R520 Infomania

We saw that with Far Cry, I know...

 

Nothing to do with that was simply because NVIDIA implemented harware elements that would be useful for Doom3? (JC's shadowing mechanism was known probably before R300/NV30 development started)

 

Yes, Ati surely missed a ship when it comes to shadowing.
There are more games than Doom3 that use that shadowing technique and it will hurt Ati every time one is launched.
It's not as much as John Carmack wanting it but nVidia offering something usefull (Curved Surfaces? when are we going to see 3Dc?)

 

Don't you mean missed the boat?

What's the probability of ATI going with an HDR format that is AA compatible? Also, does FP filtering and blending performance scale linearly with available bandwidth or do other things factor in as well?

 

These programs are made for developers to make sure thier games runs well on the given hardware, not for specific opitmizations for thier cards.

Its up to the developer if they want to make it faster for ones cards or not.

The example you gave with volumetric shadows, is kind of loose, the reason volumetric shadows are used more in recent years is because GPU are able to process the extra info needed without the memory requirements of other shadow techniques. Also they provide much better scale ability and visually they look better, and are much more forgiving in 3d scenes. Many TWIMTBP games run very well on both sets of cards, actually some of them run better on ATi cards. Far Cry and Pain Killer for example.

Doom 3 is a different example all together, that had nothing due to ID's willingness to optimize for nV's cards, nV did what it always did, optimize thier drivers to Carmack's engine. This is usually all that is done in these types of dev rel program, I've never heard or seen anyone from nV, or ATi (but I have heard stories that are contrary to this specially with Far Cry and ATi, also with Half Life 2 and ATi, this was blatently noted by Gabe Newell too).

These programs were made to make sure the games run right on thier own card and others. If a great game comes out and it doesn't run well on other companies cards it hurts the game sales, and indirectly it will also hurt the graphics card company which cards the game runs on well too.

 

These programs tends to be cross marketing deals between the IHV and the publisher - not neccessarily involving the developer at all...

 

What "stories" are these then and who were they told by? Gabes comments was the other way around - they didn't need to optimize for ati, but they spent more time optimizing for nvidia.

Why say that these things don't happen then to insinuate with "heard stories" that it does?

 

IIRC somebody from ATI write large parts of the SM2 shaders for Valve and there are only 5 changes for nVidia in this shaders.

 

Thats part of the job of developer relations isn't it? Do you think that EA wrote the shaders in Tiger Woods that didn't work on ati boards because of device id detection?

 

Sure it is. But if one IHV helps you with your shadercode you should not suprised if they don't work well on chips from an other IHV.

Id detection is allways bad because you will get in trouble and EA does sometimes funny things.

 

I think it's worth remembering that a lot of HL-2's shader code would have been written in the R300/NV3x era.

By the time NV40 turned up, was Valve in a position to re-code shaders?

I think it's interesting to ponder whether Valve has held-back HDR rendering in HL-2 until ATI is in a position to match NVidia in terms of hardware support.

Though we don't know if R520 will match NV40, in supporting FP blending.

Has anyone heard anything about FP blending making it into WGF2? I'm just wondering if ATI is set to ignore FP blending because it's years away from inclusion in any M$ spec.

Jawed

 

WTF are ATi levels then?

Frist off the source engine was never optimized for nvidia cards as it was for ATi, in no way would the performance be 10 - 15% slower on nvidia's cards then ATi's if it was, sepcially since in 90% of Half Life 2's scenes geometric complexity was around 100k per scene. Way below Far Cry, and even below Doom 3.

There has been rumours ATi gave Crytek 500k to make the sm 2.0b path for the cry engine.

Why not try to write a generic shader and run it on both the 6 series and the x800 series, which one will go faster? I'm 99% sure it will be faster on the 6 line, thats what shader mark is all about isn't it? So why is Half Life 2 faster if its so shader intensive on the ATi cards? And its not that vertex shader intesive either so that should be an equilizing factor for nV cards!

 

See here is the situation, they optimized for one card and one card only, the r300. the nv paths were never there in the final release, thats what the fp16 and pp was all about which is not that big of a deal, but then you run nV gf6's cards as ATi x800's cards you get a 3-6% boost in performance, which is not substantial but there is something else going on.

 

If the developers don't relize what is happening thats just plan ignorance. And the fault is with the developer, not that the dev rel is making the shader for them.

 

Also, I think it might be rash to conclude that the shaders in HL-2 have no optimisations for ATI hardware. Merely writing DX9-targetted code in HLSL doesn't necessarily give you optimal execution on any IHV's platform. Otherwise Richard Huddy wouldn't need to get up at places like GDC and talk about the optimisations that developers should consider. Whether that's providing compiler hints in the source code or re-ordering data access paths, etc.

As far as I can tell (I don't code for GPUs or games!) there's a big gap between shader code that works and shader code that runs well, even with optimising compilers and driver re-compilation technologies to help at every turn.

I think it's safer to say that Valve was able to produce much better optimisations for their desired effects on R300 hardware than on NV3x hardware.

Jawed

 

Really? First I've heard of that. Link?

 

Dunno, what are they? I've only heard on vague comment so far, with no detail.

as far as I know they are using full precision shaders.

Never heard those before - where they they come from? How much did nvidia give them in the first place for it being a twimtbp title? plus - so what if they did - its only there for the advantage of x800, they also put in a shader 3 path for the same tiem, so its hardly as though one was being disadvantaged over the other.

From what remember, shadermark was faster if you use pp, but not full prrecision on gf6.

 

So, that applies to valve and crytek then?

 

http://www.tcmagazine.info/articles.php?action=show&id=99&perpage=1&pagenum=10

 

Well Crytek didn't optimize only for one set of cards, they did it for both ISV's thats the difference, plus they don't need ATi making shaders for them they are very competent at making thier own.