Scientific computing and high-performance code
NDDev writes computational software and speeds up existing programs: measurement processing, modelling, scientific calculations and algorithms, plus code optimisation in C, C++ and Rust, concurrency and GPU computing. Every speed-up is confirmed with before and after measurements on your data.
Who it is for
Research groups, engineering companies and product teams whose calculations take hours and whose software is bound by CPU, memory or I/O.
- Scientific calculations and modelling
- C, C++ and Rust
- Concurrency and CUDA
- Before and after measurements
Two tasks in this area
Scientific computing: we turn a method or formulae into reliable software that calculates fast and reproducibly. Performance optimisation: we find bottlenecks with a profiler, rewrite the hot spots, parallelise the work and move calculations onto GPUs.
- turning a method into software
- profiling and finding bottlenecks
- rewriting hot spots in C++ or Rust
- concurrency and GPU computing
- before and after benchmarks
- tests and documentation for your team
We use C, C++17, C++20 and C++23, Rust, Python and Go, with CUDA, OpenCL and Julia when needed. The result comes with tests, benchmarks and documentation so your team can keep developing the code.
The order of work
Work is paid monthly and the client decides whether it continues.
Measure
We record the current speed and accuracy on your data.
Plan
We find the bottlenecks and choose what to rewrite first.
Implement
We rewrite, parallelise and check that the results still match.
Report
We show the before and after measurements and hand over the code with tests.
Questions about computing and performance
How much faster will the program get?
The first measurement tells. After profiling we say where the speed reserve is and what gain is realistic, then confirm it with a benchmark.
Will the results change after the speed-up?
Every change is checked against reference data: results must match the originals within the agreed accuracy.
Can we keep the code in Python?
Yes. Often it is enough to rewrite a few hot functions in C++ or Rust and call them from Python.
How much does it cost?
The price depends on the amount of code and the task. We give the scope of the first stage after measuring the current speed.
Computing and performance
Tell us which calculation takes too long
Describe the program, the data and the current run time. We will suggest a first stage and its scope.
We reply within 24 hours