ERCAP (Energy Research Computing Allocations Process)

ERCAP (Energy Research Computing Allocations Process) allocates compute time on NERSC's Perlmutter supercomputer at Lawrence Berkeley National Laboratory, one of the world's largest GPU-based systems with over 7,000 NVIDIA A100 GPUs alongside AMD EPYC CPUs. ERCAP is the primary mechanism for the broader DOE Office of Science research community (and partner researchers) to access leadership-class GPU resources for AI/ML, foundation model training, scientific AI, climate AI, materials AI, and AI-coupled simulation. Annual allocation cycle: submissions typically open in August and close in early October, with allocations starting January 1 of the following year. The 2026-cycle ERCAP call closed October 8, 2026 for 2027 allocations. NERSC supports both pre-trained foundation model fine-tuning and pre-training campaigns, with a growing AI/ML user community. Recipients receive both compute hours and consultation from NERSC's NESAP for AI program providing hands-on optimization, MLOps support, and access to advanced features (interactive nodes, Jupyter at scale, ML-optimized storage). ERCAP complements DOE's INCITE (largest) and ALCC (mission-aligned) allocation pathways.

NERSC ERCAP Energy Research Computing Allocations on Perlmutter for AI and ML is sponsored by U.S. Department of Energy Office of Science (NERSC at Lawrence Berkeley National Laboratory). ERCAP (Energy Research Computing Allocations Process) allocates compute time on NERSC's Perlmutter supercomputer at Lawrence Berkeley National Laboratory, one of the world's largest GPU-based systems with over 7,000 NVIDIA A100 GPUs alongside AMD EPYC CPUs.

The DOE INCITE (Innovative and Novel Computational Impact on Theory and Experiment) program is the primary mechanism for distributing leadership-class compute time on the U.S. Department of Energy's exascale systems — Aurora at Argonne and Frontier at Oak Ridge — to support high-impact AI and computational science research. INCITE specifically supports large-scale AI training including foundation models, multimodal architectures, scientific AI, and AI-driven scientific discovery. Individual awards range from 500,000 to 1,000,000 node-hours per project per year (equivalent to multi-million-dollar compute value), making INCITE one of the largest sources of leadership-class GPU/accelerator time globally. The 2026 call emphasized AI for science, foundation models for scientific discovery, large-scale generative AI, and AI-coupled simulation. The proposal review process emphasizes computational readiness, scientific merit, and ability to use leadership-class resources effectively. Open to academic, industry, and government researchers worldwide regardless of source of project funding. Annual call typically opens April-June with awards starting January of the following year. Companion programs include ALCC and ERCAP for additional compute allocations.