Explore the cutting-edge performance of DevitoPRO on Intel® Xeon® 6 6980P processors in this comprehensive benchmark study. This blog post reveals how next-generation hardware accelerates finite-difference seismic imaging simulations, delivering over 2.6× higher computational throughput and 2.1× faster compute performance compared to 5th Gen Intel® Xeon® Platinum systems. With a focus on both acoustic TTI and elastic propagator models, we examine the benefits of mixed-precision techniques that combine FP16 storage with FP32 arithmetic for optimized memory bandwidth and accuracy. Learn how NUMA-aware hybrid MPI/OpenMP configurations further boost performance, enabling scalable and efficient high-fidelity geophysical simulations. Discover the full benchmark results.
DSLPythonHPCRTM
Devito Codes has developed a new module to improve land seismic imaging by accurately modeling wavefields in complex topographies without the need for unstructured mesh generation. This innovation uses an immersed boundary method, representing free surfaces on a regular grid and enforcing boundary conditions through field extensions. This approach, showcased in their recent paper and collaboration with S-Cube, allows seamless integration into existing seismic data processing pipelines. Schism’s flexibility and efficiency make it ideal for handling topography in seismic imaging. Learn more at IMAGE'24, where Dr. Ed Caunt will present these advancements.
DSLPythonHPCRTM
AWS Graviton4 demonstrates significant performance improvements for seismic imaging using DevitoPRO's 3D acoustic wave propagation kernels. Benchmarks show Graviton4 is up to 3.6 times faster than Graviton2 and up to 81%25 faster than Graviton3, especially benefiting memory-bound HPC applications. Compiling with GCC 14.1 optimizes performance on Graviton4’s Neoverse V2 cores. While there are some performance nuances, overall Graviton4 delivers superior throughput per dollar, making it a cost-effective choice for demanding workloads. These advancements underscore Graviton4's capabilities in enhancing computational efficiency for seismic imaging and other high-performance computing tasks.
DSLPythonHPCRTM
Devito Codes is developing advanced elastic wave solvers using DevitoPRO, offering improved subsurface modeling by leveraging both P-wave and S-wave data. Experiments show the use of mixed-precision methods in DevitoPRO has nearly doubled the performance of wave propagators, showing negligible numerical errors compared to standard 32-bit implementations. Despite challenges in developing robust mixed-precision software, DevitoPRO simplifies this process through its domain-specific language and compiler pipeline, making it an essential tool for seismic imaging. Early benchmarks show a 70 performance improvement, with ongoing efforts to further optimize mixed-precision use, potentially leading to even greater speedups.
DSLPythonHPCRTM
JUDI's latest update, version 3.4.5, now supports DevitoPRO as a bring-your-own-license feature, significantly enhancing its seismic imaging and inversion capabilities. This integration introduces advanced performance and scalability, particularly for large-scale simulations, performance portability across all major CPUs and GPUs and a wide range of domain-specific optimizations. JUDI, an open-source Julia-based framework, already excels in high-performance wave propagation and machine learning integration. The update enables cutting-edge applications such as probabilistic full-waveform inversion, generative AI, carbon storage monitoring, and serverless cloud imaging. This collaboration marks a major step forward in bridging academic research and production-level seismic applications, driving innovation and excellence in the field.
DSLPythonHPCRTM
At Devito Codes, we drive innovation through collaboration, exemplified by our partnership with S-Cube. Combining our expertise in performance-portable wave propagators with S-Cube's advanced seismic imaging algorithms like XWI, we enhance subsurface imaging accuracy and efficiency. Our joint efforts ensure our solutions run seamlessly across major CPUs and GPUs, optimizing performance and cost-efficiency. This collaboration has accelerated the development of next-gen elastic wave solvers, highlighting the power of teamwork in advancing computational geophysics and beyond.
DSLPythonHPCCloud
Devito Codes introduces SYCL code generation in DevitoPRO optimized for Intel's GPU Max Series 1100 and 1550, enhancing performance in high-compute tasks like seismic imaging. This update, developed in collaboration with Intel, enables seamless use of existing Devito application code across various architectures without modifications. This enhancement solidifies DevitoPRO's commitment to performance portability and high productivity across all major HPC processor architectures.
DSLPythonHPCCloud
A framework for cross-platform benchmarking of seismic imaging workloads is described. The vision for this platform is that it will be used to support collaboration between Devito Codes, hardware vendors and Cloud providers to continuously optimize the performance of seismic imaging workloads.
benchmarkingHPCCloudAMD
Core to the Devito mission is performance portability. Of course, if it is not tested then it is assumed to be broken. We maintain a dedicated cluster for all our DevOps needs.
benchmarkingHPCCloudAMD
Devito is a popular open-source DSL for coding performance portable code in Python using symbolic computation for evolving. While previously Devito only supported AMD GPUs using OpenMP offloading, thanks to a recent collaboration with AMD, DevitoPRO now also generates optimized HIP directly.
AMDGPUINSTINCT™