RTMNU · Third Year B.Tech Computer Science & Engineering (AI & ML) · NEP 2020
System Administration GPU Computing
BCSEAI&ML6T24.1 · Elective III · Semester VI · 3 credits · College 30 + university 70
One of the Elective III options.
Unit 1: Introduction to Parallel Computing and Evolution of GPU Architectures
6 hoursTraditional computer architecture and heterogeneous parallel computing; the architecture of a modern GPU; the need for speed and speeding up real applications; parallel programming languages and models; the evolution of graphics pipelines from fixed-function to programmable and unified processors, GPGPU, GPU computing and scalable GPUs.
Examined in the 70-mark end-semester paper
- Heterogeneous programming
- GPU architecture
- Evolution of GPU architectures and GPGPU
Unit 2: Data Parallelism and CUDA
9 hoursData parallelism; CUDA program structure and a vector-addition kernel; device global memory and data transfer; kernel functions, threading, function declarations and kernel launch; predefined variables and the runtime API.
Examined in the 70-mark end-semester paper
- CUDA program structure and data transfer
- CUDA kernels and thread hierarchy
- CUDA runtime API and kernel launch
Unit 3: Execution Model
8 hoursCUDA thread organisation and mapping threads to multidimensional data; matrix–matrix multiplication; synchronisation and transparent scalability; assigning resources to blocks and querying device properties; thread scheduling and latency tolerance.
Examined in the 70-mark end-semester paper
- Threads, blocks, grids and warps
- Mapping threads to multidimensional data
- CUDA matrix–matrix multiplication
- Synchronisation in CUDA
- Thread scheduling and latency tolerance
Unit 4: Memories and Performance Considerations
6 hoursMemory access efficiency and device memory types; reducing global memory traffic; memory as a limit on parallelism; warps and thread execution, global memory bandwidth, dynamic partitioning of execution resources, instruction mix and thread granularity.
Examined in the 70-mark end-semester paper
- CUDA memory types
- Reducing global memory traffic with tiling
- Optimising CUDA applications
- Warps, memory bandwidth and resource partitioning
Unit 5: Floating-Point Considerations and Applications
7 hoursFloating-point format, representable numbers, special bit patterns and precision in IEEE format, arithmetic accuracy and rounding, and algorithm considerations; GPU applications in gaming and computer vision, and optimising GPU applications.
Examined in the 70-mark end-semester paper
- Floating-point considerations on GPUs
- GPU applications: gaming, computer vision and optimisation
Unit contents are summarised from the official RTMNU syllabus . AiCanCode.org is not affiliated with Rashtrasant Tukadoji Maharaj Nagpur University (RTMNU).