Grabberboy at 30: chasing video signals in the twilight of the Amiga

Thirty-odd years ago, me and Olof Zachrisson were engineering students on the Computer Science and Engineering programme (the "D-linjen", for datateknik) at Linköping, and like a lot of people at the time we were Amiga fans. Not just users — we wanted to make something for the platform. What we made was Grabberboy, a video grabber for the Amiga. This is the story, reconstructed from memory, from our old master's thesis (exjobb) which I recently found as a stack of scans, and from project files that survived on an old A1200 hard drive.

The thesis — "Ett digitalt system för inspelning och redigering av videosekvenser", Institutionen för Systemteknik, May 1994 — was formally about building a cheap system for recording video sequences. Single-frame grabbers existed and were affordable; recording moving video meant professional gear costing tens of thousands of kronor, or consumer products so limited (8 Hz frame rates, sequences bounded by RAM) that we judged them useless even for hobbyists. We thought we could do better by being clever instead of expensive.

(The full report is scanned here, in Swedish: Ett digitalt system för inspelning och redigering av videosekvenser (PDF, 130 pages).)

Six boards

We didn't build one grabber. Counting from the report: six boards were wire-wrapped over the years, five within the thesis work.

It started before the thesis was even an idea. GB I was built for the Digitalteknik fortsättningskurs: an expansion card for the A2000/A3000 with 32 kB of static RAM, enough for exactly one 256×256 image at 16 gray levels. The sample controller fit in a single Altera EP910. It could sample in real time but only hold one frame, so recording speed was bounded by how fast the Amiga could shovel data over the bus — in practice about 12 frames per second straight to hard disk.

That card got us hooked, and the successors grew more ambitious: programmable resolution, 25 fps independent of resolution, and a memory design where the Amiga saw the grabber RAM as true 0-waitstate fast RAM while sampling was going on. Making that work fell to two custom controllers with suitably self-confident names: BEAST, the sample controller, and RAMSES, the memory controller — which the report dryly notes "has created most of the problems". RAMSES had to arbitrate between the Amiga's memory cycles, high-rate sample bursts in fast page mode, and DRAM refresh, at 28 MHz, in a synchronous design, using programmable logic that was barely fast enough. The EP610-30 datasheets said it should work. It did not always agree.

Somewhere in here my memory had upgraded the technology to FPGA and VLSI. The paper trail says PAL and EPLD logic, wire-wrap, and a remark that JPEG VLSI chips existed but were too expensive. Thirty years will do that to a story.

When the A1200 arrived, its quadrupled bus bandwidth let us throw out the FIFOs entirely, and the final boards (GB 1200s, GB 1200s II, GB 1200d) were built around it. The design sampled the raw composite signal at three times the color clock, ~18 MHz, with color separation meant to happen in software. That's why black-and-white worked well and color remained the hard part — the color path existed mostly as a chapter of theory in the report.

Phil Collins, twenty seconds

To show what the hardware could do, we recorded about twenty seconds of Genesis "I Can't Dance" — the video with the guys in suits dancing badly — directly from MTV. There was something fitting about testing video capture hardware on the least dynamic music video ever made.

That recording became our demo loop, and it still exists: Phil.anim, a 3.4 MB IFF ANIM file dated February 1992, sitting in a directory on the old A1200 disk next to grabs of Madonna and something called CrazyDuck.

The song was Phil Collins because Olof was a fan; I was a bit of a hipster in my youth and wouldn't have picked it myself. Funny thing, though: I'm writing this thirty years later with a Dave Clarke post-techno electro set playing, and the jerky 16-gray loop fits that vibe perfectly. Some things age into place.

Köln

We traveled to the big computer fair in Köln, walking the floor and studying the competition. The trip report survives, in its entirety:

Vi hittade tre "konkurrenter" på Köln-mässan, nämligen:

VLab från Macro System, FrameMachine/FrameStore från electronic-design, 748/798 DM ?? Mkt lik Vlab, 448 DM (Ej mässpris!)

The thesis's conclusion from that study: the affordable competition maxed out at 8 Hz with limited resolution. We were, at least on paper, ahead.

What the trip report doesn't record is the visit to das Meter, a bar where — as I recall it — everything came by the meter: a meter of beer, a meter of sausage. Some field research doesn't make it into the report.

Toolbox Design

Around the project a small collective formed. Our thesis advisor Clas Tegenfeldt was an Amiga enthusiast himself — founder of user groups, PhD student in computer graphics — and he had the network. Toolbox Design HB became the commercial umbrella, founded by Clas together with Thomas Svenson, who ran Procomp, an Amiga software import business.

There was also a PhD guy who had written Effi, a system-wide scalable font engine for the Amiga — outline fonts in any application, instead of being locked to bitmap fonts or specific DTP packages. Deep system-level hacking, impressive stuff.

Datormagazin's report from AmiCon '92 at Sollentunamässan mentions us:

En rekordstor publik hade kommit för att bl.a. se nyheter som prototypen till ett 24-bitars grafikkort från M.A.S.T. och Grabberboy, från Toolbox Design, som digitaliserar direkt till hårddisk.

Between us all we had more or less the entire Commodore lineage: A1000, A500, A600, A1200, A3000, A4000. Very useful for testing — different 68k processors, different bus behavior. Several promising products, a collective of enthusiasts, machines everywhere. Stuff was happening.

When the signal faded

It fell apart, the way these things do, from several directions at once.

Productifying the video grabber turned out to be genuinely hard. A working wire-wrapped prototype is a long way from a product you can manufacture, ship and support, and we were students without the resources to close that gap. Color never got easy.

Clas got "elallergi" — electrical hypersensitivity. He kept working, which considering the workbenches and CRTs everywhere says something about his stubbornness, but it was really hard for him.

The font guy decided to stop working with us. The departure was not amicable. Effi never surfaced anywhere else afterwards, as far as I can tell — the code presumably still sits on some hard drive. A piece of lost Amiga system history.

And the timing: our report is dated May 1994. Commodore filed for bankruptcy in April 1994. We handed in a thesis about a product for a platform whose manufacturer had ceased to exist the month before. Escom bought the remains in 1995 and went bankrupt themselves a year later. There was not going to be a market.

Thirty years later

Grabberboy never made it to retail shelves. But I don't think of it as a failure. We designed real hardware as students — six boards, two custom memory and sampling controllers, a demo that could grab MTV to hard disk in real time — and stood at fairs among people who believed the Amiga was the future of multimedia. For a while, watching Phil Collins loop in 16 gray levels on hardware we'd wired ourselves, so did we.