tiny-gpu is a minimal Verilog GPU implementation designed for learning how GPUs work from the ground up. The project emphasizes the core principles of GPU architecture, parallelization, and memory management. Unlike CPU resources, GPU details are often proprietary due to market competitiveness, making resources for understanding GPU hardware limited. tiny-gpu simplifies GPU complexity to focus on fundamental elements critical to all modern hardware accelerators. With fully documented Verilog files, matrix addition and multiplication kernels, and simulation support, tiny-gpu serves as an accessible tool for exploring GPU functionality at a hardware level. Further developments may include advanced GPU optimizations and features.
https://github.com/adam-maj/tiny-gpu