Scientific computing services
NDDev writes software for scientific and engineering computation: numerical models, measurement processing, computational libraries and GPU acceleration. Our RnD practice works in C++, Rust, Python and Julia, and writes GPU code in CUDA and OpenCL. We do this for laboratories, manufacturers and research teams, and NDDev has worked since 2019.
Who it is for
Laboratories, engineering centres, university groups and manufacturers whose calculations live in scripts or spreadsheets and run into limits of speed, accuracy or ease of rerunning.
- Code in C++, Rust, Python and Julia
- GPU computation with CUDA and OpenCL
- Checked against reference answers
- Libraries and plug-ins for your software
What we do with your computation
A scientific computation often starts as a script that one person understands. We move it into software with a clear structure, tests and a repository, check the output against the original version and speed up the slow parts: we rewrite the core in C++ or Rust, parallelise the work and move it to the GPU.
- a review of the model and the original code
- a port to C++, Rust or Julia with tests
- parallel computation on CPU and GPU
- cross-platform libraries and plug-ins
- measurement processing and visualisation
- documentation on running the code and checking results
If the computation is needed inside another program, we write a cross-platform library or plug-in with a clear interface, Python included. To explore the results, we build a desktop application in Qt with scientific visualisation.
How we port and speed up a computation
Every change is checked on the same input data.
Reference
We take your current version and a set of inputs with known answers, and check every change against them.
Profiling
We find where the computation spends time and memory, and show you that map.
Port and speed-up
We rewrite the bottlenecks and check the results against the reference after every step.
Handover
We deliver the code, tests and instructions so your team can run and develop the computation on its own.
Questions about scientific computing
Do you understand our field?
We work through the mathematics of the model with you and take on the engineering: code structure, speed, parallelism and result checks. The problem statement and the physical meaning stay with your specialists.
Will the results change after the speed-up?
Every step is checked against the reference answers, and the allowed deviation is agreed in advance. If a deviation appears, you will see it in the report together with its cause.
Can the computation stay in Python?
Yes. Often it is enough to rewrite only the computational core in C++ or Rust and plug it into Python as a module. The rest of the code stays familiar to your team.
Do we need GPUs?
A GPU pays off where the computation splits well into parallel parts. We check that with a measurement on your data, and only then write CUDA or OpenCL code.
How much does the work cost?
The price depends on the size of the code, the complexity of the model and whether a GPU port is needed. The starting price is shown in the price block on this page; we name the total after reviewing the computation.
from ₽280,000
- Free warranty
- 40 days
Prices are starting prices for information only and are not a public offer. The final price is fixed in the specification.
Computing and performance
Related
Show us the computation that takes too long
Send a description of the model or a piece of the code. We will tell you where to look for a speed-up and how to verify it.
We reply within 24 hours