What we noticed is the according to in the wake region of the k-Epsilon and the k-Omega SST. We also tested other turbulence models and, in the final paper, we had compared the results. We selected to use, according to our Supervisor Ezhilmathi Krishnasamy, the RANS model based on the two equations k-Omega SST turbulence models. We used the simpleFoam and pimpleFoam algorithm in order to evaluate the solution for the steady-state and the transient simulations, respectively. We selected to use OpenFOAM, which is an open-source CFD solver widely used for research activities and industry projects. Simulations: this is the crucial part of the work.My mate, Benet Eiximeno Franch, who has been a student during this program, provides to the team the meshes for the three different geometries. Unfortunately, this is not an open-source tool, but it is available in the student version. Pre-processing: in this part, we performed our mesh, composed of up to 46 million fo tetrahedral elements.Once an overall understanding of the architecture was established, our goal was to simulate the different load and store operations (shown below) and analyse their output and performance. My primary objective was to understand how instructions, commands and data packets are to be received into the memory tile.
For example, a scalar load operation is handled like a unit stride vector load with a loop iteration of 1. Another impressive feature of this memory tile is that it allows the re-usability of some already implemented functionalities.
The microengine is responsible for generating transactions for the instructions, whereas the vector address generator generates the memory requests. The memory tile houses the MCPU (Memory Central Processing Unit), which can loosely be described as the ‘intelligence’ of the memory tile, responsible for organizing resources that are needed to perform the different memory operations.These resources are obtained from the microengine, the vector address generator (VAG) and within the MCPU itself.
Illustrated above is the basic structure of the memory tile I worked on simulating using Coyote.