From 1995 to 1998, I worked at the short-lived Acclaim Coin-Op to work on Magic: the Gathering — Armageddon, a 1- or 2-player arcade game based on MtG that was basically a 45-second real-time strategy game.

It was an adventure to work on and it’s a bummer that despite good performance in testing at Sunnyvale and Milpitas Golfland and thousands of orders, we didn’t make it to full production and Acclaim shut us down.

Blitting

In 1995 we set out with an ambitious goal — to have an arcade game running in high res (640x480, lol) and 60 Hz. With a bunch of overdraw, that meant we needed sustained blitting of around 50 million pixels per second.

While that number is laughably small today, nothing in that era could do it. We experimented with the oh-so-sexy TI TMS320c80, every graphics card from that era, you name it. Nothing came close.

We kept working and solved other problems (unlike Killer Instinct we couldn’t use a hard drive for streaming animations, so we built an ASIC for streaming jpeg decompression) but we were striking out on our core graphics pipeline.

Then we met the 3dfx folks.

While Voodoo I wasn’t quite enough, they said they had built it to be able to scan line interleave a scene with two cards, effectively doubling fill rate.

Eureka!

So, much like Atari Games, we picked 3Dfx and started building our arcade boards around it — in our case pairing it with a MIPS R5000. It was pretty sexy for the time — the R5k and the main bus — including the two Voodoo I chipsets — ran on a 200 MHz main bus. The only test setup we could use to emulate it during development was a brand new Pentium Pro and WinNT with OpenGL. In 1997.

We all learned a lot in this process, including but not limited to:

• If you don’t pay vendors on time, they’ll ship you known bad chips which in our case were R5ks with malfunctioning absolute value op codes
• When you discover this — and if you and your hardware engineers are mad enough — you can successfully de-solder a ball grid array
• Running Voodoo I at exactly 60 Hz would lead it to lock up, so we jittered around the timing
• Glide had some hilarious comments in it

But our best one was the week of the 3pm crash. Every day, around 3pm, our test cabinet would crash. Hard. Like no logs, nothing. Even hardware logging taps told us bupkis.

Readers who’ve done hardware development already know the answer. In our Mountain View office — just down the street from the soon-to-be former Alias campus Google would take over and above a really boring group of math nerds called RSA who had invented something called “public key cryptography” — the sun hit our hardware lab that month about 3pm.

Yup, thermals.

One much bigger heat sink later, we were off to the races.