I think it might be worth to note for those who updated to version 6.0.0 that the .lib and .dll files are now named differently, something I didn’t realize initially.
If you use CMake, in FindOptix.cmake you probably had something like this for 5.1.X:
the pre-compiled samples in the SDK “Optix hello” as well as
cmake built “Optix advanced samples” leads to the same error:
“Failed to load Optix Library”
Did not have the same problem with Optix 5.1 pre-compiled samples
Any ideas? I checked the FindOptix.cmake and its (i think correct)
OPTIX_find_api_library(optix 6.0.0)
OPTIX_find_api_library(optixu 6.0.0)
OPTIX_find_api_library(optix_prime 6.0.0)
You must have at least the 418.81 display driver under Windows or the 418.30 under Linux installed for OptiX 6.0.0 to work.
Right, the FindOptiX.cmake script for the OptiX Advanced Samples on github need to be adjusted to find the OptiX 6.0.0 libs and DLLs because the naming convention changed from majorminor to major.minor.micro.
the pre-compiled samples run with the -n flag (no openGL interop).
for the advanced samples that I compile, there is no error message on compilation.
on run time however, there is this message:
If OpenGL interoperability is not working although the 418.81 driver is installed, that could mean you’re running on a laptop with integrated graphics and discrete GPU and picked the wrong OpenGL implementation.
In that case try to force the application onto the discrete GPU. There should be some setting in the NVIDIA Control Panel for that.
If an application is not finding a special DLL it should tell you with a dialog automatically. If you copied all necessary 64-bit(!) DLLs into your executable folder and it didn’t complain about missing DLLs, that should be ok.
The “memory manager already entered” error is normally a follow-up problem when launching a new kernel in OptiX after the previous one hit a fatal CUDA error. Please check if that was really the first error.
thanks alot.
Its true that the .dlls was not the problem.
Also, indeed the first message was not the memory manager error, but a 304 error related to openGL support.
Am running on a windows 10 pro desktop with 2080 Ti and Intel integrated graphics, but am remotely accessing the desktop from a laptop operating windows 7 via microsoft remote desktop connection.
The problem is that openGL support goes haywire via rdc, as per this post:
Running locally without rdc I can compile and run the SDK samples, as well as run the pre-compiled samples.
We’re happy to help with specific problems, but the issues in this thread have been resolved. Will you please start a new thread and include some detail about what’s going wrong for you including whatever error messages you’re getting.
Please also include any relevant system configuration information, such as: OS version, installed GPU(s), VRAM amount, display driver version, OptiX version, CUDA toolkit version, host compiler or IDE name & version, etc.