Presented by

  • David Airlie

    David Airlie
    @DaveAirlie

    Dave Airlie is a long time contributor to the Linux graphics stack. He co-maintains the graphics driver stack in the Linux kernel, which is the second largest body of code in the kernel. He has also contributed to X.org and Mesa projects, writing various userspace drivers for various GPUs over the years. He initiated the virgl virtual GPU project, and also the radv open source vulkan driver project. He works for Red Hat and is based in Brisbane, Australia.

Abstract

One of the biggest uses for GPU compute is AI/Machine Learning applications. The tensorflow library from Google is one of the most used frameworks in the AI/ML area. To deploy tensorflow on GPUs currently, the closed source nvidia stack is required using CUDA. This talk will explore the work done and left to do to enable a tensorflow deployment on open source Mesa drivers. The use of SYCL via LLVM and OpenCL along with work towards enabling OpenCL on a broader range of hardware will be discussed.