Which path will NV40 use in Doom3?

arrrse said:

Yes and no.

Carmack has almost certainly not been using GLSL with nVidia cards, but he is on record saying that he is committed to having a GL 2.0 support (now 1.5?) path via GLSL presumably, based on early work with 3DLabs P10 chips:

http://www.webdog.org/cgi-bin/finger.plm?id=1&time=20020627230700


It will be interesting to see if Doom3 ships with that path, and how it runs on everyone's hardware.
 
Joe DeFuria said:
It will be interesting to see if Doom3 ships with that path, and how it runs on everyone's hardware.

If the "end of May" semi-rumour happens to be true ( quite possible IMO ) then I won't bet on out-of-the-box GLSL support. A patch is more likely.
 
arrrse said:
Good spotting :oops:
I meant "no way he has been using GLSL with nv3x"
AFAIK NVidia had GLSL support hidden in their drivers for some weeks or even months now, so he indeed could have tested it on NV30 (and NV40) by now.
 
:!: Doesn't arb2 path use mrt?
Oh, no, nevermind, nv has some kind of cheat/hack thingie that lets them more or less approximate that don't they?
 
Xmas said:
arrrse said:
Good spotting :oops:
I meant "no way he has been using GLSL with nv3x"
AFAIK NVidia had GLSL support hidden in their drivers for some weeks or even months now, so he indeed could have tested it on NV30 (and NV40) by now.

Well, he could have tested it...but based on the nVidia GLSL parser results, he'd likely (I would hope) be hesitant to work with it. I would assume he's using the 3DLabs chips as his "refernce" for the GLSL path, since 3DLabs seems to more or less be doing much of the leading work on it.
 
arrrse said:
:!: Doesn't arb2 path use mrt?
Oh, no, nevermind, nv has some kind of cheat/hack thingie that lets them more or less approximate that don't they?

I'm pretty sure ARB2 uses 8888 all the way definately no float point.
 
Could I just briefly remind everyone that we're all speculating here about a game that has been, "soon to be released" for a couple of years now?

Doom3 is sort of like Half-life2 for me right now, I'm giving it a big "meh." until it's actually shipping. :(
 
digitalwanderer said:
Doom3 is sort of like Half-life2 for me right now, I'm giving it a big "meh." until it's actually shipping. :(

Think JC would let down nV with not having Doom3 ready for the actual NV40 launch?



EDIT : not to mention that up until recently, it's been "when it's done", and definitely not "soon to be released". As far as I can tell, id never, ever said anything else ( apart from the recent GDC comment of course ) which is not quite the same as the HL2 situation. Any release dates or "soon"-kinda comments have been from outside sources, and as such, just unfounded rumours.
 
arrrse said:
:!: Doesn't arb2 path use mrt?
Oh, no, nevermind, nv has some kind of cheat/hack thingie that lets them more or less approximate that don't they?
I'm not sure JC ever mentioned he would use MRT, however NV40 supports MRT and NV3x supports MET and pack/unpack, which I wouldn't call a cheat or hack.
 
Joe DeFuria said:
Well, he could have tested it...but based on the nVidia GLSL parser results, he'd likely (I would hope) be hesitant to work with it. I would assume he's using the 3DLabs chips as his "refernce" for the GLSL path, since 3DLabs seems to more or less be doing much of the leading work on it.
AFAIK there is a switch to turn the compiler into "non-extended" mode, so it reports errors on any non-GLSL compliant code instead of silently compiling it. But testing on 3DLabs parts may be a safer bet.
 
anaqer said:
Think JC would let down nV with not having Doom3 ready for the actual NV40 launch?
My gut feeling is he already has let nVidia down by missing the nV30 launch with it, but that's just a hunch.
 
I imagine that Carmack reworked the ARB2 shader(s) to reduce the number of registers used, and also uses the quality hint so the drivers can make everything FP16. If it only uses like 4 FP16 registers then it will be fairly speedy.
 
If the rumors about the nV30 path being scrapped are true I'm kind of wondering how much time he wasted special coding a path for them now for no reason. :|
 
digitalwanderer said:
If the rumors about the nV30 path being scrapped are true I'm kind of wondering how much time he wasted special coding a path for them now for no reason. :|

Enough time to be pissed about it, hopefully and realise that coding for proprietary paths = no go.
 
digitalwanderer said:
If the rumors about the nV30 path being scrapped are true I'm kind of wondering how much time he wasted special coding a path for them now for no reason. :|

Time? Wasted? I wonder how long you think these things take to code...
 
Kombatant said:
Enough time to be pissed about it, hopefully and realise that coding for proprietary paths = no go.

Proprietary paths are GOOD deal with it. If the R200 (8500->9100) did not have a different path it would not look nearly as good as it will.

Also only paths that are proprietary allow new features to be used in advance. There is nothing actually wrong with this I fail to see why everyone is all uptight about it.

And DW as was said before D3 was going to be released when done, and id has not given us a date like valve did (while valve knew they were lying through their teeth). I think D3 will be out this summer sometime before Quakecon but that is about all I believe in relation to when.

Oh and BTW wouldn't it be funny if Nvdia asked JC for his proprietary path so they could code it into their drivers and then just run substiture calls from it when arb2 calls go out :LOL:
 
Back
Top