Institute of Technology

CUDA Projects

I. High Performance Systems

Optimizing Bioinformatics Operations:

The DNA matching computations for disease diagnosis have proportionately increased with increased datasets caused due to exponential growth in human population. VELVET which is the most popular assembly tool has HPC enabled versions which utilizes OpenMP. There is a scope for optimization for this assembler. BioInformatics companies like Xcelris Genomics have identified the need for optimization of available Alignment tools and need GPU base solutions for BLASTx aligner for DNA and Protien matching.  GPU based solutions for BLASTp (for protein to protein matching) and BLASTn (for nucleotide to nucleotide matching) are available however no GPU enabled solution is available for BLASTx (nucleotide to protein matching). This research work therefore focuses on parallelization of VelvetH (the first phase in Velvet assembly) using a hybrid combination of GPU+Clusters and providing a GPU based implementation for BLASTx. This research can prove to be very useful for bioinformatics community.

Faculty Involved: Prof Monika Shah, Assistant Professor, CSE Department

Efficient Indexing structures for Multimedia Sets:

Multimedia data is subjective in nature, so it is hard to characterize their features. Depending on the type of application, there exist methods to accumulate specific features from multimedia data. We studied local and global feature extraction methods for retrieval systems to characterize the feature space into metric, high dimensional and multi-dimensional vector space. As the size of mulimedia datasets increases query response time gets worse. Hence, for solving this problem we are implementing multimedia based index structures to speed-up the overall process of retrieval. We intend to validate our index approach using parameters like : a) reduction in search space b) reliability of the method for both image and video datasets c) likeliness of the output with the query image d) resistance to updation e) execution time. To exploit the level of parallelism involved in the indexing structures, we are working towards implementing it reliably on the GPGPU based systems and later compare its performance with the sequential version.

Mentors: Prof Vibha Patel, Associate Professor, CSE Department

Content Based Image Retrieval (CBIR):

CBIR has gained popularity as a reliable tool for many image database applications. The analysis of the remote sensing data requires processing of large volumes of data. The project presents a memory and run-time efficient image texture classification. The project implements algorithms for texture classification on NVIDIA GPU to exploit the possible parallelism in the algorithms to achieve a speedup in the existing computations.

Mentors : Prof Swati Jain, Assistant Professor, CSE Department

II. Scientific Domain

Collaborative project with Institute for Plasma Research, Gandhinagar

Institute for Plasma Research - a premier research organization in India, involved in research in various aspects of plasma science including basic plasma physics, magnetically confined hot plasmas and plasma technologies for industrial application. The institute is currently in the process of building a Steady State Superconducting Tokamak (SST-1) and is one of the partners in the international ITER project. The mechanisms of physical and chemical interactions of low temperature plasmas with surfaces can be fruitfully explored using molecular dynamics (MD) simulations. MD simulations follow the detailed motion of sets of interacting atoms through integration of atomic equations of motion, using inter-atomic potentials that can account for bond breaking and formation that result when energetic species from the plasma impact surfaces.

Many molecular modeling applications are well suited to GPUs, due to their extensive computational requirements, and because they lend themselves to data-parallel implementations. Likewise, in these domain, we need to process; a large volume of data,  very quickly, while modelling sets of many interrelated equations to produce results that are as realistic and reliable as possible.

Mentors: Dr Priyanka Sharma, Associate Professor, CSE Department
Prof Monika Shah, Assistant Professor, CSE Department
Mrs Sutapa Ranjan, Institute for Plasma Research


Dr Nagendra Gajjar is working on a  RESPOND Project, sponsored by ISRO, Ahmedabad. This is  related  to  “Design  & Development  of  32-bit  RISC  processor  based  IP  Core  for  Space Applications”.