[Jawed]

http://www.driverheaven.net/articles...ject/index.htm

The article is an IQ and performance comparison between X800XL and 6800GT.

I hadn't realised how serious the rendering issues are with the Nvidia cards.

One issue, the lack of water surface rendering, can be fixed by bodging the Device ID. Does that mean that the way this demo has been coded for NVidia cards prevents them from rendering the water correctly (in a similar fashion to the lighting errors on NVidia cards in Far Cry up to Patch 1.2)? Does the demo need a patch in order to function on NVidia cards?

Overall, I want to ask why is a nominally SM3 graphics card rendering this demo with such significant errors? How come years-old ATI hardware runs this fine? - none of the technology in this demo seems specific to the latest X800/X850 graphics cards.

The article suggests that NVidia has driver fixes for this already, but they haven't been released. Neither the depth of field nor the water surface seem radically different from effects seen in other DX9 games/benches, so should we presume that this boils down to (a lack of) application profiling in the NVidia driver?

I know that's a naive question, I'm just wondering what's the cause of the issues. Why does this instance of DX9 SM2 content render so poorly on an SM3 card?

The simple answer "it's an ATI-sponsored demo, dummy" is all very entertaining, but seriously, what kind of obscure DX9 tech is this demo using that's beyond the default behaviour of a 6800GT?

Jawed

[digitalwanderer]

Adam Foat in Nvidia UK PR let us know that the bugs have been fixed internally though there is no set date for that driver to be made publicly available.

That perked my interest right up!

Looks like there are some legit and weird issues, anyone want to guess as to why?

[jb]

Jawed,

NV products have had some rendering issues in farcry since day one. I remember one where the nv cards would not render the floor texture/glass correctly..it show banding in floor ties and else where. The quick way to fix that was to rename the device ID to be that of an R300. This fixed the IQ. Since FarCry was a TWIMTBP title, one can hardly say that was ATI's wrong doing. Granted this are now all fixed and IQ is the same...just pointing out that its happend before with NV/Farcry so lets not be too quick to point figures just yet

[trinibwoy]

Seems to point to Nvidia/developers trying to do something specific with NV hardware since it runs the R300 path with no issues. Weird.

[karlotta]

Or they are still criplling there 6xxx cards to get as much performance out of there 5xxx cards . "They" = Crytech/NVDA. Can we say shader replacment?

[Deathlike2]

Best idea I could come up with... is borked optimization... either by NVidia's driver team, or the developer (using multiple paths for different hardware/manufacturer in an engine can have issues... kinda wished they used multiple paths dependent on their DX version+SM paths... but whatever I guess)

[Demirug]

This is cryteks fault.

The engine selects the shader based on a chip detection. If a shaderset is missing a shader for an effect you will not see this effect or something strange instead.

In the case of "The Project" the shaders for the new effects like the water are missing in the NV4X shaderset.

The moral of this story:

Do not select effects based on chipidentity. Select based on chipfeatures.

[FUDie]

Originally Posted by Demirug

This is cryteks fault.

The engine selects the shader based on a chip detection. If a shaderset is missing a shader for an effect you will not see this effect or something strange instead.

In the case of "The Project" the shaders for the new effects like the water are missing in the NV4X shaderset.

The moral of this story:

Do not select effects based on chipidentity. Select based on chipfeatures.

How can this be Crytek's fault if NVIDIA is claiming to have fixed these issues in internal drivers? If the app were not loading a shader effect, the driver wouldn't be able to fix it no matter what.

-FUDie

[IbaneZ]

Guess i won't be getting a GT then. This demo is all i play.

[digitalwanderer]

Originally Posted by IbaneZ

Guess i won't be getting a GT then. This demo is all i play.

Ok, that's worth at least two points.

[Richard]

Originally Posted by FUDie

How can this be Crytek's fault if NVIDIA is claiming to have fixed these issues in internal drivers? If the app were not loading a shader effect, the driver wouldn't be able to fix it no matter what.

Although I think the jury is still out on this one, demirug's theory is not that far fetched. Even though ATi has continuously stated Punkbuster causes problems it still released a bugfix in the later drivers. There comes a time when regardless of whose fault is it, the IHV will "fix" it because they're the ones catching the bad light.

Anyway, I think the problem is somewhere in the middle between nVidia's optimisations gone wrong and developer releasing a IHV-sponsored demo.

[Sxotty]

Can they not capture identical frames? When doing an IQ comparison without it is kinda silly to look at when the characters are in obviously different poses.

They are just random places they decided to take pictures or what?

SOrry I don't care much except it is so sloppy why bother to have comparison shots that are not taken even from the same vantage poitn, or in the same direction.

[Demirug]

Originally Posted by FUDie

How can this be Crytek's fault if NVIDIA is claiming to have fixed these issues in internal drivers? If the app were not loading a shader effect, the driver wouldn't be able to fix it no matter what.

-FUDie

At least the driver is able to fake the caps after it detected "The Project". But there are other solutions to fix a missing shader after you have detect the application. If it is only used with one vertexbuffer you can activate the right shader at the same moment the app select this buffer.

With the right tools it is not a big thing to check why the water is not show at all on a NV4X. But until now I am not checked the rendersequenz of the ocean part from the demo in detail. In the case of demos the interest is usually only from short duration.

[karlotta]

i wonder if the new FC patch 1.31 is a fix?

[FUDie]

Originally Posted by Demirug

At least the driver is able to fake the caps after it detected "The Project".

Yeah, that's a good solution. "Hey, let's run in SM 2.0 mode for this SM 3.0 app!"

But there are other solutions to fix a missing shader after you have detect the application. If it is only used with one vertexbuffer you can activate the right shader at the same moment the app select this buffer.

I think you're way off base here.

With the right tools it is not a big thing to check why the water is not show at all on a NV4X. But until now I am not checked the rendersequenz of the ocean part from the demo in detail. In the case of demos the interest is usually only from short duration.

All you've done so far is point fingers, even against the words of NVIDIA themselves. Please let us know when you've done your homework.

-FUDie

[Jawed]

The Project appears to be an SM2 application. I can't be sure about every last effect cos I don't have an X8xx card, but the water and the depth of field both work on my 9800Pro.

Jawed

[Jawed]

Originally Posted by trinibwoy

Seems to point to Nvidia/developers trying to do something specific with NV hardware since it runs the R300 path with no issues. Weird.

The water is fixable with a Device ID change, but the article implies that the depth of field effect remains broken.

Jawed

[FUDie]

Originally Posted by Jawed

The Project appears to be an SM2 application. I can't be sure about every last effect cos I don't have an X8xx card, but the water and the depth of field both work on my 9800Pro.

FarCry can work on a 9800 Pro, too, what's your point?

-FUDie

[Jawed]

Originally Posted by FUDie

FarCry can work on a 9800 Pro, too, what's your point?

-FUDie

D'oh, depth of field and water shaders are not working on NVidia hardware. These effects have been seen before in games and benchmarks.

My point is these effects are hardly so advanced that they only run on the newest cards. Yet they're not working on NVidia cards. So where's the hitch?

I dunno where you get the SM3 stuff:

"Hey, let's run in SM 2.0 mode for this SM 3.0 app!"

in your response to Demirug. He's not suggesting that.

Jawed

[bigz]

Maybe this isn't a driver release?

Maybe the dev-tech team at NVIDIA have been working with Crytek to resolve the rendering issues?

I'm sorry, but I take a lot of what is said in DH's articles with a pinch of salt after the stuff they came out with after NV40's release, regarding the "poor" image quality that was actually better than refrast according to Microsoft.

[FUDie]

Originally Posted by Jawed

D'oh, depth of field and water shaders are not working on NVidia hardware. These effects have been seen before in games and benchmarks.

My point is these effects are hardly so advanced that they only run on the newest cards. Yet they're not working on NVidia cards. So where's the hitch?

Except that different code paths may be used for SM 3.0 parts. FarCry was updated to support these new features, remember? Perhaps these features are only problematic on NVIDIA cards because it's an NVIDIA driver problem (as mentioned previously). AFAIK, "The Project" is running on the same engine as Far Cry, which means that it may be using SM 3.0 features/shaders.

I dunno where you get the SM3 stuff:

"Hey, let's run in SM 2.0 mode for this SM 3.0 app!"

in your response to Demirug. He's not suggesting that.

Really? Maybe you should read more carefully:

Originally Posted by Demirug

At least the driver is able to fake the caps after it detected "The Project".

What caps do you think he is talking about when referring to missing shaders?

-FUDie

[CK]

Like bigz I really don't take their nVidia vs. ATI articles as being fair. It's a site started at R3D. They have personal relations with some ATI dudes, i.e. TM. A couple of them are working for/helping ATI in the Beta Catalyst program, and Zardon is one of the leading testers. I believe he has a certain title within the beta Catalyst program but I am not sure.

Say they find an ATI rendering error then they send the news to CM or others in the Catalyst beta program. If they find a nVidia error, then they make an article about where nVidia cards sux. Way too often it looks like the error they report is coming directly from ATI. There are several "nVidia can't render this/is cheating" articles, but NONE about ATi errors or cheating.

IMHO DH is a fansite (almost) like R3D - it's a ATi demo come on.

- I love the site, love their news and their forum but IMHO they are a fan site or is really close to being so.

[WaltC]

Originally Posted by digitalwanderer

Looks like there are some legit and weird issues, anyone want to guess as to why?

Please recall (as if you might forget...) that the stock answer to any and all driver anomalies the public may uncover in nV drivers is "Oh, yes, it's a bug we've known about internally for some time and we have plans to fix it soon in the future, so don't you worry."

Announcing foreknowledge of the bugs along with a proclamation that they'll be fixed at some point in the (far?) future is meant, I suppose, to create confidence in the consumer as to nV knowing what it's doing, even in the absence of any explanation as to why that foreknowledge did not prompt them to fix the problem(s) when discovered, but to go ahead and release the drivers without making any reference to these "known problems" at driver release.

And so, I've come to the conclusion that this is a stock policy A: by nVidia personnel who are Q:'ed about driver misbehavior in general. After all, who really knows what lurks in the heart of the ForceWare? In this case, I don't think even the Shadow knows...

[Jawed]

My understanding is that the Project uses a revision of the engine used in Far Cry. The engines are not interchangeable. Development work has stopped on Far Cry. This demo marks the start of a new phase of work for Crytek. (Remember this is my understanding).

While it would be reasonable to suppose that Crytek's next engine has a significant amount of SM3 capabilities planned, there's actually no reason to suppose that the Project uses any SM3 capabilities, particularly as it was built to demo SM2.0 ATI cards (SM2.0b? is there anything SM2.0b specific?).

Originally Posted by FUDie

Except that different code paths may be used for SM 3.0 parts. FarCry was updated to support these new features, remember? Perhaps these features are only problematic on NVIDIA cards because it's an NVIDIA driver problem (as mentioned previously). AFAIK, "The Project" is running on the same engine as Far Cry, which means that it may be using SM 3.0 features/shaders.

This is why I started the thread, to hear people's ideas for what's going on. Is this just a matter of Crytek coding for ATI? Or were Crytek lazy in retaining dodgy bits of device-specific code that should have been cleared out?

ATI's geometry instancing, which is supported in Far Cry (with a bit of user-fiddling), has to be capability-detected using either Device ID or that FOURCC query thingy - so Crytek is caught both ways by both ATI and NVidia tech, in trying to publish games/demos that are "bleeding edge".

Really? Maybe you should read more carefully:

What caps do you think he is talking about when referring to missing shaders?

Maybe these are ATI-specific caps? I dunno, this is why I started the thread to try to understand if we're looking at:

1. ATI's support of some subtle supra-DX9 tech, or
2. NVidia's application profiling in drivers hampering new apps, or
3. Crytek getting caught up in knots with their IHV-specific coding, or
4. ...?

I don't think Demirug is attacking. If we're talking about pure SM2 functionality in this demo (which genuinely seems to be the case) then it's curious that it's not running on SM3 hardware. I'm open to the possibility that SM3 specific code from Far Cry has been left in the Project and is somehow getting in the way.

Mind you (sorry for this Humus!) maybe it's just like some of Humus's demos, that don't work properly on NVidia until he gets some feedback, because he doesn't test on NVidia cards (not having any available to him).

If it was like that it would be purely Crytek's fault, wouldn't it? But NVidia says there are unreleased drivers that make The Project fully functional... Has an NVidia driver release ever "fixed" a Humus demo that hadn't worked on NVidia cards?

Are NVidia actually coding a shader replacement into the driver? Can anyone be bothered to test an old ATI driver, say Cat 4.1, to see if a 9800Pro can still run The Project? Perhaps ATI's sneaked some shader replacements into the last few versions of Cat just in time for The Project's release?

Jawed

[Jawed]

Originally Posted by WaltC

Please recall (as if you might forget...) that the stock answer to any and all driver anomalies the public may uncover in nV drivers is "Oh, yes, it's a bug we've known about internally for some time and we have plans to fix it soon in the future, so don't you worry."

Well this is an ATI-sponsored demo, so the chances are that NVidia's not had much of an opportunity to play until it was published.

At the same time it seems peculiar that Crytek had such a lot of trouble with the lighting in Far Cry on NVidia 6xxx cards, despite TWIMTBP (original version and Patch 1.1). Then there was the Patch 1.2 confusion which, on balance, seems to have caused more problems for ATI cards than NVidia cards.

Not all this stuff is driver problems.

Jawed