I am trying to determine whether I can use Unified Virtual Addressing/P2P for my application, given the following details:
-I currently have a Geforce GTX 1080 Ti and Titan Xp connected to same IOH on motherboard
-Application is 64-bit
-OS is Windows 7
-One of the GPUs must be used for OpenGL and connected to display monitor
From what I have searched online so far, in order for the P2P to work both GPUs must be “compatible”. What constitutes as compatible seems to vary slightly across the few sites I have seen thus far but I have gathered and imagine the microarchitecture needs to be the same. Both GPUs are Pascal so I believe I meet this requirement.
Also, it seems for Windows applications, a TCC driver must be installed. I have confirmed this both by running the simplep2p.cu example, which fails and outputs
For Windows Vista/Win7, a TCC driver must be installed and enabled to use P2P/UVA functionality.
and from forums online. So naturally, I believe that I need the TCC drivers installed, but 1) I am not sure how to confirm my GPUs support TCC mode and 2) I am not sure exactly how to download and install the drivers. TCC as whole is new to me, but seems to be a GPU configuration that optimizes for computation and ignores graphical display, as I have seen on a few sites and from a failed
nvidia-smi.exe -dm 1
call that GPUs connected to monitor cannot support TCC mode. Further, since in my case the 1080 Ti was connected to monitor it failed, but the Xp GPU mode changed from WDDM to WDDM*…is WDDM* an equivalent of TCC or does that mean the Xp failed too? And is it even worth me resolving this if I need one of the GPUs for OpenGL and display? otherwise, it seems I would need a 3rd GPU for display which is not an option for me physically or financially, or just forget about using P2P/UVA altogether (which is a viable solution but not ideal as my application requires speed and I am worried about the slow-down from memcpys between GPUs)
thanks so much for any help regarding this