major differences between NV4X and G7X?

are there any? they both support SM3.0, they both can't do MSAA with certain types of HDR, it seems like they have pretty identical features/limitations, except the G7X cards are faster of course. is this accurate? is G7X capable of anything that NV4X isn't?
 
MADD functions added to primary ALU, ((Making two full MAD ALUS))
Transparency Anti Aliasing,
Effiency improvement for FP16 filtering,
ROPS supporting gamma corrected multisampling.

Those are the main improvements over the Nv4x.
 
Don't forget improved dynamic branching performance (though still not nearly as good as ATI's).
 
isn't the transparency anti-aliasing basically a driver trick? it's just forcing SSAA on alpha blended surfaces, NV4X should be able to do this fine...

the gamma corrected AA seems like the biggest difference, other than speed.
 
Belmontvedere said:
isn't the transparency anti-aliasing basically a driver trick? it's just forcing SSAA on alpha blended surfaces, NV4X should be able to do this fine...

The general consent from previous discussions was that it's implemented in the ROP's AFAICR.
 
this is from the B3D G70 article:
As far as the operation of this mode goes, its now been independently suggested to us that when an alpha texture is in use the driver detects it and then resubmits the geometry for that state multiple times, dependant on the level of AA selected. The geometry is jittered, offsetting the texture sample position, in order to provide the Anti-Aliasing when the image is resolved down to the display resolution.

has anyone got TRAA to work on Geforce 6xxx cards?
 
_xxx_ said:
The general consent from previous discussions was that it's implemented in the ROP's AFAICR.
I wouldn't be too sure about that. AFAIK the feature that enables G70 to do proper TSAA compared to NV40 is centroid calculation after the multisample mask is applied. That means there is no need for geometry jittering, because masking all but one sample together with centroid calculation means that the texture sample position corresponds to the position of that one sample.
 
Blazkowicz_ said:
What's the pattern of the supersampling in TrAA? ordered grid or rotated grid?
Transparency AA uses the same grid pattern as normal multisampling on the G7x. So rotated.
 
PeterAce said:
Also another feature that came with G70 was a blender (blend unit) per ROP.
Nope a G70 doesn't have a blender per ROP
 
DeanoC said:
Nope a G70 doesn't have a blender per ROP
I heard someone else claim this, but thanks for clearing it up.

It didn't make sense to me, because there's no reason for 16 blenders when you consider the bandwidth. NV40 was quite intelligently designed in that sense.
 
I want to know if transparency AA works under opengl. B3D's quake 4 benchmark (in the 7600GT review) is not very useful to say the least (I don't remember good old alpha vegetation and fences in doom3, and I think they would cut the shadows :))

did someone try with the obvious games? (CS 1.5 or 1.6, in de_inferno at CT spawn, there are fences, and a "ceiling of vegetation" nearby). or quake 3, for example the duel map against Hunter, near the lightning gun (or thunderbolt, whatever you call it)
yes, I care a lot about TrAA in the HL1 engine, bringing better IQ to older games is nice for an upgrade. (and I formerly used a voodoo5 with its 4x RGSS)
 
Last edited by a moderator:
Back
Top