Executing a function on the GPU via CUDA is not the same as calling other external functions that run on the host.That's because the CUDA runtime engine has additional constraints that are not immediately satisfied by LabVIEW's multi-threaded execution.
There's an NILabs module - LabVIEW GPU Computing - which enabled calling CUDA-based functions from LabVIEW but it works with CUDA v2.x only. Unless your application and hardware allow you to use an older version of CUDA, it won't help solve your problem.
However, the module also contains documentation that explains the different ways you can use CUDA from LabVIEW. Some options to not require the use of the module but they too have limitations. Here's a link the NILab page where you can find more information: NILabs LabVIEW GPU Computing.
NI continues to research support for GPU computing from LabVIEW and is interested in specific applications that utilize GPUs. If you have a specific use cases, please joing the GPU Computing group linked to from the NILabs page and post your requirements.
