August 11, 2022

Accelerating Machine Learning Through Collaboration

PNNL’s collaboration with SambaNova aims to enhance scientific computing

Artist's representation of artificial intelligence

PNNL researchers teamed up with SambaNova to enhance artificial intelligence for scientific computing.

(Image by bettervector |

Artificial intelligence (AI) can enhance scientific discoveries across domains—from identifying potential drugs to combat new pathogens to aiding nuclear nonproliferation efforts. Now, a new collaboration between researchers at Pacific Northwest National Laboratory (PNNL) and SambaNova Systems, Inc. aims to make AI systems better and more user-friendly for domain scientists.

Though the interaction initially began through conversations with Kunle Olukotun, SambaNova’s Chief Technologist and Co-founder, PNNL’s Chief Scientist for Computing James (Jim) Ang saw an opportunity for collaboration that could meet each entity’s needs.

As leader of the Data-Model Convergence (DMC) initiative, Ang was already working to merge PNNL’s historical strengths in physical sciences and engineering with AI, machine learning, and graph analytics through the DMC’s converged applications projects. He saw the potential for SambaNova’s AI system to fit in this research ecosystem.

Ang secured support from a PNNL institutional investment to bring a SambaNova system to PNNL’s campus through the research computing capability. He then enlisted the help of Kevin Barker, leader of the High Performance Computing Group, and Roberto Gioiosa, director of the co-design Center for Artificial Intelligence-focused Architectures and Algorithms (ARIAA) to advance this partnership. The two worked quickly to bring a SambaNova system to PNNL for testing.

Testing new technologies to meet Department of Energy (DOE) needs is a specialty of Barker’s. As Program Manager of the Center for Advanced Technology Evaluation (CENATE), he frequently partners with industry, academia, and other national laboratories to assess the capabilities of emerging computing technologies.

“CENATE’s testbed is the ideal place for industry partners, such as SambaNova, to test the potential impact of novel or disruptive technologies,” said Barker.

CENATE researchers work in close collaboration with researchers from the DMC initiative and ARIAA, creating a well-designed ecosystem for advancing the Laboratory’s computing capabilities.

For ARIAA, the partnership with SambaNova complements one of its main objectives: to co-design novel architectures, algorithms, and compiler and programming abstractions to enable AI-based DOE applications. Gioiosa was already working with domain scientists to explore how AI and machine learning can support the power gridcybersecuritygraph analytics, and computational chemistry through ARIAA. Now with SambaNova, PNNL researchers can explore a holistic “hardware-software co-design” so that each partner—SambaNova on the hardware side and PNNL on the software side—could synergistically create a system that meets each other’s needs.

“Ideally, we want to create a good system for domain scientists to interact with AI,” said Gioiosa. “This collaboration is meant to create a legacy for PNNL and the scientific community, and to help bring research computing beyond GPUs in future post-exascale systems.”

CENATE and ARIAA are supported by the DOE Office of Science, Office of Advanced Scientific Computing Research (ASCR).  DMC is a PNNL Laboratory Directed Research and Development initiative.