DIY engineer builds his own GPU from 8,192 RISC-V chips

Full text

YouTuber bitluni has unveiled a RISC-V DIY Graphics Processing Unit

Builder has a library of insane DIY projects on his YouTube channel

RISC-V microcontrollers were sourced from AliExpress

YouTuber bitluni has unveiled his latest project, which might put any concerns you have over GPU pricing into context. The German DIYer, who has built up a strong following with his selection of intriguing and off-the-wall projects, has assembled a GPU using a large collection of affordable RISC-V microcontrollers (MCUs).

While the results are impressive for a DIY build, the GPU is unfortunately quite modest, and would probably have been abandoned long before now had a PCB design firm not got in touch to explore a partnership over the build.

Demonstrating the project on his channel, bitluni – real name Matthias Balwierz – expressed how difficult the build was, and how it almost drove him mad.

8000+ RISC-V MCUs output at 320x200 resolution

In his latest video, bitluni recalls a previous attempt at building a GPU in late 2025 which lead to PCB designers Altium getting involved, explaining, “The clusters I made before were already challenging my sanity. I thought I was done with the topic, but the budget and these tools would allow for a cluster of a different magnitude, and the magnitude I had in mind was just insane.”

Pushing things to the next level is a hallmark of bitluni’s videos, and the finished build has an incredible 8,192 RISC-V CH570 microcontrollers, each running at 100 MHz, with 12KB of SRAM. These are mounted on blades, which went through several revisions as the PCB company declared them “too complicated.”

Each CH570 has an array of LEDs mounted, and these correspond to each MCU, and the equivalent QVGA pixel. With 8,192 MCUs, the resulting 320x200 might seem modest, but bitluni already has plans for a more power version with 32,000 MCUs.

Graphics or hashing?

The spike in GPU prices before the AI boom was thanks to industrial-scale cryptocurrency hashing, which graphics cards a particularly suited to. It is this relationship between hashing and graphics that gives bitluni’s project some hope beyond displaying images and video.

Identifying the serial port as the bottleneck to the project’s success as a working (if unwieldy) GPU, bitluni has appealed for help from his community of viewers to find a workaround, but that didn’t bring the project to a close.

Changing tack, he used the microcontrollers for hashing and found it outperformed his PC’s 8 core CPU, with a low 4 watt power draw, so perhaps there is a use for this approach after all.