After the last progress meeting in Lugano in January 2024, GPU4GEO got granted a 1 year extension. 6 months later, we submitted a follow up project, ∂GPU4GEO, focussing on differentiable modelling with Julia on GPUs.
GPU4GEO extension
Ludovic Räss
After handing-in the final progress report, the GPU4GEO project got granted a 1 year extension to focus on topics such as:
- GPU HPC CI (running continuous integration and packages' test suite on GPU HPC infrastructure)
- Application readiness on Nvidia Grace-Hopper GH200 nodes CSCS' ALPS infrastructure
- Large scale Geodynamics application readiness
- Finalisation of the STREAM related large scale allocation
Follow-up PASC project - ∂GPU4GEO
On June 10, 2024, we submitted a GPU4GEO follow-up project to the 2024 PASC call. The project core team, composed of many GPU4GEO members, proposes to tackles differentiable modelling at scale using unique features of the Julia language, namely, close to native integration of the Julia GPU stack and outstanding automatic differentiation capabilities (of CPU and GPU code) with Enzyme.jl.
The proposed development will articulate around similar topics as developed in recent workshop such as the NuTS workshop (https://github.com/PTsolvers/Julia-AD-NuTS).