No-cost GPUs

Graphics processing units (GPUs) can be incredibly useful in helping you to speed your computation time. 

Curious about how to get access to this kind of computing power? We've developed some recommendations for what we consider to be the best resources to explore, depending on your use case or need.

• Dedicated support: NC State HPC

  Best for: NC State researchers who want local technical assistance with selecting resources, preparing job scripts, installing software, troubleshooting errors, improving performance or moving an existing workflow onto a computing cluster.

  GPU details: Hazel includes multiple generations of NVIDIA GPUs. The following are currently documented:

  • H200 GPUs with 141 GB of high-bandwidth memory
  • H100 GPUs with 80 GB of high-bandwidth memory
  • A100 GPUs with 40 GB of memory
  • L40 and L40S GPUs with 48 GB of memory
  • A30 and A10 GPUs with 24 GB of memory
  • Older P100, RTX 2080 and GTX 1080 GPUs for smaller development and testing workloads
• Getting started: NC State HPC

  Best for: Researchers, graduate students and research groups that are new to GPU computing and want a supported first step before applying for a larger regional or national resource.

  GPU details: A10 and A30 GPUs are suitable for development, inference and smaller models. A100 GPUs are a strong next step for deep learning and scientific computing. Researchers can move to H100 or H200 GPUs when they need more memory.

  Note: NCSU HPC is a shared resource that uses a job submission queue, so jobs may wait for resources to become available before running.

  Runner-up options:

  • NC State VCL: Useful for short-term, remotely accessible computing environments. VCL allows NC State users to reserve a computer with a selected software environment and connect to it online. An A100 environment may be an option when it is available to the researcher’s group.
  • Google Colab: Useful for tutorials, notebooks and short experiments. The free tier is easy to access, but GPU availability, session duration and hardware type are not guaranteed.
• Funded projects: NC State HPC (Hazel partner)

  Best for: Funded research groups that need predictable GPU capacity, higher priority or a dedicated queue.

  GPU details: The GPU hardware can be selected to match the project’s needs. Partner resources receive a dedicated queue, while the project also receives increased priority across shared Hazel queues.

• Dedicated access: NCShare

  Best for: Large AI models, advanced inference and other workloads that need high-end GPUs with substantial memory.

  GPU details: NCShare provides access to 32 NVIDIA H200 GPUs. H200 GPUs offer large high-bandwidth memory and are designed for demanding AI, machine learning and scientific-computing workloads.

• Advanced workloads: ACCESS

  Best for: National-scale AI, multi-GPU training, large simulations and projects that exceed local computing capacity.

  GPU details: Available options include A100, H100, H200, L40S and Grace Hopper GH200 GPUs. Researchers can choose Delta for general GPU computing, DeltaAI for specialized AI, FASTER for multi-GPU work or Jetstream2 for interactive GPU environments.

  For more information on ACCESS resources and getting started with an allocation, please feel free to contact the Research Facilitation Service.

Please feel free to contact the Research Facilitation Service with any questions!