Galaxy-Hunting AI Intensifies GPU Crunch

NASA has announced an accelerated timeline for the launch of the Nancy Grace Roman Space Telescope, now set to embark on its mission in September 2026, eight months ahead of its original schedule. This groundbreaking observatory is projected to deliver an astonishing 20,000 terabytes of data to astronomers throughout its operational lifespan.

This immense data stream will complement the continuous flow of information from existing and upcoming missions. The James Webb Space Telescope, operational since 2021, already downlinks 57 gigabytes of breathtaking imagery daily. Later this year, the Vera C. Rubin Observatory in the Chilean mountains will commence its survey, expected to gather an impressive 20 terabytes of data every night.

For context, the Hubble Space Telescope, once the pinnacle of astronomical observation, provides a comparatively modest 1 to 2 gigabytes of sensor readings daily. While the era of manual data analysis is long past, the sheer volume of information now necessitates advanced computational methods, with astronomers increasingly relying on GPUs to process and interpret these vast datasets.

Brant Robertson, an astrophysicist at UC Santa Cruz, has been at the forefront of this scientific paradigm shift, actively supporting and utilizing data from these transformative missions. For the past 15 years, Robertson has collaborated with Nvidia, applying GPU technology to unravel the mysteries of space. Initially, his work focused on advanced simulations to test theories about supernovae explosions, and he now dedicates his efforts to developing tools for analyzing the torrent of data from the newest observatories.

“There’s been this evolution [from] looking at a few objects, to doing CPU-based analyses on large scales of the data set, to then doing GPU-accelerated versions of those same analyses,” Robertson explained to TechCrunch, highlighting the progression in astronomical data processing.

Robertson, alongside then-graduate student Ryan Hausen, developed Morpheus, a deep learning model capable of sifting through extensive datasets to identify galaxies. Their initial AI-driven analysis of Webb data yielded a surprising discovery: an unexpected abundance of a specific type of disc galaxy, adding a new dimension to existing theories about the evolution of our universe.

Morpheus is continually evolving, with Robertson currently re-architecting its core from convolutional neural networks to the transformer architecture that underpins the success of large language models. This upgrade is expected to enable the model to analyze an area several times larger than its current capability, significantly accelerating its analytical work.

Furthermore, Robertson is engaged in developing generative AI models, trained on data from space telescopes, to enhance the quality of observations collected by ground-based telescopes. These terrestrial observations are often distorted by Earth’s atmosphere. Given the persistent challenge of launching an 8-meter mirror into orbit, utilizing sophisticated software to improve the Vera C. Rubin Observatory’s observations represents a crucial next step.

Despite these advancements, Robertson continues to contend with the global demand for GPU access. He has established a GPU cluster at UC Santa Cruz with support from the National Science Foundation, yet it is rapidly becoming outdated as more researchers seek to apply compute-intensive techniques to their work. This challenge is compounded by the Trump administration's proposal to halve the NSF's budget in its current request.

“People want to do these AI, ML analyses, and GPUs are really the way to do that,” Robertson stated. He emphasized the necessity of an entrepreneurial approach, particularly when operating at the cutting edge of technology. “Universities are very risk averse because they just have constrained resources, so you have to go out and show them that, ‘look, this is where we’re going as a field.’”