GopherCon UK 2026
On 12th August 2026, I returned to GopherCon UK to give From Go to the GPU: Integrating with CUDA. It is always a friendly and welcoming conference, and all the talks were wonderful. I had a lovely time catching up with the Go community and sharing some GPU programming on stage.
Talk Abstract
Go is an appealing language for machine learning projects: it is fast, efficient, and has a mature concurrency model. Pairing it with CUDA also makes it possible to keep the conventional application code in Go while moving specialised, performance-critical work onto an NVIDIA GPU.
The talk followed a concrete example from end to end. Go handled the Wordle game, orchestration, synthetic training
data, and evaluation. A custom neural network, designed from scratch in CUDA, handled the expensive work of running and
training the model. The two halves met through cgo.
That example gave us a practical way to examine the architecture of a mixed Go/CUDA codebase: how to divide responsibilities between host and device, how to structure and profile the project, and where libraries from Go’s machine learning ecosystem might fit as it evolves. The aim was not to wrap a Python framework or call a cloud service, but to look closely at the nuts and bolts of building and training a model with Go and CUDA.
The original abstract and session details are in the official GopherCon UK programme.
Slides
The project code and conference slides are available on GitHub. The slides were made with Calliope Canvas, Kirti B’s code-first slideshow tool.
I will add a link to the conference recording here when it is available.