Engineer builds his own graphics card because modern GPUs are too expensive

A single GPU card can cost thousands, but one YouTuber built his own for a fraction of the price, using 8,192 RISC-V microcontrollers.

bitluni, real name Matthias Balwierz, constructed a DIY GPU using CH570 RISC-V microcontrollers sourced from AliExpress. The cluster delivers a QVGA resolution of 320x200, with each MCU responsible for a single pixel. Despite the project nearly driving him "mad," bitluni pushed through, refining the PCB design and scaling the build to 8,192 chips.

The project began as an experiment in parallel computing but evolved into much more, and, at some cost, bitluni noted that prior GPU builds had already strained his sanity. This iteration, while visually modest, demonstrates the potential of microcontroller clusters for tasks beyond basic graphics.

3

bitluni found his cluster outperformed his 8-core PC CPU in hashing tests, drawing just 4 watts, a glimpse of untapped efficiency. For those who don't know, hashing was once the reason why GPU prices skyrocketed as GPUs were found to be extremely proficient at hashing during the cryptocurrency mining boom.

Though impractical for mainstream use, the build sparks interest in RISC-V for distributed computing, and the possibilities when you're willing to lose a little sanity in an attempt to manifest that idea burrowing away in your brain. bitluni has already hinted at a next-gen version with 32,000 chips, and that will undoubtedly be an interesting build, but at what cost will it be to bitluni?

3

Frequently Asked Questions

TweakBot answers common questions about this news using TweakTown's own coverage from this page and related content from our archive. Tap a question to reveal the answer, or type your own below.

Question #1

Which hashing algorithm did the MCU cluster outperform the 8-core PC CPU on, and what were the comparative hash rates?

Click to reveal answer

Question #2

How is video output generated and synchronized at 320x200 resolution from individual single-pixel MCUs?

Each MCU in bitluni’s build is assigned to generate one pixel of the 320x200 (QVGA) frame, so the full image is produced in parallel by 8,192 CH570 RISC-V microcontrollers. The cluster’s per-pixel outputs are combined to form the frame at QVGA resolution and driven as the video signal; timing and synchronization are achieved by coordinating the MCUs so they output their pixel values in the correct order for the 320x200 scanout.

Answered

Question #3

What limitations or bottlenecks did bitluni encounter that make this approach impractical for mainstream GPU tasks?

Click to reveal answer

Question #4

What differences did bitluni mention (or show) between this build and his proposed 32,000-chip next-gen design regarding layout or expected challenges?

Click to reveal answer

Have a question not listed here? Ask below and TweakBot will answer it.

Your questionAsk

Answering your question×

TweakBot is analyzing your question…

Could not get an answer

TweakBot is generating an answer

Read More

Was TweakBot helpful?Thanks for the feedback!

Hopefully not too much, as this is an extremely interesting series and worth a watch if you are into GPUs or hardware in general.