DX6 style MT?

[Ailuros]

Maybe because K2 is outdated hardware and we aren´t really that much apart from seing something far more advanced in reasonable time?

Maybe because Doom3 is still not finished and it´s release time shouldn´t be that much apart from a future product?

I´ve abandoned the good ole K2 about a year ago and as irrelevant as it may sound, I had good reason for doing so.

 

[KimB]

But the blazing stencil performance of the Kyro series would probably put K2 way ahead of GeForce2, possibly level with GeForce3... heck, it could conceivably approach Ti4200, depending on just how complicated the scene gets. K2 would whip through the stencil passes like it's nobody's business.

But without cube maps it can't support specular highlights, and thus can't approach the quality of any of these cards.

Of course, it is a moot point now, since the K2 is outdated hardware, and there probably won't even be a DOOM3 path that supports it.

 

[darkblu]

But without cube maps it can't support specular highlights

for point and spot lights. for directional lights there's no need for light vector noramlization (per-pixel).

 

[Hyp-X]

But without cube maps it can't support specular highlights

for point and spot lights. for directional lights there's no need for light vector noramlization (per-pixel).

You need to if you use tangent space normal maps.
Even with world space normal maps you need to normalize if you have skinning because it distorts the space of the normal map.

Btw, I don't see D3 having too many directional lights...

 

[darkblu]

But without cube maps it can't support specular highlights

for point and spot lights. for directional lights there's no need for light vector noramlization (per-pixel).

You need to if you use tangent space normal maps.
Even with world space normal maps you need to normalize if you have skinning because it distorts the space of the normal map.

Btw, I don't see D3 having too many directional lights...

forgot about skinning. my bad. latter pretty much confines the normal maps to tangen-space only, too (as skinning with object-space normal maps makes little sense). of course, you're right about D3, but once you leave the basement and get into the open directional light becomes the predominant factor.

 

[Tagrineth]

I'm sure if JC wanted to he could conjure up some of his patented Voodoo Magic and make a replacement for Kyro II, probably using a creative render to texture or something like that.

That's why he said in that old .plan that he was 'considering making a no-cubemap Kyro path'... but again, the driver quality he was looking for never showed up, I guess. Oh well. I'd love to see what Kristof would have to say here, considering he's part of PowerVR DevRel...

 

[Ailuros]

I wonder do you mean that part here?

The Kyro (or specifically, the Kyro2). With lack of cubemap support, and with LightDirection being a cube map texture, would disabling per pixel normalization of LightDirection enable the Kyro2 to run DOOM3? Would you do this?

I doubt it, but if they impress me with a very high performance OpenGL implementation, I might consider it.

http://www.beyond3d.com/interviews/carmackdoom3/index.php?page=page2.inc

That doesn´t sound very promising at least to me. In the question past that he doesn´t seem interested in supporting anything lower than a GF3 anyway, but that´s something that can of course change.

 

[Tagrineth]

No, I'm really sure it was listed in a .plan update a while ago... but that is one of the things I was referring to.

 

[Sharkfood]

IMHO, it's not the programming model, it's the old cards that we have to get rid of.

I couldn't agree more.. but I also think it comes down to performance vs. application as well.

Say you are going to have a fairly large outdoor scene with rolling, lush hills of grass. In order for the scene to not look flat/uninviting, you are going to want to "frost the peaks" of the grass, so to speak.. give them some variations in color depending on altitude and curve of the hills.

You can either-
a) use traditional DX6-style MT, with about 3 layers.
b) use a rather simple shader

The problem with b) is you have just removed all the Radeon 7500 and GF-MX owners from your target sales base.. but moreover, all the GF3+/8500+ owners may see reduced performance. It gets even deeper when you take different architectures into play. What works well in one pass may require multiple passes on another platform- so you may wind up with performance between the two different approaches yielding different results based on the platform you are developing on compared to your user base.

The shader solution would also likely require less schedule time- after all, the more scenes you have, the more texture artistry you need to create to fit a particular scene and make it look realistic. You can likely totally reuse simple shaders to be applied to "like" scenes as the given effect is the same. No need to produce another 15-18 new textures per scene.

So from the way I look at it- two things need to happen:
1) More "budget" lines of DX9+ cards need to be released to help phase out older hardware. ATI and NVIDIA are both solid in this committment so we are in good shape here.
2) These "budget" lines of cards need to really improve in shader performance so as to be applicable for developers. After all, a shader that runs fairly well on a 9800 Pro or 5900 may not even be playable on a GF 5200 or Radeon 9600.