HIP is the GPU programming interface in AMD's ROCm stack. It is designed to look familiar to anyone who has written CUDA, which makes porting CUDA applications to AMD GPUs quicker.
What AMD says about it
AMD's documentation states that "GPU programmers with NVIDIA CUDA experience will find the HIP API straightforward" and that "you can quickly port NVIDIA CUDA applications to run on AMD GPUs." HIPify tools can convert CUDA API calls into the corresponding HIP calls.
AMD also states the limit: "HIP is not intended to be a drop-in replacement for NVIDIA CUDA", so some code still needs rewriting and performance tuning.
Where HIP sits
HIP is one component inside the stack described in what ROCm is, listed next to HIPIFY and LLVM in the Core SDK. The conversion tooling is covered in whether AMD GPUs can run CUDA code.
For how the whole picture compares with NVIDIA, see NVIDIA's main competitors in AI.
