“GPU Work Graphs” by Kuth, Oberberger and Meyer
Conference:
Type(s):
Title:
- GPU Work Graphs
Organizer(s):
Presenter(s)/Author(s):
Abstract:
Work Graphs constitute one of the most recent feature additions to Direct3D12. It was introduced at GDC 2024. The innovation behind Work Graphs is that they enable GPU compute shaders to dispatch other GPU compute shaders directly from the GPU without going back to the CPU. This makes Work Graphs useful for general purpose computing as well as for graphics applications. They are helpful to easily map dependent, recursive, and irregular work-loads on the GPU. Unlike previous approaches, Work Graphs do this in a fast and memory efficient way. Work Graphs reduce CPU-GPU data-transfer and synchronization. Further, they allow more algorithms to be expressed in a GPU-only fashion. Ultimately, that speeds computation and saves memory. In computer graphics, a plethora of applications such as geometry generation, procedural modelling, shading, or decompression, etc. benefits from Work Graphs in terms of implementation simplicity, memory efficiency, and performance. However, there are some non-trivial shader-language additions that demand a thorough explanation. Therefore, in this course, graphics programmers gain an understanding of the necessary fundamentals, concepts, tools, programming techniques, as well as, related concepts to understand, describe, and use Work Graphs. We focus on the HLSL-aspects of Work Graphs. We cover topics like node-records, node-launches, the classify-and-execute pattern, recursions, and synchronization. We provide insight in how a hardware-implementation of work-graph could work. Further, we show advanced examples demonstrating the power of Work Graphs. As an activity, participants will experience multiple hands-on programming examples using our easy-to-learn WorkGraphPlaygroundApp. It is open-source and includes tutorials with detailed descriptions, tasks, and solutions. After this class, particiapans should understand, explain, and apply Work Graphs. As main takeaway from this course, attendees should be excited and ready to solve own programming projects with Work Graphs.
Additional Information:
- Background and Work Graphs Concepts (90 min) – Introduction and Foudations – Work Graphs in a Nutshell
- Basic Hands-On (90 min) – Nodes, Records, Launches, Classify-and-Execute
- Advanced Work-Graphs (90 min) – Recursion, Synchronization – Applications: Procedural Generation – Work Graphs under the hood – Wrap-Up
Intermediate
Prerequisite: A shader language and basic understanding of real-time computer graphics.
Topics: Games, GPU, GPU-Acceleration, Real-Time
List of topics and approximate times:
Additional Info: This course introduces graphics programmers of all levels to the latest GPU-feature “Work Graphs” and how to use in HLSL and Direct3D12 for their own projects. After this class, particiapans should understand, explain, and apply Work Graphs in their own problem domain.
