remote GPU for tesla C870?

Hi, I understand that you can only use CUDA via VNC, not remote desktop since remote desktop use another display driver. Anyway, I thought it is possible to use it remotely, so I purchased a tesla C870 and install in my office pc. I have to unplug my original video card and install the C870. Although the C870 have no video out, I thought I can still access it remotely. Indeed I can. I remote desktop and setup the VNC server, however, when I close the remote desktop and try to connect to it via VNC, the screen does not show up. Later, I find that if I attempt to connect via VNC while my remote desktop is connected, I can connect via VNC. But as I disconnect my remote desktop session, the VNC window does not react.

I guess it is because VNC want you to have a working video card and monitor, and the VNC is merely intercept the video output. So if I use C870, which doesn’t have a video out, I can’t use it remotely in anyway (I guess linux would still work), am I correct? So now the only solution for me is to have a pci video card (which is the only slot that my computer have) and connect it to the monitor, right?

In Linux you can do this with Xvnc, which emulates a graphics card for X without actually outputting to the screen (the same as remote desktop does), but for windows VNC clients you need an actual screen.