A look at the community of hobbyists of the Silicon Graphics Inc supercomputers like the Onyx, which cost up to $250K on release and weighed 250 pounds
Nearly 30 years after Silicon Graphics ruled the high-performance computing roost, its supercomputers have found themselves a new home …
Context & Ripple Effects
Silicon Graphics spent the early 1990s selling the Onyx — a 250-pound, up-to-$250K visualization supercomputer — to studios and labs that needed graphics muscle no workstation could deliver. Nearly three decades later the company is gone, but the machines persist in a hobbyist community that scavenges, repairs, and runs them, much as the IBM mainframe survived its own era of predicted obsolescence through a devoted installed base.
The niche SGI once owned — an expensive, heavy, specialized box at the top of the performance pyramid — has a clear modern heir: Nvidia's DGX H100 clusters, which pair ~300-pound enclosures with eight $25K GPUs apiece. The hobbyist story is a reminder that today's flagship AI hardware will eventually face the same afterlife.
First-order effects
- Hobbyists become the de facto maintainers of Onyx hardware: with Silicon Graphics defunct, parts, OS images, and repair knowledge circulate only through the collector community, setting the machines' survival odds.
- The Onyx's collapse from $250K flagship to surplus collectible reprices an entire class of 1990s HPC gear on the secondhand market, where value now rests on rarity rather than capability.
Second-order effects
- Vendors of today's equivalent hardware — Nvidia above all — inherit the same lifecycle: DGX-class systems built for a handful of buyers will eventually flood secondary markets, seeding future enthusiast communities the way decommissioned SGI rigs did.
- Preservation efforts around SGI machines raise the bar for documentation and emulation of proprietary architectures, since hobbyists must reverse-engineer what vendors never open-sourced.
Third-order effects
- If the mainframe and SGI patterns hold, every generation of flagship compute leaves behind a long-tail ecosystem of owners who keep dead platforms alive for decades — an argument for archival software, schematics, and toolchains even for closed, enterprise-only systems.
- The cycle also suggests vendor lock-in cuts both ways: proprietary platforms die with their makers, while the surviving machines become artifacts whose value is historical, underscoring why open standards tend to outlive the companies that championed them.
The trend: Flagship specialized computers keep outliving the companies that built them, migrating from corporate data centers into enthusiast hands as each new performance leader — from SGI to Nvidia — takes over the top of the market.