GPU Resources

Why have Graphics Processor Units (GPUs) in the Clusters?

Graphics Processing Units (GPUs) dramatically speed up the parallel mathematical operations at the heart of machine learning, molecular dynamics, image and video processing, rendering, and photogrammetry. Our clusters are built with GPU acceleration so researchers can take advantage of that speed-up without managing hardware of their own. Our GPUs live within compute nodes inside OrangeGrid and Zest, and researchers request them as part of a normal job submission.

 Access to GPUs will enhance research opportunities for Syracuse students. Undergraduates and graduate students will gain practical experience with cutting-edge computing architectures.
NSF award enhances Syracuse University computational capabilities through GPU cluster

Frequently Asked Questions:

What specific computing needs does it serve?

GPUs provide a significant speed increase over CPUs for certain classes of computation. On campus they are used for training and running machine learning and large language models, GPU-accelerated simulation in molecular dynamics, chemistry, and physics, computer vision and image processing, and rendering and photogrammetry projects of all scales and sizes. They also give students and researchers hands-on experience developing with CUDA and other GPU programming frameworks.

What hardware and system configurations are available?

Hundreds of NVIDIA GPUs across a range of generations and memory sizes. The largest share is in OrangeGrid, with a smaller pool in Zest.

Available GPU Models:

– NVIDIA H100 80GB HBM3
– NVIDIA A100 80GB PCIe
– NVIDIA L40S (48 GB)
– NVIDIA A40 (48 GB)
– NVIDIA Quadro RTX 6000 (24 GB)
– NVIDIA Quadro RTX 5000 (16 GB)

Both clusters are Linux non-interactive environments. NVLink is available with the A100 and H100 models. CUDA is supported on every GPU node, and researchers can install the CUDA version their software needs into their own Conda or UV environment as well as utilize containers via Singularity.

Please see our GPU Computing page for more specifics.

Who can use this system?

GPU resources are available at no cost to researchers affiliated with Syracuse University working on faculty-sponsored research. Access begins with a short consultation so we can match your workload to the right cluster.

Who can provide assistance in using this system?

Feel free to contact us by emailing researchcomputing@syr.edu or review more details on our Requesting Access page.

Does this system replace or complement OrangeGrid or Zest?

Neither. GPUs are part of both clusters. OrangeGrid holds most of the GPUs and suits many independent GPU jobs such as inference, parameter sweeps, and single-node training. Zest suits GPU jobs that need to span nodes over its InfiniBand interconnect or run for weeks.

Why have Graphics Processor Units (GPUs) in the Clusters?

Frequently Asked Questions:

GPU Hardware and Rendering Images