Hi,
I was wondering if I was the only one getting spammed with this message :
D3D12 ERROR: ID3D12CommandAllocator::Reset: A command allocator is being reset before previous executions associated with the allocator have completed.
I am only running the basic HelloWorld sample from microsoft. I also get awful crashes (full fatal reboot your PC types) when using the graphics debugger.
I have a nVidia mobile adapter (970m) & note that none of that is happening if I use Intel integrated GPU.
I am using the driver 364.72
I did a few more tests and it seems that on top of all the other issues mentioned above, the "Present" call does not seem to respect command list fences as I get very unusual tearing. As if the back buffer was presented mid-way during drawcalls. I know it sounds silly but I can see the clear color rip through the triangle shown on screen.
I was not able to take a screen capture of the issue. It reminds me of good old days with mode 13 where you could render directly on screen and see the pixel show up as the where updated in memory.
It could very possibly be just a driver issue. The D3D12 API is still quite new, and the hardware vendors are still very much tweaking the drivers for D3D12 interoperability. That being said (And not having seen the Microsoft HelloWorld Example), is it possible that the CPU & GPU aren't being synchronized before the Allocator is reset? (Which your error message strongly suggests)
Some flow control that prevents the function that delimiters the fence object on the GPU, and CPU waits for the event not being called? I'd step through the code (And pay special attention when you reach the Synchronization function, or code) and see if something is aloof.
p.s. That driver I believe is bleeding edge for NVidia. Wouldn't hurt if my previous suggestions get you nowhere to consider rolling back to an older version of the Driver.
Marcus
I believe its a driver bug.
The message indicate an error that would normally throw an exception which it does not. The call to reset actually return S_OK and still produces the message.
I've looked at the code and compared it to my own and I believe the sample is fine. I also ran it with warp without any issue.
Looks like ill have to live with my Intel GPU until they improve the drivers.
Thanks for the answers!
I can confirm such issues with the GeForce GTX 980M too.
On Desktop GeForce GTX 980's and the Intel iGPUs all is working fine.
What builds of Windows are you both running?
If you run 'winver' from a command prompt, it'll say something like "OS Build 14318.1000", what's that number?
I've got a GTX 970 here that I've just put on the 364.72 driver and I get no such warnings on the Debug Layer here. I'm compiling/running on the 10586 Windows SDK too, you might have some out-of-date bits?
Hi,
Windows 10 x64 Build 10586 (End-User Build)
Current NVIDIA Driver 364.72
But this bug is since Driver 350er Series I remember
But this issue seems to be only specific to NVIDIA Mobile GPUs.
My mistake, I missed the 'm' on the original post and thought we had a case of one Desktop and one Mobile part with the issue. I'll raise it with the team and see if we can have someone chase this up.
I am using windows 10 pro version 1511 (os build 10586.164)
platform sdk version: 10.0.10586.0
