[hovz]

http://www.ixbt-labs.com/articles2/gffx/fc12.html

the bottom part is very interesting. apparently vertex instancing is completely useless. and

"FarCry 1.2 example demonstrated that Shaders 3.0 are real and effective and that many games will be capable of their support in case of a flexible engine. Obtained results are somewhat ambiguous: on the one hand, we witnessed performance gain in several places; on the other hand, it seems that the improvement, which brought considerable rendering gain (several light sources at one pass), could have been implemented with Shaders 2.x (possibly even 2.0). Thus I would like to wish the developers not only to promote the latest technologies, but also not to forget and optimize the game performance with the existing video cards. However all performance gains in the above tests are actually due to pass reduction at rendering lighting. That is the current implementation has no merits achieved via the features of Shaders 3.0. "

now every time i read some uninformed jackass rant on about shader model 3 in the current gen i will just post this

[trinibwoy]

Shader Model 3.0 r0x!!!

[Mintmaster]

So, is this TWIMTBP doing it's thing again?

Man, this sucks pretty bad. The only (non-marketing) things I can think could possibly justify this are that SM3.0 has:
1. 10 texture iterators as opposed to 2 colour and 8 texture
2. More instructions, registers, constants
3. A different way of mapping between VS and PS.

I don't see any of these factors affecting their ability to put this lighting model in SM 2.0.
1. I doubt FarCry needs that many texture iterators, but I guess it's possible if a lot of textures consume many of them. Doesn't look like it to me from any screenshot I've seen.
2. Performance wouldn't be where it's at if so many instructions were needed, nor would multipassing have had much of an impact unless geometry was a factor, which it isn't judging by resolution scaling of SM3.0 perf. More lights don't need more registers, and 32 constants is plenty.
3. Not too familiar with this, but it shouldn't make a difference.

What confuses me is that they're putting 3Dc into their engine. Why would they do this if they're coding things specifically for SM3.0 that can also be done with SM2.0, i.e. if they're in NV's pockets?

[Chalnoth]

Well, Mintmaster, all I can suggest is wait until 1.2 comes out, and use some software to extract the shaders to examine them (I believe the DX SDK comes with such software....).

[karlotta]

ill be waiting for the 1.3 patch when all the 1.2/sm3 stuff is set for sm2.x.....

[Demirug]

Originally Posted by Chalnoth

Well, Mintmaster, all I can suggest is wait until 1.2 comes out, and use some software to extract the shaders to examine them (I believe the DX SDK comes with such software....).

The SDK only have D3DSpy. It is a good if you want to debug something but bad if you only want to get the shaders.

1.2 contains all 3.0 shaders in a binary version in some files. If you extract them you will get more than 80000 pixelshaders.

But in any case my good old GFFX 5900 runs the new SM3 Path very well after some work.

[Ylandro]

Eh... Do I understand correctly that the unique features of SM3 are not giving any performance gain (vertex instancing) or not used at all (branching) and that the performance gain seen is basicly just optimized SM2 compatible code? :?

[hovz]

yep, and its only on nvidia cards. its pretty pahetic that for 2 generations in a row nvidia is hurting the games industry with its ridiculous marketing campaign because it is unable to develop hardware that can compare with atis.

[Demirug]

Originally Posted by Ylandro

Eh... Do I understand correctly that the unique features of SM3 are not giving any performance gain (vertex instancing) or not used at all (branching) and that the performance gain seen is basicly just optimized SM2 compatible code? :?

Object Instancing helps you only if you are CPU limited. Take a look at the "Regulator" results at 1024*768.

[hovz]

Originally Posted by Demirug

Object Instancing helps you only if you are CPU limited. Take a look at the "Regulator" results at 1024*768.

yes a very impressive increase of 3 or 4 fps, which was probly only cuz of the 1 pass shaders

[Johnny Rotten]

Originally Posted by hovz

yep, and its only on nvidia cards. its pretty pahetic that for 2 generations in a row nvidia is hurting the games industry with its ridiculous marketing campaign because it is unable to develop hardware that can compare with atis.

And which 2 generations would these be?

[hovz]

nv30 to 40?

[Dave Baumann]

My understanding is that the lighting model shaders could actully fit in a single pass with PS2_a/b models, and 3 lights can fit in a single pass with PS2.0, leaving a second pass for the fourth light. However, I believe there is a component that isn't used, so if pack/unpack was used then all 4 lights could fit into one pass even on PS2.0.

I'm not sure that this will be changed, but I would keep an eye on changes after DX9.0c is released.

[Johnny Rotten]

Originally Posted by hovz

nv30 to 40?

NV40. I see.

[trinibwoy]

Originally Posted by hovz

yes a very impressive increase of 3 or 4 fps, which was probly only cuz of the 1 pass shaders

Well considering that some people would declare card A the "winner" over card B if A gets 300fps and B gets 297fps 4 fps might be a whole lot :P

And in response to your previous comments I think that most agree Nvidia and ATI are on par this round performance wise. This marketing push by Nvidia for 3.0 is more about feature support than performance. One more checkbox.

[hovz]

http://www.nvnews.net/vbulletin/showthread.php?t=32221

interesting thread. some people just dont understand logic.

[hovz]

Originally Posted by trinibwoy

Well considering that some people would declare card A the "winner" over card B if A gets 300fps and B gets 297fps 4 fps might be a whole lot :P

And in response to your previous comments I think that most agree Nvidia and ATI are on par this round performance wise. This marketing push by Nvidia for 3.0 is more about feature support than performance. One more checkbox.

on par before you play in high settings or have alot of pixel shader code. than the nv40 just falls behind.

[trinibwoy]

ROFL!!!

You posted a link to a thread on another site where you tried to flame Nvidia and got flamed in return? Do you relish this sort of punishment?

Somebody close this before it devolves into similar tripe to what transpired in that nvnews thread. The thread starter's intent is quite obvious.

[Demirug]

Originally Posted by DaveBaumann

My understanding is that the lighting model shaders could actully fit in a single pass with PS2_a/b models, and 3 lights can fit in a single pass with PS2.0, leaving a second pass for the fourth light. However, I believe there is a component that isn't used, so if pack/unpack was used then all 4 lights could fit into one pass even on PS2.0.

I'm not sure that this will be changed, but I would keep an eye on changes after DX9.0c is released.

Dave, some of the pixel shaders need more than 8 float inputs from the vertexshader. This is not possible with 2.A/2.B. But you are right that it is possible with pack/unpack you can save some input register to make the shader fit in the 2.A/2.B limit.

The big problem with PS 2.0 ist the instruction limit. 64 arithmetic slots are not enough. IIRC it should be possible to fit every shader in 2 passes but many shaders will even fit in 1 pass. I am not sure if 3 lights will allways work with one pass but 2 lights should fine.

I am not sure why crytek did not implement the new lighting model for SM 2.X. I need only some hours to get it work on a GF FX. SM 2.0 is more complicated but I am believe that I close.

[hovz]

Originally Posted by trinibwoy

ROFL!!!

You posted a link to a thread on another site where you tried to flame Nvidia and got flamed in return? Do you relish this sort of punishment?

Somebody close this before it devolves into similar tripe to what transpired in that nvnews thread. The thread starter's intent is quite obvious.

it wasnt a flame tho? what is wrong with you people you guys think anything that isnt positive is a flame.

[trinibwoy]

Dave, from a development standpoint does your suggestion take an equal amount of time/effort to the SM3.0 implementation?

[ChrisRay]

Originally Posted by hovz

yes a very impressive increase of 3 or 4 fps, which was probly only cuz of the 1 pass shaders

This depends on where you are, I have used this patch extensively and have found if your roaming through the vegetation in the jungle the performance increase can be up to 10 FPS, If your running across the road in the middle the vegetation. It's more like 5 FPS,

The Instancing on the grass leaves have made a difference from my experience. But the SM 3.0 patch doesnt speed things up everywhere, Its dependent on where you are and whats being rendered.

[hovz]

did u even read the article i posted?????

if you did you would see the gains arent com,ing from the instancing, but from optimized ps2.0 code.

[ChrisRay]

Originally Posted by hovz

did u even read the article i posted?????

if you did you would see the gains arent com,ing from the instancing, but from optimized ps2.0 code.

I have used the patch myself,

But the differences between Shader 2.0/3.0 pathway in my research shows a 5-10 FPS increase in the areas I mentioned. I have had access to the patch for quite sometime now (Before NDA was lifted)

Whether you can take the information I have collected myself as true or not, Thats up to you. My research was done under real world enviroments. The Grass Instancing was paticularly highlighted as point to check, And there are real world gains from 2.0 to 3.0 where high vegetation is used.

[hovz]

can u plz at least have the respect to read the article i posted?

http://www.ixbt-labs.com/articles2/gffx/fc12.html

there is no ps 3.0 code. its all optimized 2.0. this is what the whole issue is. its perfectly usable on ati cards but they arent using it. GEE I WONDER WHY