AI Galaxy Hunters Are Adding to the Global GPU Crunch

NASA is set to launch the Nancy Grace Roman space telescope in September 2026, ahead of schedule. This new telescope is expected to generate a massive 20,000 terabytes of data over its operational life. This influx of data will add to the daily 57 gigabytes already being downlinked from the James Webb Space Telescope and the anticipated 20 terabytes per night from the Vera C. Rubin Observatory. In comparison, the Hubble Space Telescope, once the benchmark, provided only 1-2 gigabytes daily.

Astronomers are increasingly relying on Graphics Processing Units (GPUs) to process these vast datasets. Brant Robertson, an astrophysicist at UC Santa Cruz, has been at the forefront of this shift, collaborating with Nvidia for 15 years to apply GPUs to space-related challenges. His work has evolved from simulating supernova explosions to developing tools for analyzing data from new observatories.

Robertson noted the evolution from analyzing a few objects to large-scale CPU-based analyses, and now to GPU-accelerated versions of these same analyses.

Morpheus: An Evolving AI for Galaxy Identification

Robertson and his former graduate student, Ryan Hausen, developed Morpheus, a deep learning model capable of sifting through large datasets to identify galaxies. An early application of Morpheus to Webb data revealed a significant number of disc galaxies, prompting a re-evaluation of theories on universal development.

Morpheus is now being updated to incorporate transformer architecture, similar to that used in large language models, moving away from convolutional neural networks. This architectural shift is expected to significantly increase the model's analysis area and speed.

Generative AI for Enhanced Observations

Robertson is also exploring generative AI models trained on space telescope data to improve the quality of observations from ground-based telescopes, which are often affected by atmospheric distortion. This software-based enhancement is seen as a crucial next step, given the challenges of placing large mirrors into orbit.

The GPU Demand and Resource Constraints

Despite advancements, the growing demand for GPU access is a significant challenge. Robertson has utilized National Science Foundation (NSF) funding to establish a GPU cluster at UC Santa Cruz, but it is rapidly becoming outdated as more researchers adopt compute-intensive techniques. He highlighted the need for entrepreneurial approaches, especially when working at the technological frontier, and the inherent risk aversion of universities with constrained resources.

Robertson emphasized that GPUs are essential for modern AI and Machine Learning analyses, and researchers must be proactive in demonstrating the field's direction to secure necessary resources.