We will gladly assist in porting CPU-based code functionality to the GPU.
We analyze the provided codebase in-deep before we define a step-by-step strategy for the desired GPU acceleration.
Example: An industrial customer had used Machine Learning algorithms on the GPU, but been using the CPU for the postprocessing of the classifier output, which had slowed down overall performance of the ML sensing application because of unnecessary PCIe bus data transfers.
Our prototype could eliminate that bottleneck.
Contact us below if you need to remove similar bottlenecks in your software pipeline.
Existing GPU codes might be yet unfinished or need further extension.
After providing us with a briefing on the current code state and the intended functionality, we would set up our systems in order to compile and run your code in its current state and then start extending it towards your goals, with exact tracking of the hours spent (billed weekly or monthly).
git-based collaboration is no problem, we can also interact via Slack, Atlassian tools and regular video meetings.
The development process can be hardened by having us specify unit tests and formal specifications that we develop in collaboration with our clients.
Contact us if contracted software development would help you best.
Our expertise with NVIDIA hardware, spanning over more than two decades, makes us proficient at spotting performance bottlenecks in GPU codes.
We excel especially in analyzing real-time vision or HPC workloads.
NSight Systems, NSight Compute are our main tools for this analysis, followed by practical advice on improving code performance.
Contact us if you worry about low hardware utilization for your current GPU codes.