The Portable Instruction Set Architecture, or PISA, is designed to sit between high-level compiler output and a GPU's native machine code. This abstraction layer means a single PISA workload can be finalised for different compatible Intel GPU architectures, allowing newer hardware features to be used without rewriting the original code. Intel notes that compatibility depends on the target architecture and the specific instructions used by the program.

PISA is not Intel's first virtual ISA, as the company already uses vISA for its compiler stack. The new standard introduces a portable programming model, with Intel listing close-to-the-metal performance and support for handwritten libraries as goals. The initial documentation covers an OpenCL 3.0-based execution and memory model, as well as definitions for data types, addresses, and registers.