Originally posted by: Blastman
Originally posted by: Rollo
I have a bunch of SM3 games already: Painkiller, Far Cry Splinter Cell, Lego Star Wars- and the Lego Star Wars and Splinter Cell have SM3 only effects?
SM3.0 in Painkiller and Farcry????? I don?t think so. HDR is not SM3.0.
The fact of the matter is that SM3 has been the Microsoft standard for a year now, most of the big developers have been coding on it longer than that,
So, it?s been what, over a year since the 6800 came out and we have maybe 3 games that use SM3.0 ?
Pitfall Harry -- water shader
Pacific Fighters - water shader
Splinter Cell CT -- 8 shaders which exceed SM2.0 -- AFAIK.
A single water shader in 2 games and a few shaders in Splinter Cell shaders that likely don?t exceed SM2.0 (96 instructions) in length but only use more registers than the SM2.0 is capable of handling. Ya, SM3.0 is just catching on like wildfire.
The few games that have anything to do so far with SM3.0 are likely a result of payola from NV anyway.
The 6800 also chokes on the SM3.0 water shader on Pacific Fighters ?
simhq
?I believe you have PF now, so tell us all how great the GF6800 Ultra can run those water settings? Bear in mind that only recently you were proclaiming the 6800 Ultra as the dogs nads and that PF would prove it. Face it mate you fell for the same line as the rest of us, I too expected a 6800 Ultra to run water=3 and shaders 3.0 like number two's of a shovel
It's a rhetorical question, I don't need an answer as water=3 runs like crap on my system. Go anywhere near a coastline and FPS drops to 15-17, with action near a costline it drops to single figures. A case of Nvidia promising lots but delivering a lot less, there is no doubt that water=3 will need at least an SLI system or the nextgen GPU's. CPU power wont make a big difference, I get the same FPS as Takasaki with a less capable CPU.
AMD64 3200
1GB DDR400 CL2 RAM
6800 Ultra (65.73)
SM3.0 is basically about longer shaders and the X800 can handle shaders many times longer than SM2.0 (its - SM2.0b = extended SM2.0.) As shaders get over 100 instructions long (over the SM2.0 spec) there are some real question marks on how well the current cards will be able to handle these longer shaders anyway whether it?s the 6800 or X800. If we get anywhere near the SM2.0b spec the currents cards will likely be choking on the shader load anyway. The X800 can render 3 lights in a single pass (SM2.0 spec is 1) compared to SM3.0 on the 6800 which can do 4 (SM3.0 spec is 4). This rarely will make any difference, so, the X800 is close to the SM3.0 spec in several ways except for dynamic branching which probably won?t that useful on the 6800 because of it?s large performance impact.
I think it?s a good possibility when developers start using longer shaders than SM2.0, they will be ?SM2.0b? shaders anyway. Well, except where marketing $ may come into play. If a developer needs a shader longer than SM2.0 spec?s -- say a pixel shader 150-200 instructions. Then you might as well make it look like a long PS2.0 ( no branching), so it?s a SM2.0b spec shader. There is no point in adding branching to the shader to make it SM3.0 (instead of SM2.0b) when that will probably kill performance anyway. Since both the 6800?s and X800?s run SM2.0b (SM3.0 is backward compatible with SM2.0b) both current generation of NV and ATI cards are supported with the longer shaders if you use SM2.0b.
Even a SM2.0 card can effectively run longer shaders than 96 instructions (SM2.0 spec) by multipassing them. My understanding is that Splinter cell does this in Chaos Theory with SM1.1, using 6-10 passes to effectively render longer shaders than the base SM1.1 spec. The interesting thing about this is that one gets the same IQ for all graphics cards that support SM1.1 and up. The downside is that older cards like the 4200 will choke on these longer multipassed SM1.1 shaders because they weren?t designed to handle longer shaders well.
Facebook
Twitter