SPPU · Final Year Computer Engineering · 2019 Course (Revised)
GPU Programming and Architecture
410245B · Elective IV · Semester VII · 3 credits · In-semester 30 + end-semester 70
One of the Elective IV options.
Unit 1: Introduction to Graphics Processing Unit (GPU)
7 hoursEvolution of GPU architectures, parallelism with GPUs, a typical GPU architecture; CUDA hardware: threads, blocks, grids, warps and scheduling; shared, global, constant and texture memory.
Marks: 15 in the in-semester exam
- GPU architecture and its evolution
- Parallelism on the GPU
- CUDA threads, blocks, grids and warps
- CUDA memory: shared, global, constant and texture
Unit 2: CUDA Programming
7 hoursUsing CUDA, multi-GPU solutions; optimising CUDA applications: problem decomposition, memory, transfers, thread usage and resource contention.
Marks: 15 in the in-semester exam
- Writing CUDA programs
- Multi-GPU programming
- Optimising CUDA applications
Unit 3: Programming Issues
7 hoursCUDA error handling, parallel programming issues, synchronisation, algorithmic issues, finding and avoiding errors.
Marks: 17–18 in the end-semester exam
- CUDA error handling
- Synchronisation and parallel programming issues
Unit 4: OpenCL Basics
7 hoursThe OpenCL standard, kernels, host–device interaction, the execution environment, the memory model, basic examples.
Marks: 17–18 in the end-semester exam
- The OpenCL standard and kernels
- OpenCL host–device interaction and execution model
- OpenCL memory model
Unit 5: Algorithms on GPU
7 hoursParallel patterns: convolution, prefix sum, sparse matrix and matrix multiplication; programming a heterogeneous cluster.
Marks: 17–18 in the end-semester exam
- Parallel convolution on the GPU
- Parallel prefix sum
- Sparse matrix and matrix multiplication on the GPU
- Programming heterogeneous clusters
Unit 6: OpenCL and Application Design
7 hoursOpenCL for heterogeneous computing; efficient neural-network training and inference.
Marks: 17–18 in the end-semester exam
- OpenCL for heterogeneous computing
- GPU acceleration for neural networks
Unit contents are summarised from the official SPPU syllabus . AiCanCode.org is not affiliated with Savitribai Phule Pune University.