CSCI 551  Numerical Methods and Parallel Programming  4 Units  

Prerequisite: CSCI 440 with a grade of C or higher, MATH 109 or MATH 120; or Classified Computer Science Graduate Standing.   

Typically Offered: Fall and spring  

Surveys numerical methods and parallel programming. Parallel architectures and algorithms for scientific programming applications, solution of simultaneous linear and non-linear equations, iterative techniques, matrix methods, and error analysis. Development of parallel algorithms, with a focus on efficient implementation and performance. 2 hours activity, 3 hours discussion.  (002332)  

Grade Basis: Graded  

Repeatability: You may take this course for a maximum of 4 units  

Course Attributes: Upper Division 

Course Description: This course focuses on numerical methods used in computer science including vectors, matrices, data transformation, methods of integration, difference equations, linear systems, non-linear, error analysis and root solving.  Methods to speed up algorithms used for numerical and data analysis will be studied in 4 main categories: 1) OpenMP shared memory, 2) POSIX threads shared memory, 3) Message Passing Interface (MPI) for distributed memory clusters, and 4) CUDA GP-GPU co-processing, along with use of libraries and compilers for parallelization.  Use of NVIDIA CUDA-Q or IBM Qiskit for quantum computing, a fifth category of co-processor parallelism, will be introduced as an optional method to complete a final parallel program with potential quantum advantage and speed-up.

Goals: The purpose of this course is to introduce students to specific techniques which are widely used numerical methods and associated algorithms as well as methods to speed up execution with concurrency, using both distributed processing and parallel tasking.  Coverage includes both high performance and high throughput computing.

Lab Resources: NUC Cluster (62 Intel NUC nodes, 2 cores each, 124 total), cscigpu A100, DGX-1 Cluster (4 V100 Infiniband nodes), Jetson Cluster (20 Nano 4g nodes, 4 cores each, 128 co-procs each, 80 cores, 10,240 co-procs), Raspberry Pi Cluster (30 R-Pi 4 nodes, 120 total).