Some LCD's use 6bit per color panels, they're cheaper and good enough for most people anyway, most won't notice the difference, TN-panels (used at least while back on most "gaming tft's") are all 6bit per color I think
Future Hardware
- Thread starter rwolf
- Start date
Some LCD's use 6bit per color panels, they're cheaper and good enough for most people anyway, most won't notice the difference, TN-panels (used at least while back on most "gaming tft's") are all 6bit per color I think
Rainbow Man
Veteran
Not saying anything about the petaHz because some day it might happen.. Maybe by using diamond transistors or whatever. though aiming for a specific extremely high clockspeed is in of itself a stupid idea just like intel's venture with the pentium4 was; reaching for actual performance is better)..
The 0 memory latency is going to be more difficult on the other hand methinks. That actually implies faster than lightspeed data transmisison speed which is sure to run afoul of a law of physics or two..
Clearly we need Scotty to help design the GPUs of the future!
peace.
Anarchist4000
Veteran
Or some really good caching.
icecold1983
Banned
heres an example of where i dont think 16x af is enough
heres an example of where i dont think 16x af is enough
That thingy doesn't look really that demanding on that shot to not allow a combination of 2x or 4x superasampling with 16xAF (I think a third party application like DXTweaker might help here). Besides that's a corner case and I sure hope that track has some curves after that too long straight line.
Blazkowicz
Legend
I want rotated grid supersampling. Why is it only supported by the voodoo5 and S3 chrome (which doesn't even have multisampling), and not by the cards with insane power? (geforce 6/7/8, and the ATI stuff).
3dfx used hardware for that (the T-buffer).. while S3 is doing it with a driver trick?
3dfx used hardware for that (the T-buffer).. while S3 is doing it with a driver trick?
It's easy to have that on D3D if you use a deferred renderer..
Blazkowicz
Legend
did the Oak5 dinsosaur did that?
I want unconditionnal support. it worked every time on V5 not matter the game, engine or API.
such RGSS might help for shader aliasing, ultra sharp texture mode (AF 16 + LOD -1.5), and the alpha textures, shadows, rendering techniques incompatible with MSAA, etc.
icecold1983
Banned
i dunno, theres a huge distance between where the texture is blurring and where the straight away ends, i hope we get at least 32x af soon.
Recursive functions and a stack...
You can easily implement a stack in DX10 already using the indexable temporaries and an integer variable as your stack "pointer". Recursion would have to be converted to a while-loop, which is an inconvenience, but it should be possible to implement most recursive algorithms as long as your "stack" fits in the register set.
Well right now it looks like we're at the point where we can pretty much brute force 16xAF on every surface and be ok. The jump to 32xAF will probably mean revisiting angle dependency....but I do agree that your shots demonstrate a need for higher AF.
icecold1983
Banned
32x af on 90 degree angles, 16x on rest
maybe r600 will offer higher af?
maybe r600 will offer higher af?
Have you tried forcing at least 2xSSAA on top of 16xAF? It's not like Trackmania is a performance hog is it?
It seems to me rather like just one corner case; I'd rather have (as s.o. else said in another thread) better mipmaps or a fundamentally different filtering method than anisotropic before we'd reach something like 32x samples. For corner cases like that (if you have fillrate to waste), one could always combine a portion of SSAA with AF. The unfortunate thing being that IHVs don't even allow garden variety OGSS in their drivers. I wonder how "complicated" it would be to have any sort of full screen supersampling, when your GPUs support adaptive Supersampling all along.
nao,
I hate you too
icecold1983
Banned
tm is a performance hog, at 1920 x 1200 with 4x/16x avg fps is like 38. also, i dont know how to force 2x ssaa because it doesnt seem to be enabled in the drivers.
this isnt a corner case btw, any track with a decently long straightaway shows the same problem, and its very noticeable when playing
this isnt a corner case btw, any track with a decently long straightaway shows the same problem, and its very noticeable when playing
Does anyone know if interpolation is currently done at FP32 precision? Perhaps it would be inefficient to put these onto FP32 ALUs if this is not the case. What could you do with programmable interpolators?
(RE: putting them in the shader)
Values are usually interpolated at full precision..
