NVIDIA CompileIQ
NVIDIA CompileIQ™ is an AI-powered compiler auto-tuning framework that uses evolutionary and genetic algorithms to optimize NVIDIA GPU compilers for individual workloads. Instead of accepting one generic compiler configuration for all workloads, CompileIQ flips the script: It generates optimal configurations for NVIDIA compiler options based on the specific workload in question.
How CompileIQ Works
Tune your compiler to the kernel, not the other way around.
NVIDIA GPU compilers use default heuristics designed to produce strong results across a wide range of applications. That broad default is valuable, but the best configuration for one workload is not always the best configuration for another. CompileIQ turns Compiler Search Spaces into candidate configurations, evaluates these candidates against a developer-defined objective function, and uses evolutionary search to find the optimal candidate for specific kernels that matter to you.
Also, there are some known configurations that work well for specific workloads. We call these premade configurations “Booster Packs,” and they’re available to use in your application without any tuning.
CompileIQ works best for:
Teams that already have highly optimized source code for their kernels
Applications that follow the 80/20 rule: 80% of the compute budget is spent on 20% of the code
New kernel developers developing variations of kernels for which a Booster Pack is provided
GitHub Repository
Browse source code, search space definitions, and discover usage examples. Includes compiler tuning scripts and reference implementations for NVIDIA® CUDA® C++, NVIDIA Triton, and Helion kernels.
CUDA Programming Guide
Learn how to use the --apply-controls flag introduced in CUDA Toolkit 13.3+ to apply Advanced Control Files directly to NVCC and PTXAS for kernel compilation.
Search Space Downloads
Download predefined compiler search spaces for PTXAS and NVCC. These define the tuning parameters CompileIQ will explore when searching for an optimal configuration for your kernel.
Booster Pack Downloads
Pre-built Advanced Control Files are optimized for common workloads. Use these without running a search. Drop them into your build pipeline and benchmark immediately.
Get Started With CompileIQ
CompileIQ supports CUDA C++, OpenAI Triton, and PyTorch Helion—pick your environment to get started.
CUDA C++
Apply an ACF to your kernel at compile time using the --apply-controls flag with NVCC 13.3+. Ideal for teams benchmarking with nvcc-compiled kernels.
NVCC Documentation
OpenAI Triton
Pass an ACF via the PTX_OPTIONS environment variable for global scope, or set ptx_options directly in your kernel call for per-kernel control.
Triton Integration Guide
PyTorch Helion
Specify an ACF directly via helion.Config, or use autotune_search_acf to let the autotuner select the best configuration from a list of candidates.
Helion DocumentationCompileIQ Learning Library
More Resources
Ethical AI
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their supporting model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
For more detailed information on ethical considerations for this model, please see the Model Card++ Explainability, Bias, Safety & Security, and Privacy Subcards. Please report security vulnerabilities or NVIDIA AI Concerns here.
Get started with CompileIQ today.