Devito Codes 4U
Devito makes it easy for scientists and engineers to write blazingly-fast performance-portable finite-difference software for CPUs and GPUs.
We provide products and services based on Devito, a popular open-source mathematical software platform. Interoperable with the Python ecosystem. Rigorous code verification for correctness. Collaborative research community across academia and industry.
Hardware vendors and Cloud
We work closely with hardware vendors to optimize and maintain Devito. Devito Codes has a dedicated cluster for development, optimization, benchmarking, and CI, which comprises nodes from ARM (Fujitsu A64FX), AMD (Ryzen, MI50), Intel (Various Xeon's), and Nvidia (RTX3090, V100, A100 with 40GB and 80 GB memory) in partnership with Dell and Supermicro. We work with Microsoft and DUG to help clients develop cloud applications.
B2B, end users and application developers
We provide bespoke technical support and development services. We work with end-users to integrate Devito into existing workflows, port to new hardware, or go Cloud-native.
We want application developers to be able to implement their code purely in terms of mathematical abstractions and high-level problem specifications. This is achieved in a fraction of the time taken using traditional methods (i.e., hours/days rather than weeks/months). The design objective is to make it easy to write sophisticated solvers. This enables algorithmic research and high-quality code verification.
The novelty of the Devito compiler is that we generate highly optimized code that is competitive with manually optimized code by lowering symbolic maths through a sequence of intermediate representations. All of this is made possible by a multidisciplinary team of domain specialists, mathematicians, computer scientists and HPC/Cloud specialists collaborating on multilayered abstractions.
Checkout Devito -- it's open source on GitHub!
Devito Codes users: checkout your subscribed repo's here.
On your left...
Devito can be used to rapidly re-engineer codes for Cloud and exascale computing. Devito is integrated into Python and its ecosystem, including machine learning - enabling rapid advances in science and engineering.
Easy to use but not a toy
From day zero Devito has been, and continues to be, a HPC project where the measure of success is GFLOPS or giga-points-per-sec. However, what we learned over time is that the greatest appeal of Devito among domain scientists is that it was easy to use and develop new ideas - great out-of-the-box performance was a welcome bonus. Ironically, the fact that Devito is easy to use often results in developers assuming that it does not perform as well as manually tuned code. Nothing could be further from the truth - Devito has come out ahead in all benchmark studies we have been involved in. Expert users can tune Devito to take advantage of specific computer platforms - even modifying the intermediately generated code. This ability to pop-the-hood is often used with vendors when performance tuning or with domain scientists when developing new features that are not yet automatically generated.
Get in touch with anyone on the leadership team if you have any queries.
Co-founder, Director of Business Strategy