Skip to main content

GPU Hackathon 2022


Duration: 5 Sept, 2022 and 12 Sept - 14 Sept, 2022

Total participants: 60+



About

The NSM GPU Hackathon was a multi-day event designed to help teams of three to six developers accelerate their own codes on GPUs using a programming model, or machine learning framework of their choice. GPU Hackathons provided exciting opportunities for scientists to accelerate their AI research or HPC codes under the guidance of expert mentors from National Labs, Universities and Industry leaders in a collaborative environment.



Clusters used

  1. PARAM Vidya, CDAC Pune
  2. PARAM Vidya, IIT Goa
  3. PARAM Vidya, IIT Kharagpur
  4. PARAM Vidya, IIT Madras
  5. PARAM Vidya, IIT Palakkad



Participated Institutes

  1. Tata Institute of Fundamental Research, Hyderabad
  2. Institute for Plasma Research, Ahemdabad
  3. Indian Institute of Technology, Bombay
  4. Indian Institute of Technology, Goa
  5. Indian Institute of Technology, Kanpur
  6. Indian Institute of Technology, Kharagpur
  7. Indian Institute of Technology, Palakkad
  8. IISER, Bhopal
  9. National Centre for Radio Astropysics, Pune
  10. Acharya Institute


Teams

  1. Team 1 (Team Fluid3d)
    • 2D spectral code: Ported the MPI code into CUDA code
    • 2D Diffusion code: Optimized existing CUDA code
    • Particle code: Optimized existing OpenACC code
  2. Team 2 (Team BasicSimulation)
    • Vlasov equation coupled to the Maxwell’s equation
    • Modelled using LCPFCT algorithm using Fortran90
    • Optimize the code using Nsight system and Nsight compute
    • Reduce the frequent data movement to and from device
    • Remove the bubbles between kernels and between two consecutive time
  3. Team 3 (Team FEASCT)
    • 3D fluid flow solver based on Artificial Compressibility Method
  4. Team 4 (Team CGMLI)
    • SRGAN downscaler
  5. Team 5 (Team DC_IITK)
    • Nonhydrostatic ICosahedral Atmospheric Model (NICAM)
    • Install the NICAM_DC on GPU platform, and run some basic test case like- Jablonowski-Williamson test
    • Optimize the model using CUDA libraries
  6. Team 6 (Team VAYU)
    • FEM Based Navier Stokes Solver
    • Make a hybrid code able to run both on CPUs and GPUs
  7. Team 7 (Team AI for FRB)
    • Classification of Fast Radio Bursts and Radio Frequency Interference
    • Algorithmic motif - Classification of FRB and RFI by combining 11 pre trained models using transfer learning
  8. Team 8 (Team CHAOS)
    • Investigation of Quantum Turbulence in Bose-Einstein Condensate
  9. Team 9 (Team NCRA)
    • SCRIPT: Semi Numerical Code for ReionIzation with PhoTon Conservation
    • The code models the cosmological phenomena called Reionization
    • Simulates the ionization map of the universe in a simulation box
    • Similar algorithms are performed over each grid cell inside the simulation box
  10. Team 10 (NCRA Pulsar / FRB Team)
    • Search for FRBs via the Giant Metrewave Radio Telescope (GMRT) in real time


Outcome

During this hackathon, maximum speed-up achieved by one of the team is 230X and minimum 17X over baseline. Teams were satisfied and expressed interest in participating in forthcoming hackathons.



Acknowledgement

The efforts of the CDAC team are gratefully acknowledged. We appreciate the efforts of following members for their contribution in support and exemplary, execution of the event

  1. Prachi Kakade
  2. Gayatri Gomase
  3. Himanshu Sharma
  4. Harikesh Shinde
  5. Ankur Agnihotri
  6. Shubham Mehta
  7. Samir Shaikh
  8. Om Jadhav