RANGE (Rice AI Networked GPU Engine), the Center for Research Computing’s next-generation NVIDIA GPU cluster designed to accelerate AI and data-driven research across all disciplines.  RANGE combines cutting-edge NVIDIA GPUs with ultra-fast NVIDIA NDR networking to provide both increased capacity and capability for Rice researchers.  RANGE has access to the CRC'S VAST high performance data share and the RHF (Research High-Capacity Facility) to store large datasets.  RANGE utilizes the SLURM batch scheduler, LMOD for managing software environments, and provides preinstalled suites of many commonly used AI/ML software tools.

Citation

If you use RANGE to support your research activities, please acknowledge (in publications, on your project web pages, …) our efforts to maintain this infrastructure for your use. An example acknowledgment that can be used follows. Feel free to modify wording for your specific needs but please keep the essential information:

This work was supported in part by the RANGE cluster operated by Rice University's Center for Research Computing (CRC).

Node Configurations 

RANGE has four nodes designed to provide capacity for educational use and the various courses, workshops, and trainings that increasingly include AI workloads. Examples of capacity workloads would include tuning, repurposing, and using existing foundation models in the 10 billion parameter range.

RANGE has 8 nodes designed to support the growing models and increasing computational complexity of modern ML algorithms. Capability systems will function both as production resources for the needs of campus research groups as well as prototyping environments for researchers interested in the national centers and their AI computing efforts. A practical example characterizing the scale of problems requiring capability systems would be training of the BLOOM Large Language Model (https://arxiv.org/abs/2211.05100) 

Software/OS Environment

• Operating system: Red Hat Enterprise Linux 9.5

• SLURM scheduler: version 24.11

• CUDA libraries: Run module spider CUDA on the cluster to see versions available