Textbook in PDF format This is a comprehensive and forward-thinking book that addresses one of the most pressing challenges in modern computing: how to leverage the immense computational power of GPUs within the stringent constraints of real-time systems. As industries demand faster, smarter, and more responsive technologies—autonomous vehicles, robotics, edge AI, and cyber-physical systems—the ability to deliver deterministic performance on inherently non-deterministic hardware becomes critical. This book dives deep into the architectural mismatch between GPUs and real-time requirements, unravelling the complexity of preemption limitations, variable execution times, synchronization bottlenecks, and resource contention. It presents a unified view of the field, combining foundational knowledge with state-of-the-art solutions, ranging from innovative scheduling algorithms and multitasking frameworks to power-aware design strategies and emerging hardware paradigms. The rapid evolution of Graphics Processing Units (GPUs) from specialized graphics rendering engines to powerful, general-purpose parallel accelerators has fundamentally reshaped the landscape of modern computing. Their massive parallel processing capabilities have made them indispensable in time-sensitive, data-intensive applications such as Machine Learning, autonomous vehicles, and robotics. This transformation has unlocked unprecedented performance gains, enabling us to tackle problems once considered computationally intractable. However, the integration of GPUs into real-time systems, environments where computations must be completed within strict deadlines, presents a unique set of challenges. Unlike traditional CPUs designed for sequential, predictable tasks, GPUs prioritize throughput over latency. This inherent architectural difference gives rise to issues such as unpredictable execution times, and resource contention, which can jeopardize a system’s ability to meet its timing requirements. This book provides a comprehensive review of the ongoing research and solutions dedicated to bridging this gap. We delve into the foundational concepts of GPU architecture and real-time systems, analyze the core challenges that impede their seamless integration, and categorize the innovative scheduling techniques proposed to date. Our discussion covers both implicit and explicit real-time scheduling methods, exploring approaches that range from software-based frameworks and compiler-driven transformations to hardware-based and hybrid solutions. The rest of this book is structured as follows: In Chap. 1, we introduce GPU architecture, focusing on the execution model where tasks are divided between CPUs (for sequential tasks) and GPUs (for parallel tasks). The chapter also covers how GPUs execute computations in parallel using threads, warps, and Streaming Multiprocessors (SMs), with details on how CUDA programming facilitates this process. Additionally, the chapter explores real-time systems, emphasizing their need to meet strict timing requirements. It describes various types of real-time tasks (hard, firm, soft, and mixed-criticality real-time systems) and key components like scheduling algorithms (preemptive and non-preemptive) and different task types (periodic, sporadic, and aperiodic). Chapter 2 discusses key obstacles in integrating GPUs into real-time systems. These include difficulties in implementation, such as managing synchronization and resource allocation, variable kernel execution times that affect predictability, and nondeterministic environments that lead to latency and resource contention. Other issues highlighted are power consumption, underutilization, and the overhead associated with context switching. Also we categorize our research on GPU scheduling techniques into two main types: proposed methods which implicitly consider real-time constraints and proposed approaches that explicitly consider real-time constraints. Each of these categories will be investigated in detail in Chaps. 3 and 4, respectively. Chapter 5 presents potential avenues for future research and applications, followed by a concise conclusion. Clear, rigorous, and research-driven, this book is an essential reference for computer scientists, embedded systems engineers, and academic researchers looking to design the next generation of real-time, GPU-accelerated systems. It doesn’t just review the field—it defines it. Preface Introduction Challenges Explicitly Considering Real-Time Constraints Implicit Real-Time Scheduling Future Directions and Conclusion