AMD Releases Orochi 2.0 With More CUDA/HIP Functions Implemented For Better Portability

AMD GPUOpen's Orochi project as a reminder is the effort for allowing dynamic runtime switching between the Radeon HIP and NVIDIA CUDA APIs to allow better cross-GPU portability. Today marks the availability of Orochi 2.0 for enhancing this API to target NVIDIA CUDA and AMD HIP hardware.

AMD's Orochi is not to be confused the effort AMD previously funded with ZLUDA for a drop-in CUDA implementation built atop ROCm. Rather, Orochi is its own API that in turn can target NVIDIA CUDA and AMD HIP use at run-time from a single binary. Applications need to be adapted to make use of Orochi.

With the just-released Orochi 2.0, "a lot more" of the CUDA and HIP functions have been implemented compared to Orochi 1.0. The release announcement says the function coverage with Orochi 2.0 should be "almost exhaustive."

Orochi 2.0 also brings new demos for Direct3D 12 interoperability as well as textures. Orochi 2.0 also improves the bindings and naming between HIP and CUDA for better maintainability.

Porting existing Orochi 1.x applications to using the Orochi 2.0 API is described as straight-forward with minimal changes.

Downloads and more details on AMD's Orochi 2.0 release via GitHub.