Neverending V-Ray GPU IPR crashes lately (CUDA error 700)

[marksmanguy]

For the passed few months or so I've been getting regular crashes while using V-Ray GPU IPR (always looks like this attachment (CUDA error 700)). Doesn't matter what scene, simple or complex, or whether I use RTX or CUDA. Doesn't matter if I use nightlies, latest public version, different drivers (both recommended and latest). It's obliterating my workflow.

I'm going to message support now, but am wondering if anyone else is going through this and if so what you found?

I'm using 2x2080ti (no overclocking)

Latest V-Ray GPU and latest 3DS Max 2021

Thanks

[nikoleta.garkova]

Hi marksmanguy

Have you tried updating your NVidia GPU drivers to the latest recommended ones? If the error persists, I would recommend using the DDU (Device Driver Uninstaller) tool to completely wipe the NVidia driver and install a fresh copy on a clean system. Sometimes after reinstall or a windows update the driver installation might mess up, and this is a good way to clean it up.
Here's a link to the DDU tool: https://www.guru3d.com/files-details...-download.html

[marksmanguy]

Thanks for the response. I used DDU and reinstalled the recommended driver. Will update here if problems persist. Thanks Nikoleta

[sebbiej]

I'm getting this error repeatedly on a couple of my PCs. I've tried rolling back to the latest supported driver (which my Adobe software doesn't like as that needs a newer driver now), and it still seems to happen. It happens a lot when trying to render an animation in the frame buffer or via batch render. There are reports of this all over google/chaos forums, what is a solid fix for it? I was having it on a large project for the past few weeks, now I've just made a fairly simple animation that takes up 2gb of memory on my cards (1060 on the remote PC, 2080ti on my local PC), and its happening over and over.

I can't run the Device Driver Uninstaller on my remote machine as you need to turn off the internet connection to run it.

Any help or updates to resolve this would be appreciated .

[nikoleta.garkova]

Hi sebbiej,

Is it possible to forward us the scene in question (along with all the assets) so we can have a look and troubleshoot if we're able to replicate the issue? You can do so via our contact form, just make sure to mention this thread in the email. Thanks!

[sebbiej]

Hi sebbiej,

Is it possible to forward us the scene in question (along with all the assets) so we can have a look and troubleshoot if we're able to replicate the issue? You can do so via our contact form, just make sure to mention this thread in the email. Thanks!

Thank you Nikoleta, that's uploading now.

[OzBoz]

I'm not sure if this issue has been resolved but have the same issue.

I loaded a vrmesh and turned on RTX interactive render. If I try to change a material and change it's surface control from Glossiness to Roughness it crashes with error 700. While that's annoying it's even more annoying you have to restart Sketchup before you can use the GPU to render again.

I created a video to show how easily it crashes. Tried another .vrmesh later and while it didn't crashed when I changed some materials to 'roughness' it did on the 5th material. So I reckon there must be at least something that can be reproduced. I'm using the recommended nVidia drivers which are installed using DDU.

Another thing I noticed, if I stop interactive render, change to roughness, start interactive render again it will render. If I change to glossiness > roughness visa versa again it won't crash. Not sure if that's of any help, but I noticed that.

BTW, I see sebbiej uses a 2080Ti + 1060 and I'm using (both in same system) a 2080Ti + GTX1070, so maybe that combination has also something to do with it.

[nikoleta.garkova]

Hey OzBoz,

What if you try to select only the RTX 2080Ti, does the same reproduce on CUDA? I did make a simple test but can't get such errors upon changing the Surface control from Glossiness to Roughness in Interactive (tests were done on a machine with a Quadro RTX 4000). Will have to investigate further.

Meanwhile, is it possible to forward us a sample .vrmesh file just to rule out the possibility of it coming from specific proxies? You can send it via our contact form, just make sure to mention this thread or my name in the email along with details about the exact V-Ray version used. Much appreciated!

[OzBoz]

Hey nikoleta.garkova

I'm wondering if there is some kind of cache involved. Before I created that video yesterday I tried it 4x and it crashed all of those times. Today I try to do the same but it doesn't crash. Doesn't make sense to me so I was wondering if there's some kind of cache directory somewhere. I did delete the things from appdata/local/temp but that doesn't seem to make a diference.

Without a scenario you can reproduce it's a bit hard to make V-Ray GPU more stable, so I'm willing to spend some time trying.

[squintnic]

havign this same issue consistently on a small scene at a low res 1280x720 after 3-4 frames it just claps out. this is on 2 x 3090s
currnet vray, recommended drivrs.

if i had a spare day to trouyble shoot this and upload and wait weeks to fix it i would but for now, again...im forced to switch to redhsift for this render.
please consider more resources for vray gpu dev or make it free or heavily discounted until its stable. or put it back into a private beta?

[Karol.Osinski]

I am using VRay GPU since 3-4 months doing some medium to heavy archviz renders and those CUDA errors is what I see daily at this point.

I am on 3090 and using recommended drivers installed after DDU clean up.

As OzBoz mentioned, the worst thing is that you have to restart the program in order to render again. With some heavy scenes you literally loose 15min for re-saving / re-opening the scene. Multiply it by few errors a day and you loose and hour or two of your time!

I know I might sound like a ranting kid at this point but I will say it again:

VRAY GPU WAS SUPPOSE TO BE A PRODUCTION READY RENDERER IN LIKE 2018.... and Chaos is keeps this *lie* since then.

I promise I will upload some troublesome scenes soon but you guys have to remember we are not your free Beta testers and all this is suppose to be fixed in house, in Sofia or wherever else your surely talented programmers are operating. (Not being cynical here).


I am sorry, all this anger comes from the love I have to Chaos.

[LeePuznowski]

I'll have to agree. After having moved away from Vray GPU for a few months because of consistent instability, I am now trying to come back again. With all the latest drivers, updates, etc. As soon as I try to render a sequence it crashes with the typical errors (ie. 700: Could not release device buffer). Yes, I know I could send a file over, but really, this is just getting tedious. I'll see if I can find time to send a scene.

[GregCooper]

Just want to second this. My experience of Vray GPU has been the same as the posts mentioned above for the past 3-4 months. Even on a brand new build with fresh installs of everything and recommended drivers, getting constant crashes in IPR and also batch renders that often end prematurely. Since this time I’ve been starting to use redshift more and more and the stability is just miles ahead of Vray GPU. As mentioned above we shouldn’t be having to send scenes whenever there’s a general issue like this, we’re not beta testers and it really feels like there’s just not enough resources being put into ensuring Vray GPU is a solid reliable product.

[pipe_mm]

I can copy this
I own three different strong machines, but all of them tend to crash in IPR, Framebuffer, and Standalone GPU.
It would be nice to hear from Chaos if they are adressing this stability issue in the near future.

I dont use another render engine, but would like to know if this hookups are "normal" to GPU renderes,
as to the architecture of graphic cards?

[Muhammed_Hamed]

I dont use another render engine, but would like to know if this hookups are "normal" to GPU renderes,
as to the architecture of graphic cards?

GPU rendering can be very reliable, as reliable as CPU rendering in my experience
Usually any of the errors that are reported and can be reproduced get solved. The Optix 700 errors, in the official build has been solved for some time now.

The recent stable builds 5.1 for 3Ds Max and 5.1.2 for Maya are very stable in my experience