I have installed 1080Ti card on the mother board that already has integrated graphics from Intel. I want to use the GPU for computation purpose only. My display is connected to HDMI port of integrated graphics card.
Currently when I use tensorflow it shows that available graphics card memory is 9.1GB instead of 11 GB. Is possible to reclaim some memory for computation use (operating system is Windows 10 Pro) i.e. explicitly use GPU for the computation purpose only?
I have the same problem but with the Titan Xp card. I tried many different install methods using CentOS, and it always failed (different failure modes depending on what I tried). It may simply be the Linux driver for Titan Xp is not finished yet as only the beta version is available.
So I want to try it again under Windows. There is a custom install option under which I can just install the “graphics driver”. If I install just this driver, can I continue to use the Intel GPU on the motherboard for the graphics and just use the Titan card for CUDA computation? I want to use simulation software which can take advantage of the CUDA card, and I do not want to lose any of the capability to graphics.