Doom3: Will it have a separate SM3 codepath ?

[darkblu]

Yes. But I've never actually used GLSL, so I haven't yet taken the time to get a test program up and running. I suppose you could either wait for me to get around to it, or somebody else

wait for you to come up with an infinite-lenght shader? ok, i think i have the time.

 

[KimB]

Um, to test it? There's a reason this is the 3D Technology & Hardware forum, not the Games forum.

 

[darkblu]

Um, to test it? There's a reason this is the 3D Technology & Hardware forum, not the Games forum.

don't get touchy, that was meant to be humorous

 

[Mintmaster]

Have I ever said this game runs well on the 6800U and X800 at 1024x768 with small amounts of AA and AF applied?

Hmm, that's not the greatest news. So my 9700 Pro, with 1/3 the fillrate of a X800XT, would be lucky to run it at 640x480? Looks like I'll be playing at 512x384 with 4xAA/16xAF Maybe it'll look okay on the TV.

I wonder how CPU dependent it's going to be? Anyone think a 2GHz AXP will make it? All that silhouette finding will sure be stressful.

Man, that is one hell of a "brute-force" based graphics engine.

 

[KimB]

A Radeon 9700 Pro should run the game decently at 1024x768 with no FSAA/AF, at the very least. It'll probably run it with some small amount of FSAA/AF okay.

Oh, and by decently I mean ~30 fps.

 

[Arun]

If it only ran on a few pixels, what is the point.

To test if the hardware is capable of it? Of course, if it wasn't, there isn't much we could do... except bitching at nV, of course! (BTW, don't call ME biased - I got a 6800GT Unless you want to call me pro-ATI, but I don't think anyone got called that in a while, eh.)

Uttar

 

[Doomtrooper]

A Radeon 9700 Pro should run the game decently at 1024x768 with no FSAA/AF, at the very least. It'll probably run it with some small amount of FSAA/AF okay.

Oh, and by decently I mean ~30 fps.

Really now, this 'preview' numbers shows otherwise. Medium Quality settings. ATI never got a chance to have some driver optimizations on this stunt either, and Terry M. stated they would improve performance from the driver version used.

http://anandtech.com/video/showdoc.aspx?i=1821&p=22

Even with FSAA, a 9800 Pro was over 40 fps, a 9700 Pro will be in the same ball park..especially if they overclocked..

 

[anaqer]

Really now, this 'preview' numbers shows otherwise.

Ain't that test run kinda... old? As in, "no longer relevant" or as in "Doom 3 will behave nothing like this"? (ok, that's an exaggeration, but still-)

 

[Doomtrooper]

Only a year old, not like the E3 leak. In fact if anything performance is probably BETTER.

 

[KimB]

Doomtrooper, did you notice my comment, "at the very least?"

Please read the context of my statement (meaning the above posts).

 

[anaqer]

Only a year old, not like the E3 leak. In fact if anything performance is probably BETTER.

A lot has happened in a year... materials system, anyone?

 

[Ostsol]

Only a year old, not like the E3 leak. In fact if anything performance is probably BETTER.

A lot has happened in a year... materials system, anyone?

There's also the elimination of the NV30 path.

 

[KimB]

I don't think that'll affect R3xx performance

 

[Ostsol]

I don't think that'll affect R3xx performance

True, but it may go toward the comparison.

 

[KimB]

Nah, since it's bound to be a major benchmark game, for the NV3x (or, at the very least, for the NV30-NV34), nVidia will likely just do some shader replacement to get some integer precision in there.

 

[Ostsol]

*shrugs* I'm betting on those cards just using the NV20 path.

 

[KimB]

Whoops.

 

[KimB]

No. The problem is they're much faster with integer precision than FP precision, not that they're much faster with fragment programs vs. register combiners.

 

[Ostsol]

So register combiners are not just emulated via the -very- decent FX12 processing power the cards have?

Well, I guess that the register combiner path would lead to more render passes. . . which could potentially slow things down a bit.

 

[Rolf N]

So register combiners are not just emulated via the -very- decent FX12 processing power the cards have?

I think reg combiners are emulated. From my performance observations (on NV36) this looks very likely, and it does make a lot of sense from a hw design POV -- why would you implement separate register combiner hardware if you can provide the same functionality with the more general stuff that you already have?