PoCL 5.0-RC1 is out today as the newest feature release being brewed for this "Portable Computing Language" implementation that allows for OpenCL code to run on CPUs as well as running OpenCL code on other back-ends such as atop NVIDIA CUDA and AMD ROCm and other LLVM back-ends.

PoCL 5.0 is quite interesting in that it will be the first release to ship PoCL-Remote as a new experimental back-end allowing for OpenCL across networked systems . The new networked daemon for PoCL allows for OpenCL distributed computing with more OpenCL code and doesn't require any specialized interfaces. There will be latency/overhead issues and the like of concern for some OpenCL programs, but this back-end allows for OpenCL acceleration between networked systems on the LAN to work for any platforms with TCP/IP and a working OpenCL implementation.

PoCL-Remote is an interesting addition for PoCL 5.0 and will certainly lead to some novel uses.

PoCL 5.0-RC1 also adds new NVIDIA CUDA driver features for providing partial OpenCL 3.0 support. The NVIDIA back-end also adds support for program-scope variables, OpenCL 2.0 atomics, and partial subgroup support. PoCL's CPU driver also enables cl_ext_float_atomics for FP32/FP64, updated cl_khr_command_buffer support, and improved SPIR-V handling. PoCL 5.0-RC1 also extends compatibility to LLVM/Clang 17.0.

Downloads and more details on this PoCL 5.0 release candidate via GitHub .