The M40 is a 1st Generation Maxwell board and is not supported for vGPU.
The oldest GPU you should be working with for vGPU would be an M10 which has 4x 8GB GPUs on it and is currently a fully supported board with the latest vGPU Software. This will run CAD workloads (assuming that the other components of the Server are up to Spec), however it should be noted that this is a low performance GPU, but it should allow you to run the software to some degree.
I would strongly recommend that you stay well away from anything older than the Maxwell architecture. This means avoiding any GPU starting with a "K" (Kepler).
Depending on the CAD applications you will be using (AutoCAD ?), you will also benefit from a QvDWS license to get the best performance out of it. This will allow the maximum amount of Framebuffer to be used for each User.
That said, although the M10 will run CAD workloads, it is not the ideal board for them. You would typically be looking at something like a T4 in combination with QvDWS, assuming the Generation of DL380 is compatible with the T4 GPU.
You’ve mentioned Passthrough / DDA a few times. I’m assuming this is because you are running Hyper-V which currently does not support GPU Virtualisation. The best way forwards would be to either use an M10 with each of the GPUs in Passthrough, and then licensed accordingly, or change your Hypervisor for one that supports GPU Virtualisation. vSphere / XenServer / AHV / KVM all support GPU Virtualisation. That way, you can use a much wider selection of GPUs, and if you’re trying to save costs, you could buy 1x GPU and use the vGPU Software to share it between all of your Users.
As a reference, for CAD, you should be looking at the following VM Spec for each User:
vCPUs: 4-6 Cores @ 3.0Ghz (3.0Ghz at least - Faster is better)
System RAM: 12GB (or higher)
vGPU: 4GB & QvDWS License (4GB at least or higher depending on model size)
Storage: SSD / NVMe