Why 3dfx 1 cards had such a washed out color output?
Early 3D accelerator cards like the 3dfx Voodoo often produced “washed out” images because they used a nonstandard gamma curve and analog signal quirks that didn’t match how CRTs actually responded to voltage. Commenters contrast those legacy pipelines with modern practice, where textures are authored in sRGB, converted to linear space for lighting and blending, then mapped back to display color spaces to preserve contrast—especially in dark areas. The thread branches into related topics such as colored lighting in 1990s game engines, HDR’s real benefits and pitfalls, and how evolving APIs and hardware design shaped the visual “style” of early 3D games.
3dfx gamma and the “washed out” look
- One explanation: VGA output was effectively linear (gamma ≈ 1), while CRTs had gamma ≈ 2.2.
- 3dfx defaulted to gamma ≈ 1.5, so combined with the CRT’s 2.2 this moved the overall system closer to linear light than standard VGA, but still not “correct.”
- This shifted precision away from dark tones and could make images appear flatter or “washed out,” especially with 16‑bit (565) color.
- Some suggest this was partly a “brighter looks better” marketing choice or to avoid “wasting” bits in deep shadows.
- Gamma could be changed via a driver call or DOS environment variable to match ordinary VGA behavior, but most developers and users didn’t understand or adjust it.
Analog output, DAC tricks, and voltage issues
- Beyond gamma, several comments point to analog signal level mismatches between cards and pass‑through setups (e.g., 2D card → Voodoo → monitor).
- Different cards had different black/white voltage levels and sync timing, causing differing brightness and image shifts; this is independent of gamma.
- 3dfx DACs also applied a blur/filtering pass that smoothed dithering (“22‑bit” marketing), further softening the image and slightly altering colors.
Quake‑era rendering and color
- Software Quake/Quake II often lacked proper colored lighting; OpenGL paths added it, producing a characteristic reddish/yellowish tint that some players noticed.
- Attempts at “overbright” lighting and hacked gamma ramps increased highlight pop but worsened banding in dark areas.
- Commenters marvel at how efficient the original software renderers were on limited CPUs.
Color spaces and linear vs non‑linear workflows
- Multiple posts discuss sRGB’s non‑linear transfer function and its relationship to CRT behavior; it approximates but doesn’t perfectly match CRT gamma.
- There is debate on when to work in linear space (lighting, exposure, calibration) vs perceptual spaces (e.g., OKLab‑derived models) for better editing controls.
- Some note that naïve “correct” linear workflows can introduce banding when constrained to 8‑bit outputs.
HDR, banding, and video
- HDR is framed mainly as extended luminance range (higher peak nits) plus new transfer functions (e.g., PQ, HLG).
- Implementation on PCs and cheap TVs is described as messy: poor tone mapping, large dimming zones, and global transforms that ignore local contrast can all cause bad results.
- Banding in dark video streams is attributed more to codec behavior and display setup than to linear color, since consumer video is almost never stored in linear space.
3dfx vs. Nvidia and industry context
- 3dfx is remembered as the early PC 3D leader that stumbled by clinging to its proprietary API and moving into board manufacturing, alienating partners.
- Nvidia (and ATI) leveraged open APIs (OpenGL, Direct3D) and faster iteration to win the market.
- CUDA vs. OpenCL is likened to Glide vs. competitors: easier to start with, early‑mover advantage, and long‑term support creating a strong moat.