4 ms·

Are you talking about affine transformation vs perspective correct texture mapping? Because the n64 for sure had the later, and ps1 the former.

by amlib 4y ago

midnightclubbed 4y ago
I was confused by this too. N64 texture mapping hardware has the same capabilities of current generation GPUs - perspective correction, automatic mip-level selection, bilinear/trilinear interpolation. The blurriness on the N64 was down to the hardware requiring each texture fit into a 4KB cache (and games using low resolution video output to control pixel fill-rate). The texture unit on the N64 is really powerful but the 4KB size limit and additional usage restrictions leads to its signature look: - If you want texture repeats/wraps/clamps the texture dimensions must be power of 2. - If you want mip-maps then you halve the size of the most detailed texture mip. - If you want detail textures then you halve again (I think, bit fuzzy on that mode). - If you want to use a color lookup (from 4 or 8bit pixel values) then the LUT takes 2 of your 4KB, even for 4bit pixels. Most of the pixel formats have options for 1 or 4 bits of transparency, but transparent pixels render at half speed (and the alpha bits further eat into your 4KB limit). For something like Goldeneye you are likely going to use as many 4 or 8bit mip-mapped, repeating, greyscale textures as possible (64x64 or 64x32) and use vertex tinting to add color. Everything will look a little monochromatic but you will be controlling blurriness as best you can! On early 3d games the art team would spend enormous amounts of time hand tweaking individual pixels and carefully specifying texture formats per texture in order to get the most detail out of the available resolution. It wasn't all bad though, they had a couple orders of magnitude fewer textures to produce (than a current generation game) and their polygon counts were 2-3 orders of magnitude lower too!
- amlib 4y ago
  Such a shame the SGI engineers didn't try allocating more silicon to that tiny texture memory, this is perhaps (not an expert so don't quote me) the biggest bottleneck and annoying constraint in the system. It gets even crazier when you compare how much space the VI filter/blur takes on the silicon, removing it could easily allow for double the cache memory.

Are you talking about affine transformation vs perspective correct texture mapping? Because the n64 for sure had the later, and ps1 the former.

by amlib 4y ago

Are you talking about affine transformation vs perspective correct texture mapping? Because the n64 for sure had the later, and ps1 the former.

midnightclubbed 4y ago

I was confused by this too. N64 texture mapping hardware has the same capabilities of current generation GPUs - perspective correction, automatic mip-level selection, bilinear/trilinear interpolation. The blurriness on the N64 was down to the hardware requiring each texture fit into a 4KB cache (and games using low resolution video output to control pixel fill-rate). The texture unit on the N64 is really powerful but the 4KB size limit and additional usage restrictions leads to its signature look: - If you want texture repeats/wraps/clamps the texture dimensions must be power of 2. - If you want mip-maps then you halve the size of the most detailed texture mip. - If you want detail textures then you halve again (I think, bit fuzzy on that mode). - If you want to use a color lookup (from 4 or 8bit pixel values) then the LUT takes 2 of your 4KB, even for 4bit pixels. Most of the pixel formats have options for 1 or 4 bits of transparency, but transparent pixels render at half speed (and the alpha bits further eat into your 4KB limit). For something like Goldeneye you are likely going to use as many 4 or 8bit mip-mapped, repeating, greyscale textures as possible (64x64 or 64x32) and use vertex tinting to add color. Everything will look a little monochromatic but you will be controlling blurriness as best you can! On early 3d games the art team would spend enormous amounts of time hand tweaking individual pixels and carefully specifying texture formats per texture in order to get the most detail out of the available resolution. It wasn't all bad though, they had a couple orders of magnitude fewer textures to produce (than a current generation game) and their polygon counts were 2-3 orders of magnitude lower too!

amlib 4y ago
Such a shame the SGI engineers didn't try allocating more silicon to that tiny texture memory, this is perhaps (not an expert so don't quote me) the biggest bottleneck and annoying constraint in the system. It gets even crazier when you compare how much space the VI filter/blur takes on the silicon, removing it could easily allow for double the cache memory.