This post, ATI eyes audio acceleration on the GPU, had me remenicing about the early days of Direct3D again.

What most people don't realize is that in the early days on Direct3D many of the 3D accelerators (or sometimes “decelerators”1) where in fact very general purpose DSPs. At that point in time (mid-1995) many of the hardware guys were very surprised at the rapid mindshare growth of consumer 3D hardware. We're talking people like Evans and Sutherland, SGI, etc…

Brief interlude.

Regarding Direct3D and the following OpenGL v. Direct3D wars, one point I feel is worth making is that if it were not for Direct3D, we would not have the gaming platforms that we have today.

In 1995, 3D was stalled, noone was innovating and OpenGL was stagnant. It was only after the release of Direct3D that OpenGL started to make any headway.

In fact, at lunch during the hardware guys' day at Aftermath event, we put a pre-release PS1 on the main video screen running Tekken in demo mode. You could see the palpable fear in most of the hardware guys' faces. “What is that?” “That's not realtime.” These guys, with very few exceptions, were missing the boat.

End of interlude, now where was I?

Many hardware startups in the Valley at that time were putting together DSP based ISA (or even the new PCI) boards to offload modem, audio and other functionality from the overwhelmed CPU.

These guys recognized a good thing when they saw it and decided that they could add 3D graphics too. It's all just vector math isn't it?

The Chromatic Mpact part was a case in point. It was a modem, an audio card and a 3D accelerator all rolled into one! Step right up! Well, until you tried to run a few thousand triangles per second through it. Mind you, they had some very cool booths at the shows.

None of the parts of this generation performed triangle setup (apart from the 3DFX Voodoo 1) and they all sucked to one degree or another.

But a special place in my heart is taken by the Rendition Veritee 1000.

Rendition was a true entrepreneureal company built by engineers. Those guys were great. They were the only guys to have hardware ready for Comdex 1995 where we (or rather Ty Graham, our hardware evangelist) showed hardware-accelerated Direct3D for the first time.

Servan (one of the founders of RenderMorphics) and I had locked ourselves away in an office on campus after the Aftermath event and we had a week until Comdex. We finished the driver model, an API that could drive it and had the famous2 “tunnel” sample running on it.

But for a driver model we needed a driver and some hardware. Here was where Rendition really stepped up. One of the guys from Rendition came up to Redmond and basically lived with us. He was writing driver code while we were writing the driver model. Talk about bleeding edge.

The prototype, hot off the chip foundry, V1000 was mounted on a red prototype board with a fan glued to it. Unfortunately, when plugged into the bus on the machines we had, the board was upside down and every hour or so the glue would melt and the fan would fall off. Much hilarity ensued. Much pizza was eaten.

At four in the morning on the first day of Comdex, Ty walked into the office, already late for his flight to Vegas, and witnessed tunnel running at lightning frame rates. We packed up the dev machine in a flight case and off he went - by all accounts everyone was awed by this cheap piece of consumer 3D hardware.

I still have that board.

But that's not the point of this narrative.

The point is that the V1000 was a DSP based part that stored it's microcode in it's onboard memory in an address space just below VGA memory.

Which meant that any bug in the driver, especially the clipping code, caused Direct3D to render triangle fragments all over the microcode. It didn't just blue screen. Can you say hard lockup, push the big red reset button?

And here's my other point. Man, was that fun.

[Thanks to Ars Technica for the link.]

1 This term was originally applied to one the first parts out of S3 and I think, ahem, that I coined it…