I am part of a team developing a desktop OpenGL program on Windows. One use case we are supporting is running the application in a Remote Desktop Protocol connection (RDP) on the Amazon EC2 cloud. I am aware that Microsoft’s old OpenGL software driver, which only supports OpenGL 1.1, is typically loaded in this scenario. To provide OpenGL 2.1 support over RDP, we are manually loading a binary compatible opengl32.dll software driver built from the Mesa project. This is better as we now have off-screen rendering and GLSL 1.2 shader support over RDP, but it is still slow.
My hope is that it may be possible to achieve the same success by manually loading a true hardware GL driver. The EC2 machine will always have an NVIDIA GRID “Kepler” GK104 GPU running on Windows Server 2008 R2 or 2012. Is it possible to manually load the NVIDIA OpenGL driver/libraries from within my process? If so, which libraries must be loaded and in what order (if known)?
Any other thoughts or suggestions would be greatly appreciated.
No. There is no device attached to a Remote Desktop. That’s running an own software rendered desktop with no relation to display adapters.
You would need a system with something allowing hardware rendering on the server. Search for VNC or some GPU virtualization like RemoteFX, NVIDIA GRID, etc.
Can’t agree with @droettger. You should try with creating OpenGL context using [NV_gpu_affinity].(http://developer.download.nvidia.com/opengl/specs/WGL_NV_gpu_affinity.txt). This works perfectly with RDP for me, on local Windows server and Quadro K620 GPU. Just check if the NV_gpu_affinity is supported on EC2. Normally, it’s supported for all Quadro-based cards.
However, I can’t manage to get it working with local access to the server, when main display is Microsoft Standard Graphics Adapter.
[Added 5 mins later]
Ups… very old thread, probably the answer is irrelevant.
Yup, blast from the past.
You would need to be able to load the NVIDIA OpenGL driver in the system at all before you could use NV_gpu_affinity, which itself is only available on workstation boards.
That is not the case if the main display is running the Microsoft Standard Graphics Adapter and not the NVIDIA implementation, which you could only get on Quadro boards or the resp. NVIDIA GRID driver solutions in the original case.
When you require Windows Remote Desktop OpenGL hardware acceleration on GeForce today, have a loot at this link: https://developer.nvidia.com/designworks