The atmospheres of distant planets have been described in unprecedented depth in a report published today.
Scientists investigated 25 exoplanets (planets that orbit stars outside our solar system) in an effort to learn more about the universe. To be more specific, scientists looked at exoplanets with temperatures over 3,000 degrees Fahrenheit, common on hot Jupiters, which are the largest and simplest to find.
High-performance computing with NVIDIA GPUs was utilized to better comprehend the atmospheres of all planets, including Earth.
New Lights are Shining on Jupiter
Quentin Changeat, the paper’s lead author and a research fellow at University College London, said that hot Jupiters “provide a fantastic chance to study physics under climatic circumstances practically hard to duplicate on Earth” (UCL).
Large questions are clarified by looking at patterns among many exoplanets.
For Ahmed F. Al-Refaie, co-author of the research and head of numerical methods at the UCL Centre for Space Exochemistry Data, “this work can help develop better models” of how Earth and other planets came to be.
Analyzing Hubble’s Extensive Information
1,000 hours of archival observations, mostly from the Hubble Space Telescope, were used in the study of exoplanets.
When it came to deciding which models to run against data from all 25 exoplanets, Changeat found this to be the most challenging and exciting component of the project.
NVIDIA GPUs allowed him to “explore all kinds of sometimes odd ideas, but it was incredibly fast to acquire the answers,” he noted.
Infinite Attempts to Solve the Problem
Their overall findings necessitated a lot of complex arithmetic. All 25 exoplanets have to be run 250,000 times by each of around 20 models.
NVIDIA A100 Tensor Core GPUs on an NVIDIA Quantum InfiniBand network was employed in the Wilkes3 supercomputer at the University of Cambridge.
According to Al-Refaie, “I was expecting the A100s to be twice as powerful as the V100s and P100s I had previously used, but honestly, it was like an order of magnitude.”
CPUs were outperformed by an A100 GPU by 200 times.
With 32 processes running simultaneously on each GPU, the team was able to achieve a 6,400x speedup over a conventional CPU According to Wilkes3, each node on Wilkes3 delivered the equivalent of up to 25,600 CPU cores with four A100s on each node.
Because their application is so multi-threaded, the speedups are enormous. An exoplanet’s atmosphere is simulated on a GPU using hundreds of thousands of light wavelengths.
Work that would take weeks on CPUs is completed in minutes by their models on A100s.
There were so many intricate physics models to run that their bottleneck was a CPU system that had to handle a far easier task of finding out where to go next.
“Simulating the atmosphere was not the hard part — it allowed us the ability to truly see what was in the data,” he added. “It was humorous and somewhat astonishing.”
Variety of Software
A CUDA profiler, PyCUDA, and cuBlas were used by Al-Refaie in order to optimize the team’s code and speed up some math processes.
Now that the NVIDIA software suite is readily available, the team is producing articles more quickly than before since they have the necessary resources, he explained.
Because the work is only going to get harder, they’ll need all the support they can get.
Acquiring a High-Quality TELescope
In the month of June, the James Webb Space Telescope will go live. Instead of looking for Earth-like planets like Hubble did, Kepler will be looking only for exoplanets.
The team is already working on ways to accommodate the expected data at greater resolutions. A few examples of this include the usage of two or three-dimensional models, as well as the inclusion of more variables, such as time.
It’s possible that current data won’t reveal a storm on a planet, but Changeat is confident that next-generation data will.
Studying AI and HPC
Because of the increasing amount of data, the group’s AI scientists are looking at how deep learning might be applied.
Changeat, an ESA fellow who will join the Space Telescope Science Institute in Baltimore to work with the institute’s scientists and engineers, described the current period as “interesting.”
I enjoy working with professionals on a wide range of topics. As a result of this team’s expertise in space observation, data analysis, machine learning, and software, this work was achievable, Changeat stated.