2.92) and have been experimenting with it for the last few days. Most of my scenes were really, really simple - a ball and a plane, some cubes and a plane, some sculpting and a plane with an HDRI and/or a single light source. In each case, I was able to render to the viewport, but when I attempt to do a "full" render (F12) I get a BSOD *every* time (WHEA error) - usually after a good bit of calculations (partially completed image). The GPU isn't even sweating - running 58C even at 100% and only 4.5GB VRAM used.

I even tried some of the demo scenes - and again, they work in the viewport render, but crash the system when I try to do a full render (usually at around 800-900 samples). This is on a Windows 10 Pro x64 system, 32GB RAM, 8GB RTX 2070 with the latest drivers.

Interestingly enough, I can take the same scenes to an older machine (Windows 10 Pro X64, 24GB RAM, 6GB GTX980ti) and they will render all the way.

Is there a known issue with the RTX cards or maybe a particular Nvidia driver? I've searched the forum, but haven't found any questions similar. I would think I was missing some configuration setting ... but even the demo scenes crash reliably.

Any suggestions would be welcome.

[Dade]

It sounds like a driver or hardware problem (GPU, Power supply, etc.). I would try to run LuxMark in stress mode for a while to check if it works.

[JoeW]

Thanks for the reply Dade. To make a long story short, you are correct - it was apparently a driver issue. I did a clean install of the latest nVidia drivers and it seems to have solved the problem.

Interestingly enough, for people out there with older nVidia cards, render times between my RTX 2070 and my GTX 980Ti aren't that different. The "Cannelle Et Fromage" demo scene rendered (same settings) in 15m 43s on the 980Ti and 13m 56s on my RTX 2070 - so less than 2 minutes difference - which really surprised me.

Thanks again for the reply.

[Dade]

Interestingly enough, for people out there with older nVidia cards, render times between my RTX 2070 and my GTX 980Ti aren't that different. The "Cannelle Et Fromage" demo scene rendered (same settings) in 15m 43s on the 980Ti and 13m 56s on my RTX 2070 - so less than 2 minutes difference - which really surprised me.

You a need a lot (more) triangles to see an appreciable difference with RTX (in the range of 50+ millions of triangles, it should be start to be measurable after 10M triangles).

[JoeW]

Good to know, Thanks, Dade.

I think I spoke too soon on the driver issue. While it seems more stable in general now, I've still managed to BSOD on some pretty simple scenes. I suppose it *could* be my PSU, but it's a 1200Watt PSU, so nominally I would consider my system overpowered.

I should be able to test more in about a month. I have a newer system on the way, so should be able to test against that (another RTX card).

[Martini]

Good to know, Thanks, Dade.

I think I spoke too soon on the driver issue. While it seems more stable in general now, I've still managed to BSOD on some pretty simple scenes. I suppose it *could* be my PSU, but it's a 1200Watt PSU, so nominally I would consider my system overpowered.

I should be able to test more in about a month. I have a newer system on the way, so should be able to test against that (another RTX card).

Sorry to hear about your troubles

Unless you're water cooling and/or running a few (like 10+) case fans or HDDs, yes your PSU is very likely under-loaded. They typically perform at their best when loaded at approximately 50-75%, although around that it should just be less efficient, not unstable. It could still be a "noisy" supply, though. Are you able to test with a different PSU and the same GPU, or different GPU with the same PSU?

You mentioned 900 samples, how long does that take, and what is your CPU temp after that time? Are you overclocking any components in your system? Are you using the recommended XMP Profile for your memory (if supported)? Did you change the voltage your CPU is receiving? If yes to any of the above, try again on stock settings.

Other things to try: Can you get a BSOD running Cycles renders at similar resolution/sample counts/durations? Do you get BSOD when running LuxMark? What about other stress tests like Cinebench, Unigine, 3D Mark, OCCT, etc.?

Don't overlook stress-testing your CPU and memory too, as instability can also occur there, and may cause BSOD if so. LuxCore is heterogeneous, which means that it will use your CPU as well as GPU for rendering at the same time, while Cycles does not by default.

I know it's a lot to test, sorry. Just try to be methodical and eliminate one thing at a time.