RTMNU · Final Year B.Tech Artificial Intelligence · CBCS
High Performance Computing
BTECHAI802.2T · Elective V · Semester VIII · 3 credits · College 30 + university 70
One of the Elective V options.
Unit 1: Modern Processors
6 hoursStored-program architecture, cache-based microprocessors, performance metrics and benchmarks, Moore's law, pipelining and superscalarity; memory hierarchies, cache mapping and prefetch; multicore, multithreaded and vector processors, design principles and programming for vector architectures.
Examined in the 70-mark end-semester paper
- Stored-program and cache-based microprocessor architecture
- Performance metrics and scaling
- Moore's law, pipelining and superscalarity
- Memory hierarchies, cache mapping and prefetch
- Multithreading and latency hiding
- Vector processing, VLIW and array processors
Unit 2: Basic Optimisation Techniques for Serial Code
10 hoursFunction- and line-based profiling and hardware performance counters; common-sense optimisations — eliminating common subexpressions, avoiding branches, SIMD; the role of compilers — optimisation options, inlining, aliasing, accuracy and register optimisation, compiler logs, C++ optimisations; data-access optimisation.
Examined in the 70-mark end-semester paper
- Profiling: function-level, line-based and hardware counters
- Common-sense serial code optimisations
- Role of compilers in optimisation
- Data access optimisation
Unit 3: Parallel Computers
8 hoursA taxonomy of parallel computing; shared-memory computers, cache coherence and hierarchical systems; networks — buses, switched, fat-tree and mesh networks and hybrids; data and function parallelism, parallel scalability and what limits parallel execution.
Examined in the 70-mark end-semester paper
- Flynn's taxonomy
- Shared memory and cache coherence
- Interconnection networks: buses, switched, fat-tree and mesh
- Amdahl's and Gustafson's laws
Unit 4: Distributed-Memory Programming with MPI
6 hoursMessage passing and MPI; point-to-point, collective and non-blocking communication; virtual topologies; MPI parallelisation of a Jacobi solver and its performance.
Examined in the 70-mark end-semester paper
- MPI
- MPI collective and non-blocking communication
- MPI virtual topologies and the Jacobi solver
Unit 5: Shared-Memory Programming with OpenMP
6 hoursOpenMP parallel execution and data scoping, work sharing for loops, synchronisation and reductions, loop scheduling and tasking; a parallel Jacobi algorithm, wavefront parallelisation, profiling OpenMP programs and performance pitfalls.
Examined in the 70-mark end-semester paper
- OpenMP
- Synchronisation and race conditions
- OpenMP loop scheduling and tasking
- OpenMP performance pitfalls
Unit contents are summarised from the official RTMNU syllabus . AiCanCode.org is not affiliated with Rashtrasant Tukadoji Maharaj Nagpur University (RTMNU).