PoCL 5.0 has been officially released as the latest version of this "Portable Computing Language" implementation that provides OpenCL support for CPU-based execution as well as various other LLVM-based hardware back-ends like for NVIDIA PTX and oneAPI Level Zero. With PoCL 5.0 is also a new experimental back-end for targeting OpenCL support distributed across networked systems.

PoCL-Remote is the new code in PoCL 5.0 that allows for OpenCL to be transparently used across networked systems . Permitting OpenCL support on the other networked systems, PoCL-Remote allows seamlessly using OpenCL on those remote hosts. This PoCL-Remote handling is done over TCP/IP but there is no encryption/authentication so it's just intended for LAN uses. The performance also may not be all that great depending upon how latency sensitive your OpenCL application is and other possible performance bottlenecks.

PoCL 5.0 also adds new NVIDIA CUDA driver features for providing partial OpenCL 3.0 support, the NVIDIA back-end now supports program-scope variables and OpenCL 2.0 atomics, and various other new OpenCL extensions are implemented. PoCL 5.0 also brings improved SPIR-V handling. There is also better RISC-V processor support with PoCL 5.0 as well as an experimental built-in kernel library based FPGA back-end.

Downloads and more information on today's PoCL 5.0 release via PortableCL.org .