Woot ATI to have SM3.0 on next major card release :P

I find this quite dissapointing. Developers taking sides and implementing effects for a certain IHV in exchange for money/benefits is bad.

 

That's not at all what DemoCoder was saying. The point is that many effects are simply easier to generate for SM3, and they run faster to boot. It may just not be worth it to create a version for SM2 hardware.

Beyond that, it was always known that ATI would produce GPU's that support SM3 at one point, so this isn't really supporting one manufacturer at the expense of another. I mean, if people didn't think the GeForce 6x00 would be more futureproof than the ATI Xxxx cards, they were rather deluded.

 

Nope. I guess this thing all boils down to is SM3.0 needed/valuable today or in the near future or is it something that won't be around for a year or so.

 

We're talking about writing an algorithm that supports non-proprietary standard features of DX9 - ps3.0, HDR blending, that only NV supports at this time (or supports well). You want all these devs to support fallbacks for your IHV, which takes extra development work. SM3.0 is easier to program to boot.

Back when ATI was sending major payola to Valve to support SM2.0 early, they had pretty much the only real SM2.0 implementation on the market. Valve had to spend extra work support non-SM2.0 cards, or poorly performing SM2.0 implementations. HL2 might have arrived sooner if they didn't have to spend all this time trying to make SM2.0 effects run on SM1.1 (many of the shaders were ported)


The fact of the matter is, developers are feeling the pain of development costs, and the timecrunch from publishers, and they simply can't afford to support 3 shader models plus oodles of other caps bits they may or may not be present. Hence, you cannot blame them if their initial release takes the easy road?

And what's the easy, low cost, fast development road, for soft shadows? It appears SM3.0.

 

Well, of course it'll be around in a year. Even with the introduction of a new version of DirectX, which I believe is expected to happen sometime next year, SM3 will clearly be supported. In the meantime, since ATI seems poised to produce SM3 parts from top to bottom this year, and nVidia already has them, it seems clear that there will be a large enough saturation of SM3 parts for significant support.

 

The nearly 2000 shaders in HL-2 seems to be a direct side-effect of having written the game for SM2.

Obviously I don't have any idea how many of those are SM1/SM1.1 shaders.

But I do wonder if Valve had had SM3 would it have ended-up writing so many thousands of shaders? It seems to me that Valve ended-up coding shaders specific to each scene, which means in many cases, specific to each room, or space where the lighting effects are more complex than "lit by the sun".

I suspect many of the shaders were auto-generated for different combinations of lighting. But still, Valve was stuck maintaining a vast set of shaders.

Crytek seems to have fallen into a similar "trap", having written hundreds (if not thousands) of shaders for Far Cry. Again, I suspect many of them are auto-generated:

http://www.ati.com/developer/gdc/D3DTutorial08_FarCryAndDX9.pdf

page 8.

Crytek seems to be saying that for n light sources, dynamic branching is no good because there's no mechanism to parameterise n lights. Have I understood that correctly?

Jawed

 

In some ways, yes. The basic problem is that the vertex shader does some of the work. At the same time, however, dynamic branching can help. It can allow you to use just one lighting shader, multiple times if needed, and merely use dynamic branching to make that shader not spend too much time if fewer than the number of lights supported for that shader have a significant effect on the scene.

 

What about the actions ubi is taking with SC:CT? They're not even including an SM2 path despite the fact that there is rather a large base to justify one. In either case, yes writing fallbacks does require more development time, however I can't imagine it taking more then a couple days, granted depending on the effect, complexity and number of shaders which need to be coded.

ATI did nothing of the sort. They paid Valve in order to include HL2 in their product's package. Valve also gained a relationship with ATI because they didn't like how nvidia was handling their GF FX problem.

HL2 was developed with SM2 in mind because it opened up a whole new area of graphics, ATI had nothing to do with that. HL2 also was not the first game to include SM2 support, ever hear of TR AOD? Far Cry? They also spent some time writing backends so that the GF FX could run half decent.

 

Except that writing the fallbacks won't be just as simple as writing new shaders. It'll also require that you render differently.

 

Please, if you don't think that Payola bought anything else, you're seriously naive. Gabe Newell was running around with a vaporware HL2 demo showing smoke and mirrors 1 year before the game was out, pimping ATI. Wouldn't they have saved themselves embarassment had they not pimped so hard, and had they not tried to time the shipment of the game for an ATI sku?


Fallbacks only take a couple of days? You're talking about potentially hundreds of shaders and even generating new content in the art pipeline. Plus, extra QA testing and extra support after the game is shipped. This all costs money, lots of money. If your company has 40 employees working on a game, that's another $200k burned by delaying the shipdate.

Reducing the number of features that must be supported significantly reduces costs.

Do you think 3Dc should be supported, or should developers just stick with DXTC?

 

The original demo of HL2 used an R300 because it was the only card truely capable of SM2 at the time and best exhibited HL2's graphics and capabilities, Mr. Conspiracy.

I did say depending on the complexity and number of shaders didn't I? I was referring to specific shaders like HDR in Far Cry and softshadows in COR.

 

Not right now, as you need to sell the game to people who have a GeForce 2MX.

I would argue, that the easy, low cost, fast development road to most things is to just stick to your flowchart, and demand hardware that does exactly that. In all other cases, you have to use what's available.

And if you want SM3.0 "because it's the best model we have", then I would suggest that you think about the parts (no floating point textures or vertex textures) that are actually within that spec, broadly useable at the moment and a real improvement over SM2.0.

Anyway, pointing to things that might have the future potential to be better is just silly.

 

You won't miss hires normal maps on a 512MB 6800 though

 

Um and you wont miss sm3.0 on a r520 .... sooooo

However with high res + high fsaa and lots of normal maps you will still be missing out on the space savings given to you by 3dc

 

One problem with 3Dc is distribution. You either ship two sets of normal-maps or one set of high-precision normal-maps that requires conversion as part of the installation process.

 

You could ship two sets or one set and convert .

i don't think anyone would care . Takes another 5 mins to changea disc or it costs the publishers another 10c for another cd or another dvd ?

 

If you sell one million, those 10 cents just cost you $100,000.

 

Not if you raise the price by 10 cents, unless you think thats to much.

 

No, they'd want to raise the cost 10 cents and still put it on CD's.

 

Due to the concept of limit pricing, this isn't going to happen.