Gridworkz Documentation
Gridworkz designs open hardware for AI research. The work lives in three repositories under github.com/kubeworkz:
| Repository | What it is | Status |
|---|---|---|
| grx930 | RISC-V SoC: 4 × RV64IMAC cores, 2 × 8×8 systolic NPUs, coherent L2, AXI4 fabric | Verified in simulation; timing closed on Artix-7 FPGA; GF180MCU test chip in progress |
| grxgpu | G100: open RISC-V GPU built on Vortex, with tensor cores, DXA, ray tracing and graphics. Also hosts the photonic tensor accelerator (PTA) design proposals | Simulation (SimX, RTL); discovery configuration under review |
| grxcp | GRXCP: CUDA-style runtime, grxcc compiler, grxBLAS and tools for the G100 and GRX930 |
In active development |
Where to start
- Run something now: Quickstart. Clone grx930 and run the NPU test suite in Icarus Verilog.
- Program the GPU: GRXCP
- Read the hardware: RISC-V CPU and memory map · NPU registers · G100 GPU · Photonic tensor accelerator · Board pinout
All three repositories are licensed under the Apache License 2.0. Third-party tools, scripts and PDK files included in them keep their own terms.
These pages summarize the repositories. When they disagree, the repository is right.