Electrical engineering YouTuber bitluni has designed and assembled a custom GPU cluster powered by 8,192 CH570 RISC-V microcontrollers. Despite numerous technical challenges, the project successfully drives a 320x200 QVGA RGB LED display, with plans for an expanded build featuring 32,000 microcontrollers.
- Homemade GPU cluster built with 8,192 $0.13 RISC-V MCUs
- Custom PCB design faced manufacturing and design hurdles
- Next iteration will feature 32,000 MCUs for greater scale
What happened
Bitluni, an electrical engineering and software development content creator, embarked on a project to build a GPU cluster using 8,192 RISC-V CH570 microcontrollers running at 100 MHz each. These MCUs were arranged on custom six-layer PCBs in a grid layout connected by another set of larger control cores. The result was a system capable of lighting a 320x200 resolution RGB LED display, effectively acting as a homemade GPU.
Throughout about six months of design, construction, and troubleshooting, the project encountered issues including manufacturing difficulties, intermittent MCU failures, signal interference, and wiring errors between communication lines. Despite these setbacks, bitluni completed a functional cluster and documented the process in a detailed video. The initial design was limited by PCB manufacturer constraints, prompting the splitting of boards and plans to build even larger clusters.
Why it matters
This DIY GPU project demonstrates an alternative approach to graphics processing hardware by leveraging large numbers of inexpensive, open-architecture RISC-V microcontrollers. It highlights the potential for hobbyists and engineers to experiment with scalable compute arrays outside traditional GPU ecosystems, especially amid ongoing chip shortages and rising hardware costs.
Moreover, the project underscores challenges in modern PCB design and manufacturing when dealing with highly complex, large-scale embedded systems. By sharing the design and experience, bitluni contributes valuable insight into the feasibility and engineering considerations of massively parallel microcontroller clusters for graphics or other parallel processing applications.
What to watch next
Bitluni plans to release a follow-up video showcasing the operational capabilities of his homemade GPU cluster, alongside open-source design files for the community. This will enable other enthusiasts to replicate or build upon the project and potentially explore new applications for large-scale MCU arrays.
Additionally, the next iteration of the project aims to significantly increase the number of microcontrollers to 32,000, pushing the boundaries of DIY GPU clusters. Monitoring how these larger-scale builds address previous challenges and achieve performance improvements will be key to assessing practical uses and inspiring further innovation in accessible, modular computing hardware.