ATi shader replacement?

[Terra]

I stumbled across some info that I would like to get either confirmed or rejected.
Please don't take this as an attempt to miscredit ATi, I know that in the graphics card market there are more FUD/misinformation than during the cold war, so this is why I am bringing this up to the [H]ard so someone that has Halflife2 and a NVIDA card and would be willing to test this out, so that we may get facts into the light!

In this thread on Anandtech(I'm not registred there, so I can't post there) I read the following:

Funny how HL2 runs slower on nV40 than it does ATi cards. Even runnier how it suddenly runs faster if you use 3d-Analyser to force an ATi ID on nVidia cards... Couldn't be something to do with the shaders ATi wrote could it now?

Please observe that I am NOT accusing anyone, not the original poster and not ATI of anything, I would just like to get the facts about this, and then take it from there.
And I can't do this my self as I dont have the game HalfLife2, so I can't test this myself, otherwise I wouldn't bring this up.
Hope you guys understand

Terra - Sorry if anyone feel offended by this, but I am just trying to sort out the facts in the "GFX-war"

 

[NumbLock]

I would like to know the skinny on this as well. BTT

 

[Eva_Unit_0]

Hmmmm that IS very interesting, I must admit. Granted, it's a well-known and well-accepted fact that different video cards do indeed run different shader codes to produce the same effects. That's how nvidia got the Geforce FX to perform well...they had the drivers re-compile shaders into code that the GFFX was more comfortable with. (NV3x in particular was especially sensitive to how shaders were written) So I'm not suprised that nvidia and ati cards run different shaders...now why the ati shaders run faster than the nvidia shaders ON nvidia cards is beyond me.

 

[tranCendenZ]

I think this is bogus, the poster is likely referring to getting DX9 quality on GeForce FX 5900 cards by using 3d analyze to have the card detect as ATI, then using the FP16 patch to speed it up.

 

[Eva_Unit_0]

I think this is bogus, the poster is likely referring to getting DX9 quality on GeForce FX 5900 cards by using 3d analyze to have the card detect as ATI, then using the FP16 patch to speed it up.

yeah that is something else that had crossed my mind. The main reason why NV3x didn't perform well in DX9 HL2 is because it ran all shaders at FP32. Perhaps they're using 3d-analyze to mess with the bit depth of the shaders. NV40 also runs all shaders at FP32 (though nvidia claims it no longer takes a performance penalty for doing it) and applying an ATI device id would probably(?) cause hl2 to pass FP24 shaders (the highest bit depth that any ati card supports right now)...and who knows what else would happen there. I'd imagine that image quality would get kinda messed up with such a scenario.

 

[legcramp]

isnt this subject kinda old...I mean really old... dont we already have a thread about this lol... I think it was called Nvidia sucks as the title....try and search it lol.

 

[Sly]

Indeed, this is old news

http://forums.steampowered.com/forums/showthread.php?s=&threadid=285231

Using 3DAnalyze to speed up your game for nVidia cards could also get you banned. Atleast ATI has that covered.

 

[Elios]

yeah that is something else that had crossed my mind. The main reason why NV3x didn't perform well in DX9 HL2 is because it ran all shaders at FP32. Perhaps they're using 3d-analyze to mess with the bit depth of the shaders. NV40 also runs all shaders at FP32 (though nvidia claims it no longer takes a performance penalty for doing it) and applying an ATI device id would probably(?) cause hl2 to pass FP24 shaders (the highest bit depth that any ati card supports right now)...and who knows what else would happen there. I'd imagine that image quality would get kinda messed up with such a scenario.

FP24 not 32 if it was 32 the ATi cards would be the ones having a fit as there all FP24
the NV cards FX up are FP32 full and FP16 half presison

but yea there was a trick to force it to use FP16 and that did speed up the FX line and it would look ok for the most part though now and then you can see were it gliches

 

[Eva_Unit_0]

The FX's run source in DX8.1 mode by default (FP16 shaders). My 5950U actually performed very well in it...unless I decided to force DX9 mode lol.

 

[Jbirney]

Terra,

for awhile you could do the same trick to get slighly faster frame rates and fix IQ issues with the GFX cards in Farcry back in the day. And with Farcry being a TWIMTBP game...

I guess what I am saying is just because you change the device ID and you see some faster results or better IQ does not mean it's an evil plot to slow down the NV cards. Maybe with this change it slows down other parts of HL2? Maybe there is some IQ issues in some cases with this change?

 

[Scali]

There is a valid reason why the ATi path is faster on NVIDIA cards...
I don't recall what it was exactly, but it comes down to NVIDIA cards (well, FX series in particular, which is what this is for) not supporting a certain feature, so having to use slower workarounds. The ATi path is faster, but does not render correctly.

 

[Terra]

Terra,

for awhile you could do the same trick to get slighly faster frame rates and fix IQ issues with the GFX cards in Farcry back in the day. And with Farcry being a TWIMTBP game...

I guess what I am saying is just because you change the device ID and you see some faster results or better IQ does not mean it's an evil plot to slow down the NV cards. Maybe with this change it slows down other parts of HL2? Maybe there is some IQ issues in some cases with this change?

And

There is a valid reason why the ATi path is faster on NVIDIA cards...
I don't recall what it was exactly, but it comes down to NVIDIA cards (well, FX series in particular, which is what this is for) not supporting a certain feature, so having to use slower workarounds. The ATi path is faster, but does not render correctly.

Thank for clearing that up for me guys!

Terra - This is why I asked the [H]ard!

 

[Terra]

Indeed, this is old news

http://forums.steampowered.com/forums/showthread.php?s=&threadid=285231

Using 3DAnalyze to speed up your game for nVidia cards could also get you banned. Atleast ATI has that covered.

Now that strikes me as an incredible arrogant approach by Valve?
I mean, with what right do they justify banned people for doing something that when:

A: Dosn't give an unfair egde when playing..it's not like a wall hack ect.?

B: And what buisness is it of theirs, to dictate how the graphical rendermode on peoples LOCAL machines has to be?

Terra - Now I know for sure that I will never buy a Valve product..I don't like "dictators"!

 

[Scali]

Now that strikes me as an incredible arrogant approach by Valve?
I mean, with what right do they justify banned people for doing something that when:

A: Dosn't give an unfair egde when playing..it's not like a wall hack ect.?

B: And what buisness is it of theirs, to dictate how the graphical rendermode on peoples LOCAL machines has to be?

Terra - Now I know for sure that I will never buy a Valve product..I don't like "dictators"!

Valve doesn't specifically block 3DAnalyzer. It's just that Valve detects that 'something' is modifying 'something'. Normally that means some kind of cheat. It's impossible to figure out what exactly is being done, so they are just rigorous and block anything that is not 100% 'clean'. Which means that 3DAnalyzer happens to be blocked aswell, as a side-effect. If you want to do it right, you'll have to hex-edit the card's name in the driver binaries, I suppose.

 

[Scali]

I think just remembered what it was...
It's the user clipplanes that were not available on NVIDIA cards.
So the NVIDIA cards had to use slightly different shaders, that did the clipping per-pixel with a texkill instruction, eating up fillrate.