Description
OPTION A:
The assignment is to evaluate the performance of two machine learning
workloads on a parallel hardware/software platform. The applications studied should
present significant computing challenges and should require high performance to
achieve good throughput. Some examples of potential machine learning tasks include:
classification, clustering, regression, association and dimensionality reduction.
Before starting the project, you should submit a project proposal, which includes the
following information:
1. The workloads you will evaluate (e.g., I will study logistic regression and k-means
clustering).
2. The input datasets you will use (e.g., I will use the XYZ data set from the UCI ML
repository).
3. The platform you will use in the study (e.g., I use MPI and evaluate a multi-GPU
implementation on Discovery using multiple nodes).
4. Experiments you will run (e.g., I will generate timing data on the platform, as well
as consider model accuracy. I will use 3 different input data sets).
5. The results you will generate, and the associated grade you would expect to
receive. For example:
A = two workloads evaluated, 3 different inputs used on the platform, all
results reported and analyzed thoroughly in the project writeup.
A- = two workloads evaluated, 3 different inputs used on the platform, all
results reported, but little analysis of the data included in the project
writeup.
B+ = one workload evaluated, 3 different inputs used on the platform all
results reported and analyzed thoroughly in the project writeup.
etc…..
OPTION B:
Review the state of the art in GPU microarchitectural enhancements. You
can find these described in many conferences and workshops proceedings, including
GPGPU. Based on the selected paper, your goal is to reproduce the results in the
paper. You can use GPGPUsim, Multi2sim or another GPU simulation infrastructure for
your assignment.
The range of GPU architectural features include the L2 cache, shared
memory, streaming multiprocessors, and many others. You should evaluate the
proposed design using the simulation framework. Then you will run programs (e.g.,
benchmarks) to evaluate the effectiveness of the studied feature.
Before starting the project, you should submit a project proposal, which includes the
following information:
1. Problem you will study (e.g., cite the paper your work will be based on, and
include a describe of the microarchitectural feature you plan to study).
2. Tools you will use in the study (GPGPUsim, Multi2sim or another simulator).
3. Experiments you will run (e.g., 3 benchmarks, 12 different configurations)
4. The results you will generate, and the associated grade you would expect to
receive. For example:
A = 3 benchmarks studied, 12 different configurations modeled, all results
reported in project writeup.
A- = only 1 benchmark studied, 12 different configurations modeled, all
results reported in project writeup.
B+ = only 1 benchmark studied, 6 different configurations modeled, all
results reported in project writeup.
Etc…..
OPTION C: Select a research topic in the field of high performance computing. The
topic could focus on hardware, software or middleware. Complete a literature review on
that topic, carefully characterizing and comparing a minimum of 10 different papers
related to the topic. The project write-up should include 1-page summaries of each
paper, following by a 7-10 page discussion comparing the various approaches and
suggesting what future directions should be considered on this research problem. It is
expected that the paper will include more than 20 cited papers in the citation list. The
write-up should be single-spaced and use 12pt font.
GENERAL:
You can work in teams of 1, 2 or 3, but of course, a team of 2 should produce 2.25X as
much output, and a team of 3 should produce 3.5X as much output. All team members
will receive the same project grade. You will have the option of submitting your project to
be presented in class. I reserve the right to ask any team member to present the
presentation.
When you are done with your project, you should submit your completed project in a
report. Your report should include the original proposal, as well as a well-written
description of the work completed and analysis of the results obtained. We will try to
leave time during the last class of the semester for students to give a 10-minute
presentation on their project. This is not a required component of the project, but can
add up to 10 points of extra credit.
Please pay close attention to citing your sources, documenting any tools or work that
you use in your project, and produce a document that could be shared with a future
employer.
