Undergrad Research Project - Selectively Inconsistent Architecture

Fall 2016

Christopher Meredith
Brandon Lucia
Project description

I would be working on finding a multi-core processor. Then modifying it to work with a neural net and to have performance indicators. Specifically, the idea is to see if we can speed up computations by not caring about all of the data we are reading from cache and ram. A neural net will tell if parts of our data needs to remain consistent or be valid at the end of the program. If it does not have to be valid, we can let other cores begin modifying that place in cache. Even if the current task is using it. The method would be to use open core to find a multi-core processor with a relevant instruction set. Then I would work on adding counters and performance indicators to see how performance is being effected. Then I would be assisting in putting the neural net into our processor and testing the performance gains compared to the invalid data. The result should be data on when the performance gain is worth the risk of invalid data.

Return to project list