Building a VGA at Home Without a Graphics Processing Chip: The Crazy RISC-V Based Project

Creating a GPU without using a traditional graphics processor might seem like an out-of-the-box idea, but that’s exactly what Matthias Balwierz, known in the community as Bitluni, set out to do. The experimental project replaces the classic graphics chip with thousands of inexpensive RISC-V microcontrollers, each responsible for processing a single pixel and displaying it through an RGB LED.

The idea is to combine graphic hardware and display into a single structure. Each chip functions simultaneously as a computing unit and as a light point on the screen. To keep costs and complexity down, Balwierz opted for simple RGB LEDs instead of more sophisticated addressable solutions, which would significantly increase the required budget.

The ultimate goal is to create a display with a resolution of 320 x 200 pixels, a setup that requires 64,000 microcontrollers. A Full HD panel, on the other hand, would need over 2 million chips, which is indeed an unfeasible milestone for this technological demonstration. The current prototype uses 8,192 microcontrollers, distributed across specifically designed printed circuits. Each board manages a block of 16 x 32 pixels, and the various PCBs are arranged in a circular configuration reminiscent of the famous Cray-1 supercomputer. The result is a surface made up of thousands of blinking LEDs, each controlled by its dedicated processor.

The stars of the project are the QingKe CH570, 32-bit RISC-V architecture microcontrollers capable of operating up to 100 MHz. Despite costing around $0.13 per unit, each chip integrates a USB controller, a 2.4 GHz transceiver, and support for Bluetooth 5.0 LE. Even with such a low price, the complete setup of 64,000 units would require an investment of over $8,000 just for purchasing the microcontrollers.

To coordinate such a large number of processors, the project adopts a hierarchical structure. Groups of 32 CH570 chips are managed by a CH32V controller, a solution that distributes control without relying on a single central processor.

One of the main challenges is powering the entire system. Although each unit consumes about 10 milliamperes, the overall consumption grows rapidly with the increasing number of units. The prototype indeed reaches a power demand of around 2,161 watts, equivalent to about 655 amperes at 3.3 volts. To support this load, Balwierz uses a Corsair WS3000 ATX power supply along with custom-designed converters to transform the voltage from 12 volts to 3.3 volts.

Almost every component of the system has been specially developed. The author personally designed the printed circuits, power system, and testing tools. This is also his first six-layer electronic board, a realization that reached the design limits allowed by the PCB manufacturer. Initially, the use of immersion cooling was also planned, a solution later shelved for economic and environmental reasons.

Particularly interesting is also the method chosen to program thousands of devices. Balwierz created a 3D-printed three-pronged tool, which is then installed on the mobile cart of the same printer. A Python script sends G-code commands to automatically position the programmer above each microcontroller, eliminating the need to manually intervene on each individual chip.

The project is still in its early stages and, of course, does not aim to compete with commercial GPUs. The goal is to see how far an architecture based on thousands of low-cost computing units operating in parallel can be pushed.