compusential
Published on

The CUDA Moment: From Bricking a RIVA 128 at 13 to the 8800 GTX

Authors
  • Name
    Yang Pei
    Twitter

The name Compusential came to me on a morning walk to work during my time as a junior web developer at NVIDIA in Theale, Berkshire.

My history with the company had started almost a decade earlier with a memorable childhood failure.

1997: The MediaMarkt Incident and the RIVA 128

In 1997, I was 13 years old, attending international school in Beijing, and looking to spend some of my hard-saved money during a trip to Germany in a MediaMarkt aisle to upgrade the graphics card of my Compaq PC, after reading the latest PC Games (https://www.pcgameshardware.de/).

The obvious darling of 3D gaming was the 3dfx Voodoo Graphics. I opted instead for the dark horse: the NVIDIA RIVA 128 (NV3)—an ambitious all-in-one 128-bit 2D/3D accelerator with 4MB of SGRAM.

Back in Beijing, I decided the card needed a firmware upgrade. One botched ROM flash later, the screen went pitch black.

On our next trip back to Germany, I carried the dead card 7,000 kilometres in my luggage, hoping MediaMarkt would replace it. At the customer service counter, I naively admitted I had flashed the BIOS. The clerk delivered classic German directness:

"Leider ist das Ihre Schuld, und deshalb können wir Ihnen nicht weiterhelfen."
(Unfortunately, that is your own fault, and therefore we cannot help you further.)

That failure sparked a decade of hardware tinkering. Through teenage years in Beijing, I worked through the RIVA TNT2, GeForce 256, and GeForce 4 Ti series, while incinerating a few bare-die AMD Socket A Athlons along the way (those exposed dies had zero thermal throttling; if the heatsink slipped, the chip fried in seconds).

After finishing my Computer Science degree at the University of Reading in 2006 (thesis: Face Recognition in C++), being able to rattle off GPU hardware specs got my foot in the door at NVIDIA.

2006: The CUDA Moment on the 8800 GTX

In late 2006, technical decks and silicon for the GeForce 8800 GTX (G80) arrived at NVIDIA’s UK office. Alongside the hardware came the first release of CUDA 1.0.

Until G80, graphics chips were fixed-function rasterizers. Doing general math meant contorting problems into RGB textures and fragment shaders. CUDA and G80’s unified streaming multiprocessors blew that away: you could write native C kernels and dispatch thousands of hardware threads directly onto silicon.

I was so captivated that I bought an 8800 GTX directly from an add-in board partner (PNY), immediately burning whatever discount I got on a beefy new power supply and dual 6-pin PCIe rails.

Seeing internal engineering demos turn hours of complex numerical simulations into milliseconds of parallel compute made one thing obvious: this was not faster gaming hardware; it was a paradigm shift that would redefine all of computing.

The Essence of Compusential

Walking to the office that morning, I wanted a canvas for that realization:

  • Computational Essence: Stripping away hype to look at raw memory hierarchies, silicon limits, and mathematics.
  • Tinkering & Craft: The same impulse that led a 13-year-old to flash a ROM, applied to reliable production systems.
  • The Lineage of Compute: From G80 streaming multiprocessors to early GPU mining rigs and today’s local 128GB silicon setups.