BFG10K
Lifer
- Aug 14, 2000
- 22,709
- 3,003
- 126
PotNoodle:
Quake3 takes the easiest method of compression and ?blanket? compresses everything under DXT1.
Not anymore.
decompression of DXT1 (and the number of bits that it results in using) has no bearing on the size of the compressed texture;
I'm not talking about decompression. I am talking about the textures while they're in a compressed state. DXT1 compression has a higher compression ratio than DXT3 (because it uses less bits), hence it's faster and uses less memory bandwidth. By making DXT1 store even less stuff than their competitors, nVidia's scheme works faster.
No, they weren?t.
Um, yes they were. Lets say two companies have identical boards, but company A uses 2 bits to store textures and company B uses 4 bits. As a result, company A's textures will be half the size (while compressed) and will use half the memory bandwidth compared to company B's textures.
Quake3 takes the easiest method of compression and ?blanket? compresses everything under DXT1.
Not anymore.
decompression of DXT1 (and the number of bits that it results in using) has no bearing on the size of the compressed texture;
I'm not talking about decompression. I am talking about the textures while they're in a compressed state. DXT1 compression has a higher compression ratio than DXT3 (because it uses less bits), hence it's faster and uses less memory bandwidth. By making DXT1 store even less stuff than their competitors, nVidia's scheme works faster.
No, they weren?t.
Um, yes they were. Lets say two companies have identical boards, but company A uses 2 bits to store textures and company B uses 4 bits. As a result, company A's textures will be half the size (while compressed) and will use half the memory bandwidth compared to company B's textures.