GPU Dev Accelerator Ultimate

Free & open source. Since 2026.8.0 the plugin is completely free — every feature is unlocked for everyone, no license needed — and the source code is available under the Apache License 2.0.

CUDA, Metal, and OpenCL development for IntelliJ-based IDEs

Write, compile, inspect, and analyze GPU code in one IDE. Every feature is backed by a real tool integration (nvcc, compute-sanitizer, Nsight Compute, nvidia-smi) or by deterministic static analysis — no mock data.

Features

Language support (.cu, .cuh, .metal, .cl)

• Syntax highlighting for CUDA C/C++, Metal Shading Language, OpenCL C
• Code completion with API signatures for 100+ builtins (runtime API, atomics, warp/simd primitives, barriers)
• Quick documentation (Ctrl+Q) for kernel qualifiers and builtins
• Line/block commenting (Ctrl+/), brace matching
• Live templates: cukernel, cugridstride, cucheck, culaunch, cushared, cutime, mtlkernel, clkernel

Toolchain integration (GPU Dev console, background tasks)

• Compile CUDA File (Ctrl+Alt+C) — nvcc with configured -arch/flags, clickable errors
• Show PTX — inspect the code nvcc actually generates for the current file
• Memory Check (Ctrl+Alt+M) — compute-sanitizer (or cuda-memcheck) on an executable
• Profile Executable (Ctrl+Shift+P) — Nsight Compute (ncu) or nvprof report
• GPU Information — nvidia-smi device query + Apple GPUs via system_profiler
• Multi-GPU Topology (Ctrl+Alt+G) — real link matrix from nvidia-smi topo -m with live utilization

Static analysis

• Inspections: GPU memory leaks (cudaMalloc/cudaFree pairing), unchecked CUDA API calls (with a Wrap in CUDA_CHECK quick fix), kernel performance heuristics
• Analyze Kernel (Ctrl+Alt+O) — strided access, warp divergence, __syncthreads() in divergent branches, missing bounds checks, missing __restrict__, device-side malloc, and more
• Occupancy Calculator — theoretical SM occupancy for CC 7.0–9.0 with limiting-factor breakdown
• Migrate Kernel — CUDA ⇄ Metal / OpenCL / HIP source-port starting points

Requirements

• Any IntelliJ-based IDE 2023.2 or newer (build 232+)
• NVIDIA CUDA Toolkit for compile/profile/sanitize features (nvcc, compute-sanitizer, ncu on PATH, or set paths in Settings | Tools | GPU Dev Accelerator)
• macOS for Apple GPU detection

Quick start

1. Open a .cu file — highlighting, completion, and inspections are active immediately.
2. Type cukernel + Tab for a kernel skeleton with a bounds check.
3. Ctrl+Alt+C to compile; errors are clickable in the GPU Dev tool window.
4. GPU Dev | Occupancy Calculator to size your blocks before profiling.

Development

./gradlew detekt test   # static analysis + unit tests
./gradlew runIde        # sandbox IDE
./gradlew buildPlugin   # build/distributions/*.zip
./gradlew verifyPlugin  # JetBrains Plugin Verifier (also run in CI)

Support

• Issues: https://github.com/JirakJ/gpu-dev-accelerator-ultimate/issues
• Discussions: https://github.com/JirakJ/gpu-dev-accelerator-ultimate/discussions
• Email: support@jakubjirak.com

Contributing

Issues and pull requests are welcome at https://github.com/JirakJ/gpu-dev-accelerator-ultimate.

License

Copyright 2026 Jakub Jirák. Licensed under the Apache License, Version 2.0.