What a 64,000‑chip DIY GPU RISC-V project really is
A massively parallel DIY GPU RISC-V project is a homebrew graphics system that uses thousands of tiny RISC-V microcontrollers wired together so each chip behaves as a pixel-sized processor, trading conventional GPU silicon for an enormous grid of low-cost, programmable nodes. Hardware hacker Matthias Balwierz, known as Bitluni, is building exactly that: a home-made GPU from thousands of low-cost RISC-V microcontrollers, each paired with an RGB LED so the graphics hardware and display live in the same physical array. His current prototype uses 8,192 CH570 RISC-V MCUs on custom six-layer PCBs, overseen by another 256 larger cores with FPUs, after six months of debugging and redesigning what he now calls his “nemesis”. The key idea is not comfort or practicality, but proving that a wall of cheap, simple chips can act like a parallel processor design for graphics.
From 8,192 to 32,000 chips: when a homebrew GPU stops being a toy
The first working version of Bitluni’s DIY GPU is already absurd: 8,192 CH570 RISC-V microcontrollers at about $0.13 (approx. RM0.60) each, plus 256 bigger controllers on top, all wired into a six-layer PCB that strained his board house’s limits. Each chip runs at 100 MHz with 12 KB of SRAM, and every one drives its own RGB LED, forming a dense display grid where compute and pixels are fused. He has not stopped there. He has already secured parts for a follow-up build with 32,000 MCUs, a fourfold jump over version one. This is where the project stops looking like a quirky LED art piece and starts feeling like a serious parallel processor experiment: debugging crossed MOSI/MISO traces, redesigning noisy layouts, and managing levels of controllers is training for anyone who dreams of custom parallel processor design rather than buying off-the-shelf GPUs.
Why 64,000 homebrew GPU chips still only buy you 320×200
The brutal reality of this homebrew GPU chips experiment is that scale has limits, and they show up first in resolution and power. A system where each pixel is its own RISC-V chip sounds like infinite parallelism, but the math is unforgiving: a Full HD display would need more than two million chips. That is not a quirky weekend project; it is a logistical disaster. So Bitluni scaled his goal back to a 320×200 display, which still needs 64,000 microcontrollers when complete. "A Full HD display would need more than two million chips, so the project was scaled down to 320×200, still demanding 64,000 microcontrollers". The current 8,192-chip prototype is a slice of that final target, and it already exposes the tension between visual ambition and physical feasibility.
Power draw: when parallel processor design meets a wall socket
The most underappreciated lesson in this parallel processor design is not about instruction sets or GPU math; it is about electricity. Each CH570 might sip only about 10 milliamps, but thousands of them turn into a small electrical heater. The current setup already pulls around 2,161 watts, or roughly 655 amps at 3.3 volts. To feed that, Bitluni resorts to a high-end ATX power supply and custom converters to step 12 volts down to 3.3 volts at massive current levels. He even considered immersion cooling but backed away due to cost and environmental concerns. This is the real constraint for the DIY GPU RISC-V dream: before you hit any theoretical limit of parallelism, you hit copper thickness, connector ratings, and the simple question of how much heat your workshop can tolerate.
What this DIY GPU RISC-V experiment tells us about the future
The project will not dethrone commercial GPUs; even its creator admits it is early and nowhere near that league. But that is not the point. It proves that, with patience and accessible tools, an individual can orchestrate tens of thousands of homebrew GPU chips into a working, if modest, display system. Groups of 32 small chips are managed by more capable CH32V controllers, forming a layered control plane instead of one central brain. That hierarchy matters: it is a hint at how future open parallel systems might be structured, whether for graphics, simulation, or art. Bitluni plans to release design files and further performance details in a future video, inviting “anyone insane enough” to try their own build. The real breakthrough is cultural: graphics hardware is no longer something you can only buy—it is something a determined hacker can build, iterate, and understand down to every trace.
