GeForceFX and displacement mapping?
- Thread starter tb
- Start date
Something doesn't add up: 125M transistors, 128-bit bus, same features as ATI, but just higher instruction counts? Surely a half-bus size would give back a bunch of transistors in the memory controller/interface, so what's those extra transistors being used for? Cache? super-scalar shader op dispatch? A per-primitive processor that is there, but disabled because of a bug in this rev of silicon, much like the 3D Volume textures in the first GF3?
Luminescent
Veteran
Maybe Nvidia implemented those extra transistors on all those high precision complex functions, such as sin/cos, etc. Nvidia may have adopted an array of programmable dsp's for the vertex processing (ala wilcat vp), which are capable of displacement mapping without the addition of extra hardware.
Wow, that's the last thing I would have expected.
I don't really get it...more major problems with implementation than I thought possible or something to schedule as a "driver implemented feature"?
*ponder*
No, I still simply don't get it, this is a DX 9 feature that seems pretty useful. I even seem to remember some comments from nVidia about deformed silhouettes that would have caused me to take the inclusion for granted.
OK, 125 million transistors...dropping displacement mapping....those feature tradeoffs as outlined in the R300 versus NV30 article are where those ~15 million transistors went?!
DemoCoder's recent guess made sense independently, but if this is true it makes a great deal more sense...
I don't really get it...more major problems with implementation than I thought possible or something to schedule as a "driver implemented feature"?
*ponder*
No, I still simply don't get it, this is a DX 9 feature that seems pretty useful. I even seem to remember some comments from nVidia about deformed silhouettes that would have caused me to take the inclusion for granted.
OK, 125 million transistors...dropping displacement mapping....those feature tradeoffs as outlined in the R300 versus NV30 article are where those ~15 million transistors went?!
DemoCoder's recent guess made sense independently, but if this is true it makes a great deal more sense...
Reverend said:
That would be rather...odd...since they entirely dropped support for those with their previous cards.
I think that what we may be seeing is a symptom of the "Programmable Primitive Processor" that didn't quite make it to the NV30. They may have attempted to remove all HOS to place it in that processor, but when it was cut from the final design, so was all HOS support.
Of course, I don't know if the "Programmable Primitive Processor" ever existed, but it does seem like a plausible explanation.
But seriously, how many games have announced support for Displacement Mapping? It's been available for a little while now...
mech said:
I believe that it is a DX9 implement... but not supporting displacement mapping doesn't disqualify the hardware from being DX9. Same thing happened with HOS support in DX8... nvidia didn't support it and still the hardware was considered to be DX8.(Of course somehow nvidias DX7 hardware also managed the DX8 label as well.... somehow..
) I am surprised there isn't a chorus of nvidia fans down-playing displacement mapping. Nvidia continues, it seems, to not support any high ordered surface. Why?
http://www.beyond3d.com/previews/nvidia/nv30gfx/index.php?p=3
According to this site, NV30 supports:
High Order Surfaces
Continuous Tessellation
Vertex Displacement Mapping
Geometry Displacement Mapping
According to this site, NV30 supports:
High Order Surfaces
Continuous Tessellation
Vertex Displacement Mapping
Geometry Displacement Mapping
Sabastian said:I believe that it is a DX9 implement... but not supporting displacement mapping doesn't disqualify the hardware from being DX9. Same thing happened with HOS support in DX8... nvidia didn't support it and still the hardware was considered to be DX8.(Of course somehow nvidias DX7 hardware also managed the DX8 label as well.... somehow..
well, GF3 was the first chip from nVidia to support all DX6 features.
yeah, it's the ghost of Bitboys hoovering around again. I am talking about DX6 BumpMapping aka. EMBM.[joke mode on]
"God damn EMBM... why Dot 3 didn't replaced it? are you sure we have to support it?? How about making PS without EMBM? is that possible? Are you sure?? sh*t... how we are going to explain this?? we never have admitted using tech invented by somewhere else... How we gonna get over difficult questions? oh... but that's great idea... yeah, right... I call to marketing right away... re-inventing it on marketing sounds pretty good. maybe we should give a picture that it doesn't work right without full Pixel Shader compability? okay... let's give the rest of planning to marketing. they know what they do... sure. bye."
[joke mode off]
EDIT: I am as well very interested to hear if NV30 really does or does not support Displacement Maps. I'll be watching this thread actively.
and Chalnoth: how about fact that Displacement Mapping is DX9 feature? Maybe it has something to do that no one supports it yet?
Or do you know if ATI or Matrox has OGL extension available implementing their DM Hardware? afaik, neither does.
