Wafer is a GPU development stack designed to streamline the process of writing and optimizing GPU kernels by integrating essential tools directly into your integrated development environment. It eliminates the need to switch between fragmented tools by bringing profiling, documentation search, compiler exploration, and AI-assisted optimization into a unified workspace.

The platform offers GPU profiling capabilities through NVIDIA Compute Utility (NCU) integration, allowing developers to run profiles directly from their editor and gain performance insights without context switching. It includes comprehensive documentation search functionality that provides fast access to CUDA programming guides, API references, and optimization best practices. The compiler explorer feature enables users to compile CUDA and CuteDSL code into PTX and SASS assembly, mapped to source code and available as context for AI agents.

Wafer provides GPU workspaces that allow developers to work primarily in persistent CPU environments while spinning up GPUs on-demand when running code, reducing infrastructure costs by approximately 95%. The system includes an AI agent that can read profiling data and suggest optimizations, with tool calling capabilities that allow the agent to perform actions like running NCU profiles, searching documentation, and executing code automatically.

The platform enables developers to review agent-suggested changes through code diff functionality before applying optimizations, and includes hyperparameter tuning features that automatically sweep common kernel parameters like tile sizes, thread counts, and unroll factors. These capabilities help kernel engineers identify performance bottlenecks, implement optimizations more efficiently, and reduce the time spent switching between different development tools.

Wafer targets GPU kernel engineers and developers working with CUDA code who want to accelerate their development workflow and optimize GPU performance. The platform integrates with popular IDEs including Cursor and VS Code through extensions, and supports development workflows involving NVIDIA GPUs with hardware counters.