Tesla connected via Ethernet with other graphics card

Hello, How about?. I wanted to know if you could help me a problem. I have a machine with four graphics cards Tesla (no video output and Linux operating system) and another machine with a graphics card if you have video output (Linux operating system). Would connect the two machines so that the Tesla machine calculates the data and send to the second machine via Ethernet or some other type of network, so this shows the results in OpenGL. Do you know any way to do it hardware or software?

Sorry if I posted the message in the appropriate forum, I’m new yet.

Thank you very much.

If you have to go over network you have to look at openMPI. You have to read the entire array you want to view back to the CPU, send it to the other computer, and then do a memcpy to the other gpu with video output.

If you’re doing some kind of time-step iterations real time is not feasible. I would write everything to binary files then reconstruct the data later on the other machine. Unless the GPU is connected directly (SLI) to your Tesla machine there’s no point in trying to do anything realtime.

US llnl.gov site has a good tutorial on MPI (I learned mpi through their site). I suggest understanding the ‘cartesian grid’ section if you’re working with any kind of spacial data. That is extremely useful. To use multiple GPU’s I expect you already know at least openmp, so openMPI is only a small step away : )

Good luck!

If you have to go over network you have to look at openMPI. You have to read the entire array you want to view back to the CPU, send it to the other computer, and then do a memcpy to the other gpu with video output.

If you’re doing some kind of time-step iterations real time is not feasible. I would write everything to binary files then reconstruct the data later on the other machine. Unless the GPU is connected directly (SLI) to your Tesla machine there’s no point in trying to do anything realtime.

US llnl.gov site has a good tutorial on MPI (I learned mpi through their site). I suggest understanding the ‘cartesian grid’ section if you’re working with any kind of spacial data. That is extremely useful. To use multiple GPU’s I expect you already know at least openmp, so openMPI is only a small step away : )

Good luck!