Crashing/Freezing/OutOfHostMemory with Nvidia or on MacOS systems

[AustinJ235]

This issue has been brought up many times and is possibly due to driver bugs. Further investigation is needed to see if this is an issue within vulkano. Pull request #1246 adds a workaround in the examples so that new users aren't put off when getting started with vulkano and aren't confused as to why it isn't work on their system.

Issues related: #1217 #959 #923 #627 #1022 #1210 #1136 #1135
Pull request related: #1246 #955 #1027

[nickwilcox]

Has anybody done a deep dive on what the possible causes are?

[AustinJ235]

#955 is about the deepest we have gone. Original thought it was an nvidia bug, but somebody posted about a validation error. Something may be incorrect about our GpuFuture handling. So if you cared to do any investigation sanity checking the gpu future code would be a good place to start.

[nickwilcox]

I found that the issue with the triangle example was fixed by an upgrade to MoltenVK 1.1.126 from 1.1.101.

Before that when I took out the workaround I was seeing the MoltenVK swapchain code constantly leaking CAMetalDrawables at 12MiB each.

[Ella-0]

I'm still having issues with this. I've tried a couple solutions and my program still only works reliably on my Intel GPU (HD 630). It crashes after a few seconds on the Nvidia GPU (GTX 1050). I am making a new ImmutableImage every frame which probably isn't the best of ideas but I don't think that's what is causing it.

OS: Arch Linux VK_KHR_display/wayland

[AustinJ235]

Id make sure you are using cleanup_finished and waiting for each future as nvidia might have issues yet it seems.

…

On Oct 5, 2020, 1:58 PM, at 1:58 PM, Ella-0 ***@***.***> wrote: I'm still having issues with this. I've tried a couple solutions and my program still only works reliably on my Intel GPU. It crashes after a few seconds on the Nvidia GPU. I am making a new `ImmutableImage` every frame which probably isn't the best of ideas but I don't think that's what is causing it. -- You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub: #1247 (comment)

[Ella-0]

Yeah I've tried both of those and it still crashes. As for memory usage, it states it was 1.6% of memory before it crashed which is 256 mb.

[Ella-0]

I think it must be my code because I've_ never had issues like this before when using an Nvidia gpu with vulkano though I've never been copying lots of data each frame with vulkano.

[AustinJ235]

Creating an image every fram the driver may not like that. I would recommend create a storage image a copy from a buffer every frame.

…

On Oct 5, 2020, 5:26 PM, at 5:26 PM, Ella-0 ***@***.***> wrote: I think it must be my code because I've_ never had issues like this before when using an Nvidia gpu with vulkano though I've never been copying lots of data each frame with vulkano. -- You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub: #1247 (comment)

[Ella-0]

Yeah, I'll do that and see if the errors still persist.

[Ella-0]

Well I switched everything to use StorageImage so now it doesn't crash, but no images are loaded when using the nvidia card. it works fine with my intel card though. weird.

[kgnlp]

I'm getting memory leaks at around 1mb/s which makes my program practically unusable at the moment. I'm testing on linux with a mobile nvidia 1070 without Optimus.

I'm using an ImmutableImage which is updated only rarely when needed, a CpuAccessibleBuffer with storage usage of which I create a new one every frame, and push constants (potential related to #1118). I'm using both future.wait(None).unwrap(), without which I get OOM very quickly with PresentMode::Immediate, and cleanup_finished after every draw call.

Edit: as mentioned in the other issue this seems to be caused by push constants and only when running it on my nvidia GPU - on Intel HD the issue doesn't occur at all

[kgnlp]

I haven't noticed it yesterday, but it seems like I'm also running into leaks caused by ImmutableImage even though I only replace it when the window is resized though I'm not entirely sure. I only know for certain that leaks occur when resizing the window

[Rua]

Is this still an issue?

[Ella-0]

I have no idea. I don't have nvidia hardware to test with anymore.

[AustinJ235]

Should be closed with #1246

#1246 (comment)

Can you update your comments to refer to #1247 instead.