NASA and AI: The GPU Race Threatens Modern Astronomy

⚡

Key Takeaways

1NASA will launch the Nancy Grace Roman Space Telescope in September 2026, eight months earlier than planned.

2The James Webb Telescope and the Vera C. Rubin Observatory are generating massive volumes of data requiring GPUs for analysis.

3Astrophysicist Brant Robertson is using AI models to process this data but is facing a GPU shortage.

💡Why it matters — The growing demand for GPUs in astronomy could hinder scientific advancements if access remains limited.

An Early Launch for the Nancy Grace Roman Telescope

NASA recently announced that the Nancy Grace Roman Space Telescope will be launched into orbit in September 2026, eight months earlier than initially planned. This telescope promises to revolutionize astronomy by providing scientists with a colossal volume of 20,000 terabytes of data throughout its mission. This technological advancement comes at a time when the collection and analysis of astronomical data are becoming increasingly important.

An Avalanche of Astronomical Data

The James Webb Space Telescope, operational since 2021, transmits 57 gigabytes of images daily, contributing to the massive accumulation of data for astronomers. Meanwhile, the Vera C. Rubin Observatory, located in the mountains of Chile, will soon begin collecting 20 terabytes of data each night. In comparison, the Hubble Telescope, once the gold standard, produces only 1 to 2 gigabytes per day. This explosion of data requires powerful tools to be analyzed effectively.

The Rise of GPUs in Astronomical Analysis

In the face of this mountain of data, astronomers are turning to GPUs for efficient analysis. Brant Robertson, an astrophysicist at UC Santa Cruz, has observed this transformation over his 15 years of collaboration with Nvidia. He has applied GPUs to advanced simulations to test theories about supernovae and has developed tools to analyze data from new observatories. According to Robertson, there has been a notable shift from CPU-based analyses to those accelerated by GPUs.

Morpheus: AI Serving Astronomy

Robertson, along with his former student Ryan Hausen, created a deep learning model named Morpheus. This model is capable of processing vast datasets to identify galaxies. Their initial analysis of data from the Webb telescope revealed an unexpected number of disk galaxies, enriching theories about the evolution of the universe. Morpheus is also evolving with the times: Robertson is shifting its architecture from convolutional neural networks to transformers, which are at the heart of the rise of large language models. This will allow the model to analyze several times more surface area than currently possible.

Enhancing Ground-Based Observations

Robertson is also working on generative AI models trained on data from space telescopes to improve the quality of observations collected by ground-based telescopes, which are often distorted by the Earth's atmosphere. Despite advancements in rocket technology, it is still challenging to place an 8-meter mirror into orbit, so using software to enhance Rubin's observations is the best alternative.

The Challenges of Accessing GPUs

Despite these advancements, Robertson faces the global shortage of GPUs. Although he has built a GPU cluster at UC Santa Cruz with the help of the National Science Foundation, it is quickly becoming outdated. The growing demand for computationally intensive analyses highlights the need for access to more powerful resources. The Trump administration had proposed a 50% cut to the NSF budget, which could further complicate access to the necessary resources for this research. Robertson emphasizes the importance of being entrepreneurial in this field, as universities, often limited in resources, are reluctant to take risks. It is crucial to demonstrate the significance of these technologies for the future of astronomy.