DevitoPRO · Enterprise Devito

Performance-portable wave propagation, generated from Python.

Write the physics once; DevitoPRO regenerates the optimized kernel for every architecture and every processor generation.

Architectures
AMD · Arm · Intel · NextSilicon · NVIDIA
Programming models
CUDA · HIP · MPI · OpenACC · OpenMP · SYCL
Deploy
AWS · Azure · GCP · On-prem
Workloads
RTM · FWI · custom propagators and workflows

Built for the hardware and clouds you run on

AMDArmAWSAzureGoogle CloudIntelNextSiliconNVIDIA

Capabilities

DevitoPRO for production

DevitoPRO extends open-source Devito for production seismic imaging and inversion. Both express finite-difference kernels symbolically; DevitoPRO adds the tuning, portability and enterprise support needed for production HPC workflows.

Symbolic DSL · custom physics

  • Write PDE solvers symbolically in Python — your physics, your discretization
  • Customize stencils, boundary conditions, sources, adjoints
  • Inspect and tune the generated kernel; integrate via the Decoupler API in any language

Performance & autotuning

  • High-level optimizations such as expanding-box and mixed precision
  • Extremely optimized code for both GPUs and CPUs
  • Advanced autotuning via DevitoTuner for any architecture
  • Altogether, speed-of-light performance

Multi-GPU & MPI

  • Optimized GPU-aware MPI; single-node and multi-node multi-GPU
  • NUMA-aware MPI + OpenMP on every modern CPU
  • Decoupler: call DevitoPRO from any language with full MPI support

Large-model memory

  • Compression-based back-propagation
  • Intelligent data streaming disk → host → GPU
  • Lossy compression for floating-point data

Backends & cloud

  • CUDA, HIP and SYCL code generation from one symbolic spec
  • CPUs: AMD, Arm, Intel · GPUs: AMD, Intel, NVIDIA
  • AWS, Azure, GCP — tuned and benchmarked

Physics coverage

  • Acoustic VTI/TTI · viscoacoustic VTI/TTI
  • Elastic VTI/TTI · viscoelastic
  • Immersed boundaries for accurate land topography

Support, SOWs & training

  • Private/NDA Slack support channels
  • Bespoke projects, training and consultancy
  • Benchmark reports for reproducibility

Use cases

Built for production exploration geophysics.

Reverse Time Migration (RTM)

Forward/adjoint propagators with checkpointing, compression-based back-propagation and data streaming.

Full-Waveform Inversion (FWI)

Express adjoint-method optimization symbolically; scale gradients across multi-node, multi-GPU with GPU-aware MPI.

Elastic, VTI & TTI anisotropy

Acoustic and elastic VTI/TTI, viscoacoustic and viscoelastic propagators from the Devito cookbook.

Embed in your stack

Call DevitoPRO from applications in other languages through the Decoupler API, with full MPI support, and integrate generated propagators into existing inversion engines.

Very large models

Run models beyond the limits of in-house codes using compression, data streaming, expanding-box and mixed precision to keep memory and transfer costs in check.

Cloud-native seismic imaging

Tuned, benchmarked deployments on AWS, Azure and GCP — portable across available instances.

Immersed-boundary topography

Accurate land topography on finite-difference grids via immersed boundary support.

Time-lapse (4D) monitoring

Production workflows for repeat surveys — scale imaging and inversion consistently across vintages.

Least-squares imaging (LS-RTM) & variants

Iterative imaging and imaging-condition variants built from the same symbolic building blocks.

Performance

Speed-of-light performance, every architecture.

From isotropic acoustic to anisotropic visco-elastic, DevitoPRO kernels are autotuned per target and benchmarked against comparable hand-coded kernels where available. As new processors arrive, DevitoPRO regenerates and re-tunes for them, so the comparison stays current. For specific benchmark numbers on your workload, get in touch.

GPU portability

CUDA, HIP and SYCL backends generated from one symbolic spec — production-grade kernels for NVIDIA, AMD and Intel GPUs, autotuned per target for speed-of-light performance.

CPU portability

OpenMP + NUMA-aware MPI on every modern x86 and Arm CPU.

Cloud price-performance

Tuned and benchmarked across AWS, Azure and GCP. Availability matters as much as price-performance — DevitoPRO helps teams move workloads across whichever instance families are available.

Large-model scale

Compression-based back-propagation, intelligent data streaming, expanding-box and mixed precision keep memory use and disk-to-host-to-GPU transfers in check for large production models.

Case studies

Results — in the field and in the press.

Thrust Belt Imaging

Land FWI in structurally complex terrain

Thrust Belt Imaging teamed up with Devito Codes to make Full-Waveform Inversion work in rugged land settings with real topography — combining structural velocity models with their Depth Weathering Solution into a workflow that delivers stable, meaningful updates where traditional approaches struggle. First real-world results presented at the Energy HPC & AI Conference at Rice University.

Read the story →

Intel Customer Spotlight

Devito Codes automates HPC code generation

How Devito Codes helps customers resolve the HPC trilemma — performance, portability and productivity — generating optimized finite-difference code from a Python DSL.

Read the story →

AWS HPC Blog

Performance gains with AWS Graviton4

A DevitoPRO case study benchmarking 3D acoustic wave-propagation kernels — including FWI and RTM propagators — on AWS Graviton4. Its Neoverse V2 cores and higher-bandwidth memory subsystem deliver substantial gains for memory-bound seismic-imaging workloads over previous Graviton generations.

Read the story →

AMD

Devito revolutionizes HPC for oil & gas with AMD

AMD profiles how DevitoPRO brings performance portability to seismic imaging — letting geophysics teams define wave-propagation kernels symbolically and generate optimized code across CPUs, GPUs and clusters, lowering the HPC barrier for domain specialists in oil and gas exploration.

Read the story →

Microsoft Research

Devito Workshop at the 2022 Rice Energy HPC Conference

A workshop and hackathon introducing Devito — the DSL and code-generation framework for highly optimized finite-difference kernels such as FWI and RTM — run on Azure CPU and GPU instances, with lightning talks illustrating Devito from end-user, cloud-provider, hardware and academic perspectives.

Read the story →

Team

Bringing together geophysics, HPC, computer science and mathematics

Gerard Gorman

Gerard Gorman

CEO & co-founder

Fabio Luporini

Fabio Luporini

CTO & co-founder

Paul Holzhauer

Paul Holzhauer

Director of Business Development & Strategy, co-founder

Mathias Louboutin

Mathias Louboutin

Senior Solution Architect

Edward Caunt

Edward Caunt

Research Scientist

Jack Betteridge

Jack Betteridge

Research Scientist

Discuss your wave-propagation workload.

Whether you run a geophysics team writing your own RTM/FWI kernels, build seismic software, or design the hardware underneath — tell us about your target architecture and imaging pipeline, and we will benchmark it and show you what DevitoPRO can do.