In a tech tale filled with ambition and a pinch of regret, Jeff Geerling embarked on a daring journey to create an AI cluster using Raspberry Pi boards, investing a staggering $3,000 in the process. While his creation is undeniably impressive, it seems the dollars didn’t translate into the computing power he had hoped for.

Jeff's build is compact and energy-efficient—attributes that often make the Raspberry Pi a favorite among makers. However, the glaring absence of a GPU in his configuration left it lacking when it came to performance. Despite multiple rebuilds—Jeff tinkered twice, adjusting SSDs and affixing heatsinks—it still didn’t meet the mark.

Performance Puzzles

With ten compute modules, his cluster managed to achieve around 325 gigaflops during testing. At first glance, that sounds decent, but when you stack it against a Framework Desktop x4—which boasts an impressive 1,180 gigaflops—the weaknesses become glaringly clear. The Framework not only outperformed it significantly but also offered better value per dollar spent.

AI Aspirations Fall Flat

In the quest for AI capabilities, the Raspberry Pi 5’s built-in GPU unfortunately couldn't support the Vulkan version needed for speed optimizations. Even a budget-friendly consumer PC showcased superior performance, outpacing the cluster by six to seven times without a dedicated GPU, and an astonishing 14 times with one.

The Verdict: Proceed with Caution

So, is this ambitious Raspberry Pi build worth pursuing? Jeff’s answer is a resounding no—unless, that is, you possess a very specific use case that justifies it. Yet, for hobbyists and creators who thrive on experimentation, sometimes the journey is just as valuable as the destination. Have you ventured into similar projects? Share your experiences below!