HARDWARE – GPP

General Purpose Processor (GPP)

A high-performance, configurable out-of-order RISC-V core and chiplet, optimised for HPC workloads in European supercomputers.

Lead partner

At a glance

Out-of-order

RISC-V core

High-performance, configurable and customisable IP core.

Hardware/software co-design

Developed together with the software and applications that will run on it.

Chiplet and System-in-Package

GPP chiplets are integrated with AIPU chiplets in one package.

Full software stack

Middleware, OS porting, compiler and optimised compute libraries.

Software Development Vehicles

Software work starts before the chip exists.

What we are building

The General Purpose Processor (GPP) is one of the three RISC-V chiplets developed in DARE. Led by Codasip, it is a high-performance, configurable and customisable out-of-order RISC-V core designed for classical HPC workloads in European supercomputers.

The GPP is built as a chiplet. Together with the AI Processing Unit (AIPU), it forms a complete System-in-Package (SiP): GPP chiplets run classical HPC workloads, while AIPU chiplets accelerate AI inference. The SiP is mounted on a PCIe card for integration into HPC systems.

A full software stack is developed in parallel with the hardware, including middleware, operating-system porting, compilers and optimised compute libraries. Software Development Vehicles (SDVs) let developers port and test their code from day one.

From CPU core to chiplet, System-in-Package, PCIe card and HPC rack

Image shown is illustrative.

Why it matters

European-designed processor

A configurable RISC-V processor designed and developed in Europe, reducing reliance on non-European technology for critical computing infrastructure.

Built for heterogeneous systems

Chiplet design lets the GPP be combined with AI accelerators in one package, with greater efficiency, scalability and cost advantages than monolithic chips.

Tested on real applications

European applications including IFS and NEMO (weather and climate), Alya and Bettair (air quality), large language models and SuSy are being optimised for the GPP.