Shearwater has announced a key milestone in the continuous development of its Reveal seismic processing and imaging software, with the platform now running on the Arm architecture using the NVIDIA GH200 Grace Hopper Superchip.
Early benchmarks demonstrate improved performance and energy efficiency across both machine learning and geophysical workloads. A beta release is planned for November 2026, followed by a commercial release in May 2027.
This work builds on the five-years technology collaboration between Shearwater and NVIDIA. Reducing cycle time and lowering emissions are key objectives for Shearwater, and enabling Reveal to run on the power-efficient Arm-based Grace Hopper Superchip is a significant step forward in this work, helping energy companies deliver faster time-to-image while reducing the power and cost of compute-intensive seismic processing.
Shearwater engineers ported the full Reveal platform, a complex package of more than 350 tools spanning several programming languages, to run on Arm, and then benchmarked it on a NVIDIA GH200 Grace Hopper system against an x86 system with NVIDIA Hopper GPUs. The early results are encouraging across both AI and geophysical workloads.
On the deep-learning training benchmark, the tightly integrated NVIDIA NVLink-C2C interconnect between the Grace CPU and Hopper GPU delivered the largest benefit, particularly for data-movement and launch-sensitive operations.
On the Kirchhoff Depth Migration imaging workloads, Grace Hopper delivered consistent double-digit speed-ups together with near-linear scaling across multiple nodes.
The Arm-based Grace CPU also brings the low power consumption for which Arm processors are known, supporting Shearwater’s objective of lowering emissions from compute-intensive HPC workloads.
Beyond exploration and production, Reveal is used to process seismic data for shallow hazard mapping, wind farm planning, geothermal projects, and carbon capture and storage (CCS), all of which stand to benefit from faster and more efficient subsurface imaging.