Abstract
This chapter focuses on the CUDA execution model. The heart of CUDA performance and scalability lies in the execution model and the simple partitioning of a computation into fixed-sized blocks of threads in the execution configuration. CUDA was created to map naturally the parallelism within an application to the massive parallelism of the GPGPU hardware. Understanding the streaming multiprocessor is the key to understanding GPGPU programming. The twin concepts of a thread block and warp of SIMD threads encompass the scalability, performance, and power efficiency of GPU computing. From the high-level language expression of the kernel to the replication of the lowest-level hardware units, on-board GPU scalability is preserved while many common parallel programming pitfalls are avoided. The result is massive thread scalability and high application performance across GPGPU hardware generations. The CUDA toolkit provides the programmer with those tools needed to exploit parallelism at both the thread level and the instruction level within the processing cores. Little's law and queuing theory in general, provide the theoretical foundation up on which very detailed GPU and application models can be based.
Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
