HPC researchers describe the skies of exoplanets
A paper published today describes in the greatest detail to date the atmospheres of distant planets.
In search of the origins of what lies in and beyond the Milky Way, researchers have studied 25 exoplanets, bodies orbiting stars far beyond our solar system. Specifically, they studied hot Jupiters, the largest and therefore easiest to detect exoplanets, many of which suffocate at temperatures above 3,000 degrees Fahrenheit.
Their analysis of these scorching atmospheres used high-performance computing with NVIDIA GPUs to advance understanding of all planets, including our own.
Hot jupiters shine new lights
Hot Jupiters “provide an incredible opportunity to study physics in environmental conditions almost impossible to reproduce on Earth,” said Quentin Changeat, lead author of the paper and a researcher at University College London (UCL).
By analyzing trends across a large group of exoplanets, they shed new light on big questions.
“This work can help create better models of how Earth and other planets came to be,” said Ahmed F. Al-Refaie, co-author of the paper and head of numerical methods at the Data Center. of space exochemistry from UCL.
Analyzing Hubble Big Data
They used the most data ever used in an exoplanet survey – 1,000 hours of archival observations, mostly from the Hubble Space Telescope.
The hardest and, for Changeat, most fascinating part of the process was figuring out which small set of models to run consistently against data from all 25 exoplanets to get the most reliable and telling results.
“There was an amazing period of exploration – I was finding all sorts of weird solutions sometimes – but it was really quick to get the answers using NVIDIA GPUs,” he said.
Millions of calculations
Their overall results required captivating calculations. Each of the approximately 20 models had to run 250,000 times for the 25 exoplanets.
They used Cambridge University’s Wilkes3 supercomputer, which packs 320 NVIDIA A100 Tensor Core GPUs on an NVIDIA Quantum InfiniBand network.
“I expected the A100s to be twice as good as the V100s and P100s I used before, but honestly it was like an order of magnitude difference,” Al-Refaie said.
Orders of magnitude gains
A single A100 GPU boosted performance 200x compared to a CPU.
With 32 processes on each GPU, the team achieved the equivalent of a 6,400x speedup compared to a CPU. Each node on Wilkes3 shipped with its four A100s the equivalent of up to 25,600 CPU cores, he said.
Accelerations are high because their application is surprisingly parallel. It simulates on GPUs how hundreds of thousands of wavelengths of light would pass through the atmosphere of an exoplanet
On the A100s, their models run in minutes, which would take weeks on the processors.
GPUs executed complex physical models so quickly that their bottleneck became a CPU-based system handling the much simpler task of statistically determining where to explore next.
“It was kind of funny, and kinda amazing, that simulating the atmosphere wasn’t the hardest part – it gave us the ability to really see what was in the data,” said- he declared.
A wealth of software
Al-Refaie used CUDA profilers to optimize tasks, PyCUDA to optimize team code, and cuBlas to speed up some math routines.
“With all the NVIDIA software out there, there’s so many things you can exploit, so the team is starting to spit out papers quickly now because we have the right tools,” he said.
They will need all the help they can get, as the job is about to get a whole lot harder.
Get a better telescope
The James Webb Space Telescope will come online in June. Unlike Hubble and all previous instruments, it is specifically designed to observe exoplanets.
The team is already developing ways to work at higher resolutions to support the expected data. For example, instead of using one-dimensional models, they will use two-dimensional or three-dimensional models and take into account more parameters such as changes over time.
“If a planet has a storm, for example, we might not be able to see it with current data, but with next-gen data, we think we will,” Changeat said.
The rising tide of data opens the door to the application of deep learning, which the group’s AI experts are exploring.
It’s an exciting time, said Changeat, who is joining the Space Telescope Science Institute in Baltimore as an ESA fellow to work directly with experts and engineers there.
“It’s really fun to work with experts from many fields. We had space observers, data analysts, machine learning and software experts on this team – that’s what made this paper possible,” Changeat said.
Learn more about the paper here.
Top image courtesy of ESA/Hubble, N. Bartmann