Home  
 

Main Menu
Home
About us
Project Description
Quantitative Results
Research Lines
Research Results
Impact on Society
Press room
Contact us
News
Secure Login
Events Calendar
« < October 2017 > »
M T W T F S S
25 26 27 28 29 30 1
2 3 4 5 6 7 8
9 10 11 12 13 14 15
16 17 18 19 20 21 22
23 24 25 26 27 28 29
30 31 1 2 3 4 5
Login

subTask 1.1 (UM): Internal organization of the GPU

Leader: José Manuel Garcia; Researchers: Juan L. Aragón, Kenneth E. Ostby, José Pérez

Shortly after the start of the project, we redefined the task. Our new task entitled "Internal organization of the GPU” focuses on the design of the internal organization of high-performance GPUs for general purpose calculations. We started our work by developing a simulator for the GPUs.

1. Brief Description of the Goals

As graphics processors (GPUs) are gaining popularity for performing tasks of general purpose computing, it becomes increasingly necessary to introduce profound changes in the design of the architecture of modern GPUs to see not limited their performance to general purpose applications, while still maintaining its high performance in the processing of three-dimensional scenes. As the field of GPGPUs advances, so does the need for a simulator. With the introduction of different architectures, and different programming environments, one of the biggest problems is having a simulation framework that can rapidly adapt to the changes, while allowing for rapid prototyping of novel improvements.

2. Scientific and Technical Developed Activities

Our major requirements were the following: the simulator should be vendor independent, that is, internally independent of any vendor provided formats such as CUDA or OpenCL; additionally, the simulator should separate the timing model from the functional model, thus make it easier implementing architectural behaviour in either the functional or timing model without modifying the other part.

We have developed a simulator called FATSEA (Functional And Timing Simulator for Emerging Architectures) following the guidelines described in the previous section. It was necessary to begin to develop this simulation tool from scratch. Later, we developed a compiler that converts PTX code (obtained from CUDA compilers) into an internal format that is handled by our tool. A set of benchmarks (commonly kernels and simple applications) have been compiled, executed and measured. We have compared the results obtained in our simulator to the real GPU ones, tuning our simulator and doing a characterization of various parameters of a GPU architecture. The results were published by Østby et al. in the 2nd Workshop on Rapid Simulation and Performance Evaluation: Methods and Tools, 2010. 


Publications: 

Projects funded by Public Calls:   [HiPEAC] by  national grants

External collaborations Academia: --

External collaborations Industry: --

Company Agreements: --

PhD dissertations: --

Patents:  --