Bridges and Bridges-AI are available at no cost for COVID-19 research. To apply, submit your request at the COVID-19 HPC Consortium website. Contact us at firstname.lastname@example.org with any questions.
Accessing Bridges and Bridges-AI for COVID-19 Research
The Pittsburgh Supercomputing Center’s Bridges supercomputer, including its Bridges-AI platform, enables high-performance computing (HPC), scalable artificial intelligence (AI), high-performance data analytics (HPDA), and workflows requiring advanced accelerators, large memory, web portals, and high-performance access to Big Data. Bridges and Bridges-AI emphasize a flexible software environment and interactive access for user productivity. Bridges and Bridges-AI are supported by NSF award number 1445606.
How Bridges and Bridges-AI Can Help COVID-19 Research
Bridges and Bridges-AI offer exceptional capabilities rarely found elsewhere for critical areas of research. PSC’s expert User Support staff are available to promptly help get new projects started. Areas where Bridges and Bridges-AI offer specific, unique strengths are:
- Artificial Intelligence: Bridges-AI delivers scalable deep learning for memory- and compute-intensive networks (e.g., SciBERT) through an NVIDIA DGX-2 (16 tightly-coupled Volta GPUs and 1.5TB RAM) and nine HPE servers with 8 Volta GPUs each.
- Genomics: Bridges’ large-memory servers with 12TB and 3TB of RAM are the premier resource for de novo sequence assembly, and Bridges as a whole is well-suited to variant calling and bioinformatic workflows.
- Collaborative workflows: Secure web portals can be configured using Bridges’ servers that are dedicated to serving persistent databases and websites, allowing efficient coordination among distributed teams.
Bridges and Bridges-AI are available at no cost for COVID-19 research. To apply, please submit your request at the COVID-19 HPC Consortium website. If you have any questions, please contact us at email@example.com.
Bridges and Bridges-AI Overview
The Pittsburgh Supercomputing Center’s Bridges supercomputer pioneered the convergence of high-performance computing, artificial intelligence, and Big Data. Bridges includes 752 dual-CPU HPC servers, 4 extreme-memory servers each with 12TB of RAM, 34 large-memory servers each with 3TB of RAM, and 64 GPU-accelerated servers for HPC and AI. Bridges-AI, a recent expansion of Bridges, delivers extreme scalability artificial intelligence with an NVIDIA DGX-2 (16 Volta GPUs, NVswitch, 1.5TB RAM) and 9 HPE Apollo 6500 servers, each with 8 NVLink-connected NVIDIA Volta GPUs. Bridges prioritizes user productivity and flexibility and is supported by PSC’s User Support experts.
The Bridges and Bridges-AI hardware and software configurations are as follows. Each hardware resource type includes examples of applications that it optimally supports.
Bridges-AI: Deep learning, machine learning, graph analytics
- 1 NVIDIA DGX-2 node with 16 NVIDIA V100 32GB SXM2 GPUs, 2 Intel Xeon Platinum 8168 CPUs, 1.5TB RAM, and 30TB NVMe SSD
- 9 HPE Apollo 6500 Gen10 nodes, each with 8 NVIDIA V100 16GB SXM2 GPUs, 2 Intel Xeon Gold 6148 CPUs, 192GB RAM, and 8TB NVMe SSDs
Bridges-RM: HPC and HTC
- 752 Regular Memory nodes, each with 2 Intel Xeon E5-2695v3 CPUs (14c, 2.3/3.3 GHz), 128GB RAM, and 4TB local HDD
Bridges-LM: Genomics (de novo assembly), graph analytics, data analytics, memory-intensive HPC
- 2 Extreme Memory nodes, each with: 12TB DDR4-2400 RAM, 16 Intel Xeon E7-8880v4 CPUs (22c, 2.2/3.3 GHz), and 56TB local HDD
- 2 Extreme Memory nodes, each with 12TB DDR4-2133 RAM, 16 Intel Xeon E7-8880v3 CPUs (18c, 2.3/3.1 GHz), and 56TB local HDD
- 34 Large Memory nodes, each with 3TB DDR4-2400 RAM, 4 Intel Xeon E7-8870v4 CPUs (20c, 2.1/3.0 GHz), and 16TB local HDD
- 8 Large Memory nodes, each with 3TB DDR4-2133 RAM, 4 Intel Xeon E7-8860v3 CPUs (16c, 2.2/3.2 GHz), and 16TB local HDD
Bridges-GPU: Deep learning, machine learning, GPU-accelerated HPC
- 32 GPU-P100 nodes, each with 2 NVIDIA Tesla P100 16GB GPUs, 2 Intel Xeon E5-2683 v4 CPUs (16c, 2.1/3.0 GHz, 40MB LLC), 128GB RAM, and 4TB local HDD
- 16 GPU-K80 nodes, each with 4 NVIDIA Tesla K80 GPUs (2 cards), Intel Xeon E5-2695 v3 CPUs (14c, 2.3/3.3 GHz, 35MB LLC), 128GB RAM, and 4TB local HDD
Bridges-DB and Bridges-Web: Persistent databases, portals, web interfaces
- 6 DB-s nodes, each with 2 Intel Xeon E5-2695 v3 CPUs (14c, 2.3/3.3 GHz, 35MB LLC), 128GB RAM, and 2TB SSD
- 6 DB-h nodes, each with 2 Intel Xeon E5-2695 v3 CPUs (14c, 2.3/3.3 GHz, 35MB LLC), 128GB RAM, and 18TB HDD
- 6 Web nodes, each with 2 Intel Xeon E5-2695 v3 CPUs (14c, 2.3/3.3 GHz, 35MB LLC), 128GB RAM, and 12TB HDD
- Intel Omni-Path Architecture (100Gbps), custom leaf-spine topology
Persistent Data Storage
- 10PB, Lustre
Relevant Community Datasets
- CORD-19; other community datasets as required
- CentOS 7.6 (except NVIDIA DGX-2) and Ubuntu 18.04 (NVIDIA DGX-2)
- Interactive access
- Jupyter, Anaconda, R, MATLAB, including on LM (12TB, 352c) nodes
- CUDA 10.1
- NGC Container Singularity repo
- Intel, PGI, and GNU compilers
- Singularity, VMs