Someone has posted a link to the x3dfx Forum, where they discovered some interesting news on the upcoming NVIDIA baby aka NV30. Here is what we managed to sum up:

73 million transistors;
0.13micron manufacturing technology;
450MHz chip clock frequency;
External T&L unit and TT&L unit (True Time and Lighting) implemented in a companion chip (supposedly working at 450MHz);
8 rendering pipelines (about the same amount of pipelines is expected to be implemented in ATI R300);
4 TMUs per pipeline capable of laying up to 8 textures in a single pass (loopback);
4 Vertex Shaders units;
2 Pixel Shaders units;
256bit memory access;
Up to 750MHz DDR memory working frequency;
370MHz RAMDAC;
Quad cache for vertex, primitive and pixel textures;
Dual (dounble-level, like Hyper-Z?) Z-Buffer for better lossless data compression;
12nvx – new anisotropic filtering mode (12x4=48, 48 texture samples?);
NVAutoShaper prediction unit also responsible for preliminary saving of data samples in caches, which allows operating the data location manually;
NvBlur: API Glide compatibility;
DirectX9.x compatible;
OpenGL 2.0 support.


-Glide compatibility? Glide is dead, let it rest in peace.
-I wonder why is there no mention of AA.
-nVidia has already said it won't support 256bit bus because its too expensive and would be overkill.
-The rest of the specs look interesting though.