Sanglard spent extensive time cataloging every visible computer in the film, from Nedry's SGI Crimson running IRIX 3.5 to the Thinking Machines CM-5 to the Macintosh IIfx kiosks. He treats the film as a visual archive worth documenting bezel-by-bezel, keyboard-by-keyboard, arguing that the specific hardware choices reflect a genuine snapshot of high-end 1993 computing.
By submitting the piece to Hacker News where it hit #1 with 707 points, vinhnx amplified the view that this level of hardware archaeology is worth the community's attention. The strong upvote count reflects broad agreement that the specific machines — SGI Crimson, CM-5, Quadra 950 — deserve to be remembered in detail.
The editorial argues Sanglard's post understates the darker subtext: by 2010, every hardware vendor featured in the film — SGI, Thinking Machines, Cray, Apple's Quadra line — was dead, bankrupt, or a shell brand slapped on x86 boxes. SGI's eventual $42.5M sale to Rackable was less than the price of a single loaded Onyx in 1993, making the film an unintentional monument to the collapse of proprietary high-end workstation economics.
Fabien Sanglard, the same person who reverse-engineered the Wolfenstein and Doom source code, spent an unreasonable amount of time cataloging every computer visible in Jurassic Park (1993). The [writeup](https://fabiensanglard.net/jurrasic_park_computers/index.html) hit #1 on Hacker News with 707 points, and it's exactly what the title promises: excruciating detail. Bezel shots, keyboard layouts, monitor stands, the specific shade of teal SGI used on the Crimson case.
The roster is a who's-who of early-90s specialty compute. Nedry's workstation is a Silicon Graphics Crimson running IRIX 3.5 — that's the machine behind the famous "It's a UNIX system! I know this!" scene, and the 3D file browser Lex uses is real SGI software called `fsn`. The control room has more SGI Indigos and a Mac Quadra 950. The visitor center kiosks are Macintosh IIfx. The park's actual compute muscle — Nedry references "eight connection machines" — is a Thinking Machines CM-5, the black tower with the blinking red LED grid that also cameos in the background. Industrial Light & Magic rendered the dinosaurs on a Silicon Graphics Onyx with a Cray X-MP as the render farm.
Here's the part Sanglard's post doesn't dwell on: by 2010, every one of those hardware companies was either dead, sold for parts, or a shell using a recognizable name to sell x86 boxes. SGI filed Chapter 11 in 2006, filed again in 2009, and was acquired by Rackable Systems for $42.5 million — less than the price of a single loaded Onyx in 1993. Thinking Machines filed for bankruptcy in 1994, the year after the film released. Cray was acquired by SGI (yes, that SGI), spun back out, and eventually landed at HPE in 2019. Apple killed the Quadra line by 1995.
The glib reading is nostalgia: look at these gorgeous industrial-designed workstations, weep for a lost era of ambitious hardware. The interesting reading is that Jurassic Park is a preserved snapshot of a business model that has died three times since and is currently trying to die again.
In 1993, if you needed to render a photorealistic Tyrannosaurus, you bought an Onyx. You didn't have a choice. SGI owned the graphics pipeline end-to-end: custom MIPS CPUs, custom RealityEngine graphics boards, custom IRIX operating system, custom GL libraries (which they eventually opened as OpenGL — the one part that outlived them). A loaded Onyx cost roughly $250,000 in 1993 dollars. Studios paid it because the alternative was renderfarms of PCs that couldn't do the job.
What killed SGI wasn't a competitor building a better Onyx. It was PC-class hardware plus a commodity GPU getting to "good enough" while costing 5% as much. By 1999 an NT workstation with a Voodoo or GeForce card was rendering scenes that would have needed six figures of SGI kit five years earlier. The margins collapsed. The proprietary software moat — IRIX, custom GL extensions, workstation-only tooling — became a liability the moment DirectX and Windows became credible for production work. Thinking Machines died the same way, just faster: Beowulf clusters of commodity Linux boxes replaced $30M CM-5s for scientific compute inside a decade.
The community reaction on HN caught the throughline. Top comments zeroed in on how the film's opening shot — a control-room engineer squinting at a serial-terminal login prompt — looked like the future in 1993 and looked like a museum exhibit by 2005. One thread traced how many of the specific SGI engineers who built the RealityEngine ended up at NVIDIA, ATI, or 3dfx. The people survived; the vertical-integration business model didn't.
It's tempting to file this under "cool old hardware, RIP" and move on. Don't. The exact pattern that flattened SGI is currently the load-bearing question in your AI infrastructure budget.
NVIDIA in 2026 is SGI in 1993, structurally. Custom silicon (H100/B200 vs. RealityEngine). Proprietary software moat that everyone grudgingly standardizes on (CUDA vs. IRIX+GL). Margins that make finance people cry with joy (75%+ gross vs. SGI's 60%+ at peak). A customer base that resents the pricing but has no credible alternative for the top-tier workload. If you've ever priced an 8xH100 node against what it can actually do, you've had the exact conversation a VFX studio had in 1994 pricing an Onyx2.
The lesson from Jurassic Park's server room isn't that specialty compute always loses. It's that specialty compute loses the moment a commodity stack plus a portable software layer reaches "good enough" for the median workload. SGI kept the top 5% of graphics work — feature-film rendering, high-end scientific visualization — for another decade. But the top 5% doesn't fund a public company. The middle 80% does, and the middle 80% ran off to cheaper hardware the moment the software stopped being a lock-in.
For your stack: assume the CUDA moat is on a clock. AMD ROCm is where DirectX was in 1998 — not there yet, credibly moving. Google TPUs are Beowulf clusters circa 1999 — cheaper per FLOP for the specific workload they were built for, awkward for anything else, but the workload keeps growing. The bet isn't "NVIDIA will collapse in 2027." The bet is that the workload-to-hardware coupling is loosening every quarter, and locking multi-year infrastructure to a single vendor's proprietary compilation target is how you become the studio that owned twelve Onyx2s in 2003.
Sanglard's post is a nostalgia piece if you want it to be. It's a case study if you don't. Every dominant compute vendor eventually meets a commodity competitor that's 80% as good and 5% of the cost, and the transition is faster than the incumbent's org chart can react to. The interesting question isn't whether it happens to NVIDIA — it's whether the software abstractions in your codebase today let you switch when it does. If your inference layer talks CUDA directly, you're SGI's customer in 1997, feeling smug. If it talks through a portable IR, you're the studio that quietly moved to a PC renderfarm and kept shipping.
> Some code associated with Nedryland is visible on screen. It looks like actual source code[9] with Classic Mac OS API functions callsThe source code shown is example code included with the Macintosh Programmers Workshop, Apple's original IDE for the Mac. Originally sold as a separate produ
My wife worked for Thinking Machines back then. I remember that they'd asked Cray to loan them a supercomputer for the film because that's the computer used in the book. Cray brushed them off, so they turned to Thinking Machines who were happy to do it.To thank them, the producers rented a
My uncle (John Monsour) worked on this movie as the “24 Frame Computer Sync Engineer”. Because film cameras and CRT monitors have different frame rates, you needed to use specialized electronics to synchronize them with the camera frame rate otherwise you would have banding and weird moving artifact
It was indeed a Thinking Machines CM-5 — Nedry actually mentioned them in his line about how Hammond wouldn't be able to find anyone "anybody who can network 8 connection machines".An actual assembled CM-5 actually cost closer to a million dollars.But, from what I remember the one in
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
> It is unclear how Jurassic Park crew got their hands on a Motorola EnvoyThe head of frogdesign (Hartmut Esslinger) ended up running into Spielberg on a plane and showed it to him. The one in the movie is an original mockup.Source: https://www.therpf.com/forums/threads/j