CUDA Occupancy Calculator: A web-based replacement for Nvidia's CUDA occupancy spreadsheet

Optimizing CUDA kernels is often a tedious exercise in balancing register pressure against block dimensions. For too long, the standard workflow for calculating theoretical occupancy—the ratio of active warps to the maximum supported by the SM—has relied on fragmented documentation and fragile internal spreadsheets. This calculator centralizes those variables, providing a streamlined interface to see how block size and memory usage constrain hardware utilization. From a product perspective, the value is purely in the reduction of friction. By allowing developers to toggle register counts and shared memory allocations in real-time, it exposes the 'cliff' where a single additional register can halve the occupancy. However, it is important to note that this provides theoretical maximums; it does not account for actual runtime scheduling, instruction latency, or memory throughput bottlenecks which often matter more than raw occupancy. The tool's primary strength is its accessibility. Being browser-based removes the setup overhead usually associated with profiling tools. The weakness lies in its scope; it is a calculator, not a profiler. It tells you what *could* happen based on static parameters, not what *is* happening during execution. It is a pre-flight check, not a diagnostic tool. This is a pragmatic utility for CUDA programmers who are in the iterative phase of kernel design. If you are currently manually calculating SM limits to decide between a block size of 256 or 512, this tool replaces that manual toil with an instant result. It is a small but effective addition to a performance engineer's toolkit.