Presentations and other assets from this event are presented here. For information on upcoming events, click here.
In this webinar, members of the Vulkan Machine Learning Subgroup at Khronos will share the very latest updates to the Vulkan extensions and eco-system that directly benefit Machine Learning workflows.
A proposed extension for accelerating ML workloads in Vulkan. This extension targets Machine Learning libraries and Game Engines already using Vulkan. The extension adds Tensor resource types, exposes IHV-optimised pipelines for standard ML operations (metacommands), and adds tensor load/store built-ins to Vulkan compute shaders.
Speaker Bio: Over the years, Jeff has been both architect and lead contributor for the Adreno OpenGL ES and Vulkan driver teams. He represents Qualcomm graphics inside Khronos and authors Adreno extensions for OpenGL ES and Vulkan. Jeff acts as a liaison between Qualcomm's development teams and outside developers.
Matrix multiplication is a fundamental building block for recent advances in machine learning algorithms. Modern GPU have introduced dedicated hardware units which can accelerate this operation, and can be accessed in compute shaders via this extension.
Speaker Bio: Pierre is a software architect at Nvidia, driving software and hardware innovation for the professional markets; over the years, he made contributions to the OpenGL and Vulkan specifications, as well as hardware micro architecture improvement to implement them. He is also the elected chair of the Khronos working group dedicated to machine learning in vulkan. He teaches Machine Learning and Quantum computing to master students at ENSEEIHT, a top ranking engineering school in France, and to PHD students part of ANITI, a research institute dedicated to ML.
IREE is an end-to-end compiler that progressively compiles ML models to both host scheduling logic and device executables using the MLIR infrastructure. IREE uses the Vulkan API and SPIR-V representation to target a broad range of GPUs for mobile, desktop, and server deployment scenarios.
Speaker Bio: Lei is a senior software engineer in Google. He is a main contributor to many Vulkan and SPIR-V toolchain projects---he has initialized/led various efforts for bringing better compilers or tools to the ecosystem. Most recently he is leading Vulkan/SPIR-V efforts in MLIR and IREE for ML on GPUs.
Here’s your opportunity to put all questions to our panel of speakers and Vulkan ML experts. During this session we will also encourage feedback from application and framework developers on any requirements which are not currently addressed by the existing Vulkan ML solutions, or development pain points which could be alleviated.
Conference Code of Conduct: The Khronos Group is dedicated to providing a harassment-free conference experience for everyone. Visit our Code of Conduct page to learn more.
Khronos videos, presentations, and upcoming events. Skip to the Footer
