Redtail: An AI supercomputing platform for Utah
Utah's AI supercomputer, Redtail, is a statewide AI resource created to provide Utah's institutions of higher education, state organizations, and commercial sector with access to advanced computing resources, training, and support. Redtail will accelerate and support research, education, innovation, and AI workforce development across the state.
Learn more about Redtail
More information about Redtail, including opportunities to access the platform, is available on the overview page.
Infrastructure and technologies
The Redtail AI Factory will provide researchers and innovators across Utah with access to powerful technologies in support of AI applications.
Systems in the Redtail AI Factory, which will support research and innovation across Utah
Computational resources
In aggregate, the Redtail system has
- 33 HPE Cray XD670 nodes
- 264 NVIDIA H200 GPUs
- Each node has eight H200 SXM5 GPUs, and each GPU has 141 GB of high-speed VRAM to support demanding AI workloads and large models
- 3,696 CPU cores
- Each node has two Intel Xeon-P 8570 56-core CPUs
- 66 TB of RAM
- Each node has 2 TB of DDR5-5600 RAM
The architecture is optimized for distributed AI training, large language models (LLMs), multi-node GPU computing, and tightly coupled HPC workloads.
High-speed GPU and node connectivity
The system is designed to enable GPUs to work efficiently together, both within and across nodes:
- Within nodes, there is high-speed GPU-to-GPU communication, providing up to 900 gigabytes per second of bidirectional bandwidth
- Between nodes, each GPU is connected at 400 gigabits per second to an InfiniBand fabric
with 24 NDR switches, enabling large AI workloads to scale across nodes
- The system uses a rail-optimized fat-tree topology to reduce latency and improve scalability
The Redtail platform is also connected to the Utah Education and Telehealth Network (UETN) and the University of Utah's Downtown Data Center at 100 and 400 gigabits per second, respectively, to facilitate data movement to and from other computational infrastructure in Utah.
High-speed storage
The Redtail AI Factory includes high-performance temporary scratch storage for active research workloads, with approximately 1 PB of usable scratch capacity, in addition to high-speed storage local to each compute node. The WEKA filesystem supports high-throughput AI pipelines and scalable, data-intensive research applications.