AMD officially launches HIP SDK: expand the ROCm ecosystem and support running Nvidia CUDA applications
Thanks to CTOnews.com netizens South China Daniel Wu, imisaka, OC_Formula for the clue delivery! CTOnews.com, July 30, AMD HIP SDK is now officially launched as part of the ROCm ecosystem, providing CUDA support for professional and consumer GPU.
According to AMD, developers of GPU accelerated applications often need to maintain two separate code bases: one for NVIDIA's proprietary CUDA API and the other for other GPU, but this is not easy, especially for small teams.
Based on the open source ROCm solution, AMD uses HIP SDK to enable consumers to run CUDA applications on a variety of GPU.
CTOnews.com Note: HIP is a free, open source runtime API and kernel language, while AMD HIP SDK essentially converts CUDA applications into simplified C++ code, making it easier to compile and run on AMD or NVIDIA GPU. In addition, AMD extends its HIP SDK not only to its professional GPU, but also to consumer hardware such as Radeon (such as APU).
Although HIP is also part of the open source platform for AMD ROCm computing. However, the AMD ROCm platform is mainly aimed at HPC and AI, especially server solutions, while HIP is designed for desktop applications.
What HIP SDK can do to convert existing CUDA applications to run on AMD GPU
This can be done without contacting AMD for support
Porting GPU accelerated graphics and simulation tools to AMD hardware
According to AMD, HIP SDK can work on all professional, workstation and gaming GPU, and is also supported by APU. The full support GPU list for AMD HIP SDK includes:
Radeon VII
Radeon RX Vega series
Radeon RX 5000 Seri
Radeon RX 6000 Seri
Radeon RX 7000 Seri
Radeon Pro WX 9100
Radeon Pro W6000 Seri
AMD also mentioned that HIP SDK only optimizes the code for developers, but you still need to fine-tune it if you want to realize the full potential of its software. In addition, it is not suitable for machine learning or artificial intelligence tools, such as PyTorch or TensorFlow.
