Maximizing seismic insights: how Petrobras uses S-Cube’s cloud native XWI solution built on AWS Graviton

Introduction

Seismic processing is a high-performance computing workflow to enhance, image, and understand subsurface structure. Modern high density 3D seismic data acquisition methods need 10-100 times more compute power for processing and imaging as compared to previous generation seismic acquisition methods. In return, geologists get improved geological resolution details such as thin beds, subtle faults, and stratigraphic traps. Higher data density leads to clearer imaging workflows in complex subsurface environments such as salt domes, fault zones, and fractured formations. Full Waveform Inversion (FWI) is a sophisticated seismic processing technique to employ high density seismic data to derive high resolution subsurface models. In this post we present how S-Cube and Petrobras are harnessing AWS Graviton4 processors on Amazon Elastic Compute Cloud (Amazon EC2) Spot instances to run cloud-native FWI workflows, significantly enhancing the efficiency of large-scale earth imaging.

S-Cube Technologies is a forward-thinking, research-driven company, redefining seismic imaging by overcoming the conventional limitations. Founded on the belief that conventional FWI is progressing too slowly, we have identified a clear gap in the market applying fully FWI-based seismic imaging to both marine and land datasets for all subsurface energy targets.

Petrobras is a publicly traded corporation operating in an integrated and specialized manner in the oil, natural gas, and energy industry.

S-Cube’s advanced algorithms and research-driven approach deliver improved seismic accuracy, using the computational power and flexibility of the cloud. This collaboration not only reduces costs and compute time but also pushes the boundaries of geophysical analysis with cutting-edge methods applied in complex settings.

FWI

FWI is a physics-based iterative, gradient-based optimization approach used to construct highly accurate 2D and 3D models of the subsurface, as shown in the following figure. This accuracy is crucial for reducing exploration risk and optimizing production in the energy industry. Therefore, it has quickly become the algorithm of choice. However, due to the amount of seismic data and timelines of projects, traditional on-premises hardware often struggles under the weight of growing data volumes, driving the need for flexible, scalable solutions. Cloud-based compute power opens a new world of possibilities for faster, more cost-effective FWI.

S-Cube XWI is a collection of advanced FWI methods that allow machines to extract the most detailed and accurate velocity models possible from raw signals in exploration seismic datasets.

Figure 1. FWI workflow

Amazon Web Services (AWS) offers on-demand, HPC infrastructure that can be tailored to meet the needs of computationally intensive workflows such as FWI. From the sheer breadth of AWS services to the economic model of pay-as-you-go, the cloud provides a compelling platform for companies looking to accelerate seismic processing projects. AWS Graviton processors further amplify this advantage by delivering optimized performance and cost benefits tailored for HPC workloads.

FWI using S-Cube Cloud

The following figure shows how S-Cube Cloud (SSC) is deployed on AWS. SCC is a cloud infrastructure built on AWS and specialized for HPC workloads. It can be automatically deployed through infrastructure as code (IaC) and supports both CPU and GPU instances. A custom developed scheduler runs on an on-demand EC2 instance and automatically provisions a fleet of Spot Instances for the main compute tasks, with each attaching ephemeral Amazon Elastic Block Store (Amazon EBS) volumes as needed.

EC2 Spot Instances can be interrupted at any time, thus fault tolerance is essential. The workload must recover and continue with minimal downtime when capacity fluctuates. As new Spot Instances become available, the system must seamlessly incorporate these resources. For these reasons, traditional Message Passing Interface (MPI) can be challenging in this environment. Therefore, we developed a TCP-based communication interface that delivers enhanced fault tolerance and flexibility for cloud-native HPC workflows.

All input and output data are stored in Amazon S3, while logs and status updates are continuously streamed to Amazon CloudWatch Logs. Furthermore, Amazon Elastic File System (Amazon EFS) houses job-specific logs in near-real-time, making it easy to check progress and troubleshoot issues during execution. Traditionally, FWI needs highly parallel file systems, such as Lustre. Using Amazon S3 allows S-Cube to save on storage and management costs.

AWS Lambda functions manage essential orchestration tasks such as job-completion notifications and automatic instance termination, making sure that resources are deprovisioned when no longer needed. AWS Identity and Access Management (IAM) roles carefully control the permissions for Lambda functions, EC2 instances, and the EC2 Spot Fleet, maintaining a secure and efficient workflow.

Figure 2. Architecture of S-Cube Cloud on AWS

Geophysics results

Working closely with Petrobras, we deployed the FWI infrastructure in both the sa-east-1 and us-east-1 AWS Regions, running production workloads on large-scale seismic data from the Santos Basin, as shown in the following figure.

Figure 3. FWI Imaging results at 20 Hz and 45 Hz

Using 50- and 25 meter grids, the workflows processed up to 3.3 TB of seismic data covering a 700 km² subsurface area extending to 10 km in depth. The modelling used both tilted transversely isotropic (TTI) acoustic and (TTI) elastic wave equations, the latter of which can be 5- to 10 times more computationally demanding. Despite the higher complexity, the fully data-driven results closely matched and even improved upon those from conventional methods, demonstrating the power of cloud-based FWI for high-fidelity subsurface imaging.

Each job iteration drew on 163 to 325 EC2 Spot Instances—a mixture of c6g, c7g, and c8g Graviton families—to process 325 seismic shot gathers. Capitalizing on the on-demand elasticity of AWS meant that each iteration was completed in about 72 minutes, maintaining rapid turnaround times even as the computational load increased significantly. This scalable approach proved especially valuable for elastic FWI, where compute needs often spike due to the multiple wave modes involved.

Running FWI at this scale on traditional on-premises systems would typically involve substantial capital expenditure, hardware maintenance costs, and long lead times. In contrast, the pay-as-you-go model of AWS—augmented by Spot Instances—substantially reduced compute costs while still delivering robust HPC performance. Automatically bidding for unused AWS capacity allowed Spot Instances to enable S-Cube and Petrobras to run large-scale FWI jobs at lower prices, yet they still maintained the fault tolerance and data durability essential for mission-critical workloads.

“Using the scalable environment of AWS through S-Cube’s XWI software and SCC infrastructure allowed us to process our seismic imaging workflows efficiently, paying only for the resources we use. This flexibility accelerates data-driven decisions and helps us optimize costs.” – Gustavo Catão Alves, Geophysical Advisor, Petrobras

Conclusion

In this post we demonstrated how users can run large scale cloud-based FWI workflows using hundreds of Amazon EC2 Spot instances and serving terabytes of seismic data directly from Amazon S3, thus eliminating traditional parallel storage such as Lustre. S-Cube’s solution used AWS Graviton resources efficiently and produced high fidelity seismic images. The AWS Graviton CPU demonstrated performance in a real user use case, highlighting the latest technological advancements in HPC and creating a feasible alternative to traditional x86 CPU and GPU seismic processing methods.

Contact S-Cube today to find out how FWI on the cloud can improve your subsurface understanding and reduce uncertainty risks.