For the 11th consecutive year, NVIDIA is awarding substantial grants to an elite group of graduate students engaged in pushing the frontiers of GPU computing research.
A total of 12 grants of $25,000 each were made under the NVIDIA Graduate Fellowship Program to students selected from hundreds of applications from three dozen countries. The students are using GPUs to confront a variety of technical challenges, including compiler techniques and optimizations, heterogenous computing, formal verification and symbolic analysis, visual sonification and molecular dynamics.
“NVIDIA has a long commitment to supporting outstanding academic research, because we know it fuels innovation,” said Chris Malachowsky, co-founder of NVIDIA and senior vice president of research. “We’ve invested millions of dollars in support of ground-breaking research. We’re delighted to support the work of these exceptional graduate students. Their efforts will help define the future of computing.”
Without further ado, the 12 recipients of the 2012 NVIDIA Graduate Fellowship Program are:
Albert is developing compiler techniques and language extensions to support both synchronous and asynchronous forms of computation on exascale architectures that consist of hierarchies of accelerators and other parallel components.
Ashwin’s research explores parallelizing strategies and performance models for large-scale CPU+GPU systems. He plans to apply his findings in accelerating high-end applications in the field of epidemiology simulations and bioinformatics.
Belen’s research focuses on applying computational models of human perception to the design of new-generation light field (3D) displays. This allows Belen to exploit the characteristics of the human visual system to overcome the inherent limitations of these displays.
Ben is studying molecular dynamics, the computational chemistry method designed to simulate the motion of biomolecules. GPU-accelerated molecular dynamics is being used to understand crucial interactions of proteins, DNA, phosholipid membranes and other biomolecules that make up the molecular machinery of all cells.
Dominik’s research focuses on the area of programming frameworks for heterogeneous multi-core systems, specifically CPU-GPU systems. The focus of his work is on automatically mapping parallel programs across processors in a heterogeneous system. In his approach he uses machine learning techniques to guide the mapping process based on previously observed information.
Haicheng is developing compiler optimizations for accelerating large-scale data warehousing applications on cloud architectures that are augmented with GPU accelerators. Specifically, three optimization techniques under work – kernel fusion, kernel fission and GPU-CPU co-scheduling – are trying to solve the problem caused by the fine-grained nature of relational algebra operators, since their data access time outweighs the computation time.
Jason’s work focuses on developing computer architectures that can efficiently run computer vision applications. We are decreasing the execution time of mobile vision applications while also decreasing the energy required to perform tasks such as augmented reality and gesture recognition. These techniques are enabling an application called visual sonification that dynamically converts a scene into an audio representation for the visually impaired.
Meng’s current research focuses on the design of cache coherence for heterogeneous systems (e.g., CPU and GPU) with a unified memory space. She seeks to optimize the traditional cache coherence protocol designed for homogeneous systems to make it more efficient for heterogeneous systems.
Peng has designed and developed a new tool framework named GKLEE that is the first symbolic analysis assisted checker and test generator tailored for C++ CUDA programs. It locates correctness and performance bugs. GKLEE is different from existing tools that are based on conservative static analysis or heuristic-based testing.
Steve’s research concerns efficient implementations of algebraic multigrid (AMG) methods on GPUs, with a specific focus on the sparse matrix data structures and operations required for smoothed aggregation-based AMG.
Wilson’s research focuses on architectural and micro-architectural improvements to GPUs to efficiently support parallel programming models that simplify communication between threads. He seeks to provide GPU application developers with a programming model that allows them to create properly working code sooner, thereby reducing software development turnaround. Freed from debugging deadlocks and data-races, the programmers can devote more time to optimizing application performance and expanding GPU computations to more domains.
Yunsup’s research explores ways to provide better performance and energy efficiency while retaining programmability and flexibility in data-parallel processors. Currently, he is developing new techniques to more efficiently support irregular control flow.
The NVIDIA Graduate Fellowship Program is open to applicants worldwide. Eligibility criteria include completion of the first year of Ph.D. level studies in the areas of computer science, computer engineering, system architecture, electrical engineering or a related area. In addition, the student must hold a current membership on an active research team.
For more information, please visit our website.