Powering the Future of Meteorological Computation

Where precision meets performance: Explore the engine behind India's advanced weather and climate predictions.

The Compute Infrastructure of the ARKA HPC System at IITM, Pune, sets a new benchmark in computational excellence. With a peak performance capacity of 11.77 Petaflops and an additional dedicated 1.9 Petaflops for Artificial Intelligence and Machine Learning tasks, this facility is a cornerstone for reliable weather forecasting and climate research.

These cutting-edge resources empower the Ministry of Earth Sciences to:

  • Increase the horizontal resolution of weather prediction models to 6 km globally.
  • Enable regional simulations at resolutions as fine as 1 km.
  • Deliver unparalleled accuracy for extreme weather forecasts, including cyclones, heavy rainfall, and drought predictions.

36 Compute Node Racks: Hosting 3,021 CPU nodes equipped with AMD Milan 7643 processors, each offering 48 cores at speeds up to 3.60 GHz.

26 GPU Nodes: Featuring NVIDIA A100 GPUs, specifically designed for high-performance parallel computing.

Memory Powerhouse: Over 773.37 TB of memory to handle extensive simulations.

State-of-the-art Technology: Integrated with InfiniBand high-speed connectivity for seamless data processing

BullSequana XH2000 / Global Architecture

Sequana2 Networks cabling

Compute Blade

  • The basic compute blade is built with 3 identical compute nodes per blade.
  • External interfaces to Sequana2 are:
    • Signal connection + high speed interconnect,
    • Management and sideband (SBC) connections.
    • Power (up to 9600W) centralized power to receive through the busbar.
    • Input and output water connection to rack manifold for liquid cooling.

Dragonfly+ Topology

  • Several "groups", connected using all to all links
  • The topology inside each group can be any topology
  • Reduce total cost of network (fewer long cables)
  • Utilizes Adaptive Routing to for efficient operations
  • Simplifies future system expansion