compile ColorSpace.cu without cuda toolkit

Hi, the previous versions (couple of years max)of the NVIDA SDK allowed us to load NV12ToARGB_drvapi_Win32.ptx file at runtime for colour conversion of decoded video frames.

The new v9 SDK requires the NVIDIA toolkit and a compile option for ColorSpace.cu as part of the VS workspace (which requires an NVIDIA card to develop) whats going on? how can I deploy our product software so that our sdk can compile on different hardware?

Ive taken the latest sample apart removing all toolkit dependencies, this is our final issue.
Basically, why cant i compile sample:
C:\NVIDIA\Video_Codec_SDK_9.0.18\Video_Codec_SDK_9.0.18\Samples\AppDecode\AppDecD3D
on an Intel NUC for example?
Am I missing something?
Can I use an intermediate object file compiled on an NVIDA machine?

Our software determines the hardware platform at runtime, which i thought was pretty common?
Cheers for any thoughts on this.

We are looking into this.

Have you considered using nvcc to compile the CUDA kernels into PTX? You can use nvcc on a machine without GPU, if needed. The PTX can be bundled with the application and then the driver on target machine will execute it.

Hi, thanks for the reply, this is the avenue we’ll be pursuing next yes. We also noticed that the decode sample mentioned above works on a P3000 MXM but not a Quadro K model (I’ll have to check again) it just renders a blank black screen.

We’ll have to perform much development to upgrade your SDK to the latest version in our security product :( cheers for the response Abhijit.