Back in April 2022 was the announcement by AMD's GPUOpen team of [1]Orochi as a library for HIP/CUDA API run-time switching . Making use of Orochi allows for dynamically targeting either AMD HIP or NVIDIA CUDA at run-time to ease the distribution/usage of apps wanting to target both AMD and NVIDIA GPUs from the same software build. Today Orochi 3.0 was released.

Nearly one year ago was [2]Orochi 2.0 with more functionality for this dynamic run-time switching between the NVIDIA CUDA and AMD HIP libraries. With the just-released Orochi 3.0, it's bringing official support for HIP 6.0+ with ROCm 6.0+ as well as NVIDIA CUDA 12.2 support.

Supporting the newer HIP and CUDA versions is the primary benefit of Orochi 3.0 but a bit surprising it took AMD so long to put out this release for HIP 6.0+ support. Mind you ROCm 6.0 released [3]at the end of 2023 . The latest NVIDIA CUDA Toolkit upstream is also on the 12.6 series. Without timely support for new versions of CUDA and HIP, it's rather difficult to convince developers to make use of this helper run-time library, isn't it?

Orochi 3.0 also brings updated unit tests around WMMA (Wave Matrix Multiply Accumulate) and other features. Plus Orochi 3.0 has added some additional build flags.

Downloads and more details on the Orochi 3.0 release via [4]GPUOpen on GitHub .



[1] https://www.phoronix.com/news/AMD-Orochi

[2] https://www.phoronix.com/news/AMD-Orochi-2.0-HIP-CUDA

[3] https://www.phoronix.com/news/AMD-ROCm-6.0-Released

[4] https://github.com/GPUOpen-LibrariesAndSDKs/Orochi/releases/tag/v3.00