Here is what ShugahShack says.
- M-Buffer? : "M" stands for multi-sample. ... in the multi-sampling method, used by Spectre, only one texture lookup per *pixel* (so that all subsamples use the same texture lookup value) is required. ...it reduces the amount of texture lookup by a factor of 4 compared to the VSA-100 method...
- Recursive textures : Texturing flexibility by supporting N (8 for Spectre) independent, unique textures applied per pixel in a single rendering pass... Also reflection on water.
- Overbright (52-bit color) : Increased internal precision of 52-bits (signed 13-bits per RGBA color component), generating rendered images of substantially higher quality. The higher dynamic range reduces darkening resulting in more vibrant and realistic images.
- Perlin Noise :
- Vertex cache :
- Natural cubic spline :
- Higher Order Surfaces (HOS) : Curved surface maps representing textures. Includes Bezier, B-Spline, Catmull-Rom spline patches
- Curved patches : A curved surface created from two or more curves.
- B-Spline :
- Catmull-Rom spline patches :
- Bezier :
I do not know every one of these things, but ill talk about what i do know in a sec. For now, lets talk about that wonderful little 8 texel per clock number
We have a 2 chip card here. Therefore, I am going out on a limb and assuming that each chip has two pipelines. I want to say that it will have 1 pipeline per chip, but that would leave the pixel fillrate pitifully low, as you will see later. The card has been rumored to be running at 200/200, so that means our final numbers come out to be 800mpix and 6400mtex. This seems unnaturally imbalanced as you look at 3dfx's past, where the V5 5500 has 2mpix/2mpix per chip per clock. However, very few of the games will use more than 3/4 textures per pixel, so that number is probably pretty overkill, even though John Carmack was looking for 8tex per pixel for Doom 3. This info comes from a .plan update.
Either way, that is what the rumors are telling us.
From here, we need to look at memory bandwidth. 200mhz DDR memory on a 2 chip solution will be competing with a 250mhz DDR (expected) memory/1 chip solution from NVIDIA. I am unsure how to calculate the bandwidth's, but this obviously gives 3dfx a huge advantage.
There will also be HSR, and it will probably be slightly more potent than Nvidia and ATi's, as 3dfx can use tricks from Gigapixel.
Lets look at the features now:
-Overbright seems like the same thing that Matrox uses to increase 2d quality, but i may be wrong.
-Multi-Sample is using the same texture value and offsetting it slightly, rather than supersampling, which created 2/4 (2x/4x FSAA) values independently and offsetted them. Look for HUGE increases here in FSAA, making it an essential. 8x FSAA will be there, because it was on the V5 6000... Nvidia has this feature also.
-Curved surfaces will be created in hardware
My question, what is the vertex cache? Someone wanna jump in on this one?
Finally, a disclaimer. This is completely too early to make this thread, but i thought it would be fun. It is probably wrong, but it seems to add up fairly well. Lets talk about it guys, what do you think?
- M-Buffer? : "M" stands for multi-sample. ... in the multi-sampling method, used by Spectre, only one texture lookup per *pixel* (so that all subsamples use the same texture lookup value) is required. ...it reduces the amount of texture lookup by a factor of 4 compared to the VSA-100 method...
- Recursive textures : Texturing flexibility by supporting N (8 for Spectre) independent, unique textures applied per pixel in a single rendering pass... Also reflection on water.
- Overbright (52-bit color) : Increased internal precision of 52-bits (signed 13-bits per RGBA color component), generating rendered images of substantially higher quality. The higher dynamic range reduces darkening resulting in more vibrant and realistic images.
- Perlin Noise :
- Vertex cache :
- Natural cubic spline :
- Higher Order Surfaces (HOS) : Curved surface maps representing textures. Includes Bezier, B-Spline, Catmull-Rom spline patches
- Curved patches : A curved surface created from two or more curves.
- B-Spline :
- Catmull-Rom spline patches :
- Bezier :
I do not know every one of these things, but ill talk about what i do know in a sec. For now, lets talk about that wonderful little 8 texel per clock number
We have a 2 chip card here. Therefore, I am going out on a limb and assuming that each chip has two pipelines. I want to say that it will have 1 pipeline per chip, but that would leave the pixel fillrate pitifully low, as you will see later. The card has been rumored to be running at 200/200, so that means our final numbers come out to be 800mpix and 6400mtex. This seems unnaturally imbalanced as you look at 3dfx's past, where the V5 5500 has 2mpix/2mpix per chip per clock. However, very few of the games will use more than 3/4 textures per pixel, so that number is probably pretty overkill, even though John Carmack was looking for 8tex per pixel for Doom 3. This info comes from a .plan update.
Either way, that is what the rumors are telling us.
From here, we need to look at memory bandwidth. 200mhz DDR memory on a 2 chip solution will be competing with a 250mhz DDR (expected) memory/1 chip solution from NVIDIA. I am unsure how to calculate the bandwidth's, but this obviously gives 3dfx a huge advantage.
There will also be HSR, and it will probably be slightly more potent than Nvidia and ATi's, as 3dfx can use tricks from Gigapixel.
Lets look at the features now:
-Overbright seems like the same thing that Matrox uses to increase 2d quality, but i may be wrong.
-Multi-Sample is using the same texture value and offsetting it slightly, rather than supersampling, which created 2/4 (2x/4x FSAA) values independently and offsetted them. Look for HUGE increases here in FSAA, making it an essential. 8x FSAA will be there, because it was on the V5 6000... Nvidia has this feature also.
-Curved surfaces will be created in hardware
My question, what is the vertex cache? Someone wanna jump in on this one?
Finally, a disclaimer. This is completely too early to make this thread, but i thought it would be fun. It is probably wrong, but it seems to add up fairly well. Lets talk about it guys, what do you think?
Facebook
Twitter