Developers will do special codepaths for FX cards?

[tEd]

Since futuremark released the new patch for 3dmark03 and we can finally see how the nv35 really performs in the ps2.0 test which is though improved over nv30 but compared to the r300,350 still not very good, i'm wondering how developers will handle such situation in games.

Is there gonna be a special codepath for nvidia fx cards in every game that uses some kind of ps2.0 shaders? Doom3 has one , future ps2.0 games like stalker and breed which both are "how it's meant to be played" games will have special fx optimations which means for me after all the findings using lower precision for ps2.0.

Do Developers even have a choice? I don't know how much additional work developers will have because fx cards needs special coding to get decent performance in ps2.0 shaders on fx cards.

I would like to hear some thoughts from people maybe facing this problem.

 

[BRiT]

Sure, the developers have a choice. It's an ugly one though.

A) Code to the standards (PS2.0)

If they go route A, the result is consumers with GF-FX cards will say "This game sucks. It runs like crap." No developer is too keen on having this happen.

Of course they can try to use the Nvidia Dev-Rel, but there's always the chance they won't have time for their game if they're working on a larger title. It can certainly take more time. Also, they'll most likely end up with lower precision shaders (INT12 or FX16). They'll have to trade off visual image quality for speed. Depending on what they're doing with the floating-point buffers and register, the lower precision option might not be acceptable at all.

 

[assen]

Realistically, developers have ALWAYS had to write separate codepaths for different chips. This will be the situation until rapid development in the GPU area stops, and improvements start coming out more gradual and just in terms of raw speed (as in CPUs today).

 

[AndrewM]

Has anyone seen the hoops that Q3 had to jump through to be supported on certain hardware?



It's the same as now, except things are more complex, because we not just talking about pixel formats or blending modes.

 

[Dave Baumann]

Actually tEd, why do they need a special codepath? In this instance if they want to mix full and partial precision for DX9 work they can do it in one codepath but one set of hardware will be ignore it while the other won't, so it will balance somewhat.

However, We've sent a few questions slightly related to the precision and shader mess to a number of developers, so stay tuned we hope to have some more in a little while.

 

[LeStoffer]

However, We've sent a few questions slightly related to the precision and shader mess to a number of developers, so stay tuned we hope to have some more in a little while.

Good idea. I haven been wondering lately whether developers really have as much focus on rendering precision (int12 over FP16 and vs FP24) as we typical have on beyond3d.

After all nVidia already had a massive dev-rel network when they designed CineFX so one would assume that they would try to give the developers want they wanted the most. And FP power apparently wasn't high on the list... :?

 

[AlphaWolf]

However, We've sent a few questions slightly related to the precision and shader mess to a number of developers, so stay tuned we hope to have some more in a little while.

Good idea. I haven been wondering lately whether developers really have as much focus on rendering precision (int12 over FP16 and vs FP24) as we typical have on beyond3d.

After all nVidia already had a massive dev-rel network when they designed CineFX so one would assume that they would try to give the developers want they wanted the most. And FP power apparently wasn't high on the list... :?

Are you sure it wasn't just nvidia telling developers what they wanted?

 

[LeStoffer]

Are you sure it wasn't just nvidia telling developers what they wanted?

Yes. It doesn't work that way because the darn developers always want more features and performances than they can get from the IHV's silicon budget.

 

[Randell]

isnt this what nVidia want Dev's to use Cg for?