The goal of VeriGPU is to Build an opensource GPU, targeting not only ASIC tape-out but also for machine learning. Hopefully, it will soon work with the PyTorch deep learning framework. Creating an open source GPU for machine learning is long over due with the market leader Nvidia facing little competition in the industry.

 

The eventual goal of VeriGPU is to implement a HIP API, that is compatible with pytorch machine learning framework. VeriGPU is also open to provision of other APIs, such as SYCL or Nvidia’s CUDA. Currently VeriGPU Core ISA is loosely compliant with RISC-V ISA.

 

Users can download the VeriGPU IP from GitHub here