GPU Computing
GPU (Graphical Processing Unit) computing accelerates complex computational tasks by leveraging highly parallel architectures designed for massive data processing. These systems are particularly effective for AI training, deep learning, image processing, scientific simulations, and data analytics. By utilizing GPUs, researchers can significantly enhance performance compared to traditional CPU-based computing.
Key Components of GPU Computing
- GPU Hardware: High-performance GPUs from vendors such as NVIDIA (A100, H100) and AMD, designed for large-scale parallel processing.
- Software and Frameworks: GPU-enabled software through CUDA or OpenCL, and specialized AI/ML frameworks like TensorFlow, PyTorch, JAX, and RAPIDS optimize workloads for GPU acceleration.
- Multi-GPU and Distributed Computing: Capability to scale computations across multiple GPUs and nodes, improving efficiency for large-scale workflows.
- Job Scheduling and Resource Management: HPC systems implement job schedulers (such as Slurm) that efficiently allocate GPU resources among users.
Available Resources
Quest
Quest is Northwestern’s high-performance computing (HPC) cluster, providing robust and powerful GPU resources for researchers. Quest offers both no-cost General Access to University-supported GPU resources, as well as the option to purchase dedicated GPU resources for use by your research group.
Common Use Cases:
- Deep Learning, AI Training, and Inference
- Scientific Computing and Simulations
- Data Analytics and Processing
Key Features:
- Multi-GPU Support: Quest includes NVIDIA Tesla A100 and H100 GPUs in various configurations in General Access:
- 24 nodes, each with 4 H100 SXM (80 GB) cards
- 18 nodes, each with 4 80 GB A100 SXM cards
- 16 nodes, each with 2 40 GB A100 PCIe cards
- CUDA Toolkit: A parallel computing platform and programming model for NVIDIA GPUs, to increase computing performance. Quest GPU resources can support multiple versions of the modern CUDA Toolkit.
- Framework Compatibility: Works with TensorFlow, PyTorch, JAX, RAPIDS, and other machine learning and AI libraries.
Further details on Quest’s architecture are available in the Quest Technical Specifications.
The Genomics Compute Cluster (GCC) resides within Quest infrastructure and provides GPU resources for Northwestern researchers from any school or center who are engaged in genomics research.
External Resources
Several federal programs provide access to GPU resources.
- ACCESS: Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) is a National Science Foundation (NSF) program providing researchers access to extensive GPU computing resources.
- NAIRR: The National Artificial Intelligence Research Resource (NAIRR) Pilot program offers access to a wide variety of GPU resources and architectures. While some of the available resources are also part of the ACCESS program, NAIRR also includes GPU systems from commercial cloud computing providers.
- National Laboratory Clusters: Researchers working on projects funded by the Department of Energy (DOE) may have access to computing clusters operated by national laboratories. DOE computing facilities with GPU resources include the Oak Ridge National Laboratory (ORNL) Leadership Computing Facility, Argonne Leadership Computing Facility (ALCF), and the NERSC (National Energy Research Scientific Computing Center).
Research Computing and Data Services computational specialists can assist you in evaluating external resources and submitting an application. Connect with a computational specialist to discuss your needs and the options.