ADSP Call Details

Call for Proposals

The ALCF Data Science Program (ADSP) employs a competitive proposal process that awards allocations of compute time and data storage on ALCF supercomputers. All proposals are peer reviewed by a panel of experts for both the scientific impact and computational readiness. ADSP awards span two years and researchers from academia, government research facilities, and industry are welcome to apply. PIs will be required to submit a renewal application for the second year of the award.

If you have any questions, contact

SUBMISSION DEADLINE: February 26, 2021, 6pm US Central Time

Program Overview

Ongoing and past ADSP projects span a diverse range of science domains, e.g. Materials, Imaging, Neuroscience, Engineering, Combustion/CFD, Cosmology; and involve large science collaborations (APS, LSST, DESC, LIGO, DES, ATLAS) and smaller research groups developing machine learning methods at scale. ADSP projects benefit from hardware and system architectures at ALCF which support data analysis and machine learning with a common software stack to allow for large scale science campaigns. The program also benefits the selected projects by offering directed assistance and computational time to gain experience scaling codes on ALCF systems, enabling future INCITE and ALCC proposals for DOE supercomputing systems.

Scientists at the ALCF partner with each ADSP project assisting in code and methods development, optimization, workflow creation, data analysis, and visualization. Example techniques and areas of research that leverage ALCF’s experience include hyperparameter optimization, uncertainty quantification, statistics, machine learning, deep learning, databases, pattern recognition, image processing, graph analytics, data mining, real-time data analysis, and complex interactive workflows.

ALCF Computing Resources

ADSP project teams can request allocations on the following ALCF computing resources:

  • Theta has 4,392 KNL nodes and, newly added in Fall 2020, 24 DGX-A100 nodes:
    • Each KNL node has a Intel Knights Landing (KNL) processor with 64 1.3GHz cores, 16 GB of high-bandwidth in-package memory (HBM), 192 GB of DDR4 RAM, and a 128GB node-local SSD. The aggregate peak compute speed is 11.69 petaflops.
    • Each DGX-A100 node has 8 NVIDIA A100 GPUs and 2 AMD Rome CPUs. More details to come. 
  • Cooley is a visualization and analysis cluster with 126 compute nodes; each node has 12 CPU cores and one NVIDIA Tesla K80 dual-GPU card. The entire Cooley system has a total of 47 terabytes of system RAM and 3 terabytes of GPU RAM.
  • ALCF systems are all backed by our 10PB Lustre and 7PB GPFS systems. 
  • NEW Polaris system is expected to arrive in the Summer 2021 timeframe. Polaris is planned to be a hybrid CPU/GPU machine that will be available to ADSP teams. More details to come.

Anticipated Polaris Configuration

System Peak
35-45 PF
Peak Power
<2 MW
Total System Memory
>250 TB
System Memory Type
Node Performance
>70 TF
Node Processors
1 CPU; 4 GPUs
System Size
>500 nodes
Node-to-Node Interconnect
200 Gb
Programming Models
OpenMP 4.5/5, SYCL, Kokkos, Raja, HiP
GPU tools, PAPI, TAU, HPCToolkit, DDT
Python/Numba, TensorFlow, PyTorch

ADSP Resources

  • Staff and Postdoc Support: The chosen ADSP projects will receive support from the Data Science group, a multidisciplinary team of scientists and high-performance computing software engineers. 
  • Training and Hardware Access: The ALCF will offer one-on-one assistance for R&D. Depending on the requirements of the projects, this may include a detailed introduction to the hardware and software stack, access to early hardware, deep dives on specific hardware features, and customized tutorials.
  • Computing and Storage Resources: ADSP projects will be awarded compute time and storage space on ALCF systems. 
    • Annual awards are expected to be roughly:
      • 20-100 thousand node-hours on Theta-KNL nodes,
      • 1-5 thousand DGX node-hours on Theta-GPU nodes, or
      • 10-50 thousand node-hours on the new Polaris system.
    • The size of the request should be commensurate with the size of the scientific impact.
    • Second year awards will be based on progress in the first year and consultation with the project’s PI. 
    • Initial storage requirements may be up to 100 terabytes; additional project needs will be accommodated in consultation with the ALCF.

Eligibility Requirements

This call is open to US- and non-US-based researchers and research organizations in universities, academia, industry, national laboratories and other research institutions needing large allocations of computing time, supporting resources, and data storage. DOE sponsorship is not required to participate. Note that there are federal laws regulating what can be done on ALCF systems. As an example, Classified Information, National Security Information or Unclassified Controlled Nuclear Information cannot be stored on our systems.

Reporting Requirements

ADSP project teams are expected to provide quarterly progress reports, to participate in update calls, collaborate with ALCF staff, help prepare highlights of notable accomplishments and results, and provide a written report at the end of the project.

Submission Instructions

Evaluation of Proposals

Proposals will be evaluated on the strength of:

  • Potential impact of proposed science on the respective domain.
  • Demonstrated scalability of the target application on the targeted resource or on a comparably sized computing cluster.
  • Describe the datasets to be used or generated, including size and resources needed to generate it.
  • Appropriateness of development team: the likelihood that project member’s expertise and person-hours proposed are likely to accomplish the science goals or software development described.
  • Overall diversity of science domains and algorithms.