Performance

Lightning-Fast Performance

Leverage the latest technologies to provide fast retrievals on consumer grade computers and high-performance computing facilities.

Highly optimized python

Exceptional with optimized python design using numpy and numba: TauREx uses vectorized and C-compiled libraries to provide the fastest models, even on laptops.

GPU-accelerated

NVidia GPU-accelerated models allow 100x gains in your calculations: these functionalities are deployed via the TauREx-Cupy plugin.

Parallel High Performance Computing

TauREx retrievals are made for parallel computing on HPC! Using MPI, multiple CPUs can work on the same problem.

Designed for your computing needs

1000+ CPUs across multiple nodes, Low memory nodes, multi-GPU facilities? TauREx is always at its best and adapts to your facilities.

Built for Speed

Our code is engineered from the ground to optimize performances on large sciencific computing facilities. Sucessfully deployed over 20+ National and University clusters, 1000+ CPUs and GPUs can work together in order to find the atmospheric composition of alien worlds!

For more information about TauREx Computing, see press release from our partners NVidia and DiRAC HPC:

NVidia Press Release Link to NVidia PR DiRAC Press Release Link to DiRAC PR

Performance Metrics

  • 1-core CPU compute (1 molecule): 0.6s
  • 1-core CPU compute (15 molecules): 2.3s
  • GPU compute (15 molecules): < 20ms
  • HST Retrieval (24-core CPU): ∼800s
  • Ariel Retrievals (24-core CPU): ∼7,500s

Note: The CPU used for these test is a 24-core Xeon E5-2697 at 2.70GHz, and the GPU is an NVidia V100. This is with the Numba library installed.

See it in action

See how our optimized TauREx retrieval code delivers lightning-fast atmospheric models and retrievals! Coming soon!

Demo

TauREx 3

The most advanced exoplanet retrieval platform.