High Performance Computing (HPC)

Understanding High Performance Computing (HPC)

High Performance Computing (HPC) systems are engineered to deliver extraordinary processing power. Such systems are outfitted with fast networking, sophisticated storage solutions, cutting-edge GPUs, and substantial memory allocations to meet the rigorous demands of today’s most challenging computational and memory-intensive tasks.

Introducing Innovator: SDSU RCi's New HPC System

SDSU RCi is excited to unveil its newest HPC system, named Innovator. Functioning on the robust Linux Rocky 9 platform, Innovator is a testament to modern computing, with an array of nodes powered by advanced Intel processors. These nodes are interlinked by a state-of-the-art, high-velocity communication network that facilitates impeccable data transfer and coordination of complex tasks.

Innovator represents a multifaceted HPC cluster, integrating nodes outfitted with standard CPU processors alongside select nodes equipped with the formidable NVIDIA A100 GPUs. This amalgamation is unified by a high-speed InfiniBand network, which is key to enabling efficient parallel processing. Delineated into four distinct partitions—bigmem, compute, gpu, and quickq—the cluster spans an impressive 105 nodes in total. Each partition is fine-tuned for specific computational requirements, thus offering a dynamic and robust platform for a multitude of scientific explorations and intense computational proceedings.

Detailed Overview of Cluster Partitions

1. Bigmem Partition:

   • Usage: Designed for tasks requiring large memory.
   • Nodes: Includes 4 nodes named bigmem001 to bigmem004.
   • CPUs: Each node has 48 CPUs.
   • CPU Details:
     • Architecture: x86_64, supporting both 32-bit and 64-bit modes.
     • Cores/Threads: 24 cores per socket, 1 thread per core, and 2 sockets per node.
     • CPU Frequency: Ranging from 800 MHz to 3500 MHz.
     • Cache: L1d cache of 2.3 MiB, L1i cache of 1.5 MiB, L2 cache of 60 MiB, L3 cache of 72 MiB.
     • NUMA Nodes: 4, for optimized memory access.
   • Memory: 2TB per node.
   • Unique Features: High memory allocation, suitable for memory-intensive tasks.

2. Compute Partition:

   • Usage: General-purpose computing tasks.
   • Nodes: Includes 41 nodes (node001 to node041).
   • CPUs: Each node comes with 48 CPUs. The CPUs are Intel Xeon Gold 6342 CPUs @ 2.80GHz, offering robust processing power.
   • CPU Details: Identical to Bigmem node.
   • Memory: 256 GB per node.
   • Unique Features: Wide range of memory availability, catering to a variety of computing needs.

3. GPU Partition:

   • Usage: Tasks requiring GPU resources.
   • Nodes: Includes 14 nodes (gpu001 to gpu014)
   • CPUs: Each node has 48 CPUs.
   • CPU Details: Identical to Bigmem node.
   • Memory: 512 GB per node.
   • Unique Features:  Equipped with NVIDIA A100 80GB PCIe GPUs (2 per node), ideal for parallel processing and graphics-intensive applications.

4. Quickq Partition:

   • Usage: Short-duration tasks (12 hours max).
   • Nodes: Includes 46 nodes (node001 to node046).
   • CPUs: 48 CPUs per node.
   • CPU Details: Identical to Bigmem node.
   • Memory: 256 GB per node.
   • Unique Features: Shorter maximum job time limit compared to other partitions, optimized for quick-turnaround tasks.

The Potential of HPC at Your Fingertips

Our HPC system provides immediate linkage to high-performance storage and a plethora of software resources, including but not limited to widely-utilized programming languages such as Python, R, Matlab, and various C compilers. This confluence of technology creates a computational ecosystem that is both adaptable and potent, capable of catering to any specialized requirement. Researchers from myriad academic departments throughout the university harness the power of our HPC assets on a daily basis, significantly propelling the velocity and efficiency of their research endeavors.

Questions or Problems

You can reach us anytime by filling out this form or by email at SDSU.HPC@sdstate.edu

SDSU RCi