Exposed: [UT2004] ATI vs nVidia

[spyke]

Okay, just to start off, this isnt one of those ATI sucks threads that most everyone been doing around here lately. This is an ATI praise thread.

Okay, a about half a year ago or so I owned a Radeon 9000Pro, UT2004 performance was just playable, but everything did look good, I never had any rendering issues or anything like that. Then I get myself a crappy nVidia 5200FX, and then I notice some different things in UT2004. Later on that year I upgraded to a nVidia 6800GT, while figuring my problems were just related to the 5200FX being a crappy card, far worse then a Radeon 9000Pro.

But I was proved wrong, the problem still existed. So really, after months of bringing this up to the nVidia developers, I still get poked around and its like they don't know what this problem is or if its their problem or not. So I needed the ATI screenshots to prove them wrong. This here is proof that ATI's FGLRX has more accurate rendering then nVidia's drivers.

Lets start off with those pretty ATI screenshots that wind gave me in the other thread, thanks wind!




So as you can see, with ATI, everything looks normal, nothing strange, just looks good.

Now, this is where we can all start scratching our heads when it comes to nVidia and their drivers. Performance isnt everything as far as i'm concerned. When things don't render properly when I paid over $400 for a graphics card it gets personal!

Here are some screenshots just taken by me on nVidia.
Remember, quality is all up high as it can go, everything is turned up full.




Now, you don't even have to question as to what is wrong here.
Bottom line, nVidia's drivers arent all the hype they turn out to be just because they perform good. As far as I see it, drivers are only as good if they perform good and render things properly. Oh, I been pestering the developers, and I didnt get anywheres really. Oh well.

Score a big point for ATI for getting things right in rendering things the way they should be rendered.

[wind]

Originally Posted by spyke

As far as I see it, drivers are only as good if they perform good and render things properly.

Agreed. This whole this is kind of a double edged sword: improve quality -> lose performance, improve performance -> lose quality. Of course, sometimes things can be done better in both areas but that is likely to require changes to the hardware as well.

I think ATI is finally getting the base for their drivers in shape. Once the base stuff is right (did I hear someone mentioning word "rewriting"?), optimizing the drivers should be a whole lot easier as changes are less likely to break something. Speaking for my own experience. ;-)

Just you wait - we're gonna have the best drivers around!

[spyke]

Indeed it is a big step forward for ATI.

[WhO_KnOwS]

As always: Don't start flaming or anything. And this goes for flaming ATI and nVidia! (nobody can say I'm favouring ATI over this). And spyke, nice job figuring this out!

[skirst]

Just to be fair, I have my own applications that can prove the exact opposite. Everything rendered correctly on my nvidia parts, and total crap with ATI's. The same exact code is executed on either card, only 100% ARB standard extensions were used. This is the nature of a driver, and for that matter, software development folks. It's software just like anything else, and bound to have bugs. Bugs that only surface when the conditions are just right, or an exact sequence of events occur that cause them to rear their ugly heads. These type of bugs can be extremely difficult to track down. Trust me I've personally spent many nights pulling my hair out and cursing the world over bugs like that If nvidia is aware of this, I'm sure it's on a TODO list somewhere.

[DImGR]

some pics of UT 2004 with 9600 XT works good and with the previous driver version !

[mczak]

Originally Posted by spyke

Lets start off with those pretty ATI screenshots that wind gave me in the other thread, thanks wind!




So as you can see, with ATI, everything looks normal, nothing strange, just looks good.

Now, this is where we can all start scratching our heads when it comes to nVidia and their drivers. Performance isnt everything as far as i'm concerned. When things don't render properly when I paid over $400 for a graphics card it gets personal!

Here are some screenshots just taken by me on nVidia.
Remember, quality is all up high as it can go, everything is turned up full.




Now, you don't even have to question as to what is wrong here.

Are you sure that the ati pics are correct and nvidia ones are wrong? To me nvidia's pics look perfectly fine. Open source DRI driver for r200 produce in fact the same picture.
It might be possible ut2k4 doesn't use the exact same extensions on every card and could end up with different output (in fact, I wouldn't be surprised if r200 DRI looked different, since it lacks ATI shader extensions and other things like crossbar, though I'm surprised it looks the same as the nvidia pic in that case).
I don't say that nvidia's driver isn't at fault here, but so far I've seen no evidence that this couldn't be an application bug (i.e. it uses different extensions on different cards and produces mistakenly not the same output) or even intended behaviour (maybe ut2k4 uses ATI_fragment_shader, since it's essentially the same as PS 1.4, but the game engine might lack support for ARB_fragment_shader which would be needed if you want to do everything (and more of course) which is possible with ATI_fragment_shader - the old NV_register_combiners won't cut it, though some newer NV extensions would do it too).

[ctt]

Here're two screenshots from my GeForce FX 5900XT. Maybe the 5200 you have is just as crappy as you describe.

Screenshot 1
Screenshot 2

...seriously, though, maybe your Nvidia card (the hardware) lacks support for the extensions UT2004 uses to render this.

- chris

[DImGR]

1 more

[skirst]

Originally Posted by mczak

Are you sure that the ati pics are correct and nvidia ones are wrong? To me nvidia's pics look perfectly fine. Open source DRI driver for r200 produce in fact the same picture.
It might be possible ut2k4 doesn't use the exact same extensions on every card and could end up with different output (in fact, I wouldn't be surprised if r200 DRI looked different, since it lacks ATI shader extensions and other things like crossbar, though I'm surprised it looks the same as the nvidia pic in that case).
I don't say that nvidia's driver isn't at fault here, but so far I've seen no evidence that this couldn't be an application bug (i.e. it uses different extensions on different cards and produces mistakenly not the same output) or even intended behaviour (maybe ut2k4 uses ATI_fragment_shader, since it's essentially the same as PS 1.4, but the game engine might lack support for ARB_fragment_shader which would be needed if you want to do everything (and more of course) which is possible with ATI_fragment_shader - the old NV_register_combiners won't cut it, though some newer NV extensions would do it too).

Good call. The more I look at it, the more that explanation makes sense. Nvidia has never supported crossbar, though they have similar functionality with their NV_texture_env_combine4 extension that happens to work the same exact way crossbar does. I can see how this fact may have gotten overlooked.

[mczak]

Originally Posted by skirst

Good call. The more I look at it, the more that explanation makes sense. Nvidia has never supported crossbar, though they have similar functionality with their NV_texture_env_combine4 extension that happens to work the same exact way crossbar does. I can see how this fact may have gotten overlooked.

It is not exactly true that nvidia never supported crossbar. They STILL don't announce that extension even with GF6 series I believe, but since their OpenGL driver reports version 1.4 even for GeForce4 (and higher for newer cards), strictly speaking, they don't have to announce support for it - the driver is _required_ to support it. An application can look for ARB_texture_env_crossbar in the extension string OR just look at the OpenGL version number. No idea what UT2k4 does. Of course, if you're unlucky, you end up with software emulation - while the driver has to support every feature of a specific OpenGL version, nowhere does it say it needs to be supported in hardware...
(as a matter of fact, for instance EXT_blend_func_separate is a core feature of OpenGL 1.4, guess what happens if you use it on a GF4... this is the function you want to use to show your friends how much faster a Radeon 8500 is than a GF4 :-)).
I think you'd have to ask epic if these differences are expected. A quick grep over the ut2k4 binary shows it'll probably use ATI_texture_env_combine3, ARB_texture_env_crossbar and NV_texture_env_combine4. But nothing fragment-shader like. While combine4 and crossbar are similar, I don't think you could always easily substitute one for the other. Anyway, I've just used crossbar as an example in the previous message, it could be something completely different...

mczak

[spyke]

Originally Posted by ctt

Here're two screenshots from my GeForce FX 5900XT. Maybe the 5200 you have is just as crappy as you describe.

Screenshot 1
Screenshot 2

...seriously, though, maybe your Nvidia card (the hardware) lacks support for the extensions UT2004 uses to render this.

- chris

Hehe, you got the same problem m8. Theres a reason why I used the blue core in this example.
Notice theres no side beams or top beams trailing out of it like they are in Dimgr's screenshot? Oh and if you look closer, you will see the dots that render it. The real power core shields like in Dim's screenshot dont have those dots, and are purely transparent.
Its just harder to notice on the red core, but the result is the same as the blue on on nVidia.

And by the way, my card is a 6800GT, i'm sure it supports all the features.

[spyke]

Originally Posted by skirst

Good call. The more I look at it, the more that explanation makes sense. Nvidia has never supported crossbar, though they have similar functionality with their NV_texture_env_combine4 extension that happens to work the same exact way crossbar does. I can see how this fact may have gotten overlooked.

At one point I thought about this same fact, about extensions on one card being used, while being substituted or differently used on another card. In all, it could also be a bug in UT2004. But I figure I would bring it up here too since the people over at nvnews didnt seem to know much about it, and the developers havent said much on the matter, and icculus has complely ignored me on the issue since I posted this to his bugzilla and nothing was ever mentioned about it. I even emailed him. No reply.

[wind]

Originally Posted by DImGR

some pics of UT 2004 with 9600 XT works good and with the previous driver version !

[removed some screenshots showing framerate of 70-80]

Is it just those low-polygon maps or do you get framerates like that on all the maps? How?

[DImGR]

Originally Posted by wind

Is it just those low-polygon maps or do you get framerates like that on all the maps? How?

what polygons and [what pictures were deleted] ? Yes i get framerates like that on all type of games with the exception of onslaught games where i get around 30-40 fps

[wind]

Originally Posted by DImGR

what polygons and [what pictures were deleted] ? Yes i get framerates like that on all type of games with the exception of onslaught games where i get around 30-40 fps

"low-polygon maps" as in "maps less complicated than maps that ship with UT2004". That stuff inside the square brackets tried to inform the reader what I was ranting about. Without having to include the images in my reply, that is. A thread grows (as in renders) awfully long when people include pictures in their replies.

And thanks for your answer about the framerate.

[DImGR]

they are regular ut maps.
What fps do you get with your nvidia?

` stat fps

[spyke]

Whats really ironic about this is that UT2004 for linux was ported and tested using nVidia cards.

[spyke]

I get 80fps locked to my vsync in most normal deathmatch maps. Its when maps max out my cpu is when things get slower, since my card is far more powerful then what my cpu can dish out.

[DImGR]

i am trying to find an old 20gb hard drive so i can install Windows and benchmark my ATI there to see how it performs playing UT

[SpeedyFish]

Just so you know. I'm running a dual 2.8GHz Xeon box (533 FSB) with Radeon 9800 Pro (128MB) 8xAGP - 8.8 drivers. In order to achieve an average 40fps with UT2k4 (onslaught, pretty much any map) I have to run it with low/medium settings at 1024x768.

Performance is terrible. Much as I love it, I can't carry on playing Enemy Territory for ever ...

SF

P.S. That's as an aside to another annoyance - with my Dell Lapdog with 9700 Mobility, I can't run a DVI monitor in dual head config. The driver can only run with the VGA output when simultaneously using the Lapdog LCD. In a word, sh1te.

[Cheeseness]

I've got a 9600XT and I usually get really good performance.
A lot of people withs similar hardware in the Atari forums are talking about serious drops in frame rate (down to 10 fps and under), something I've encountered during onslaught (perhaps because I usually play with instagib on during deathmatch).

Does anybody else have this sort of problem?

[JOPS]

Originally Posted by Cheeseness

I've got a 9600XT and I usually get really good performance.
A lot of people withs similar hardware in the Atari forums are talking about serious drops in frame rate (down to 10 fps and under), something I've encountered during onslaught (perhaps because I usually play with instagib on during deathmatch).

Does anybody else have this sort of problem?

Yes, many of us have the same problem. But performance is fine on those maps with older drivers, such as FGLRX 3.2.8.

[Cheeseness]

I was never able to get the older driver working under Fedora 3, and was never able to run the game with the previous drivers under Fedora 2.

I think I may have an idea as to what is likely to be behind the poor performance with the new driver.

http://www.rage3d.com/board/showthre...6&page=3&pp=30

[spyke]

I think I might of found the problem to why these weird power cores exist.

Under nVidia's drivers, it seems theres only a maximum of 4 texture mapping units in use.

Log: 4 Texture Mapping Units found

Where as with ATI's drivers, you get this:

Log: 6 Texture Mapping Units found

Could 2 missing texture mapping units cause these weird power core renderings?

Something that needs to be asked... Why do ATI's drivers give more texture mapping units for multitexturing then nVidia's? I cannot possibly believe that these cards only have 4 TMU's...

[rinthos]

to Spyke:

Try changing the ini file to 4 texture mapping units instead of the higher default value.

Compare quality.

For some reason from what i've seen, Nvidia's look a whole lot better with 4 than when you force 4 with ati cards....
---

[skirst]

Originally Posted by spyke

I think I might of found the problem to why these weird power cores exist.

Under nVidia's drivers, it seems theres only a maximum of 4 texture mapping units in use.

Log: 4 Texture Mapping Units found

Where as with ATI's drivers, you get this:

Log: 6 Texture Mapping Units found

Could 2 missing texture mapping units cause these weird power core renderings?

Something that needs to be asked... Why do ATI's drivers give more texture mapping units for multitexturing then nVidia's? I cannot possibly believe that these cards only have 4 TMU's...

No nvidia cards support more than 4 texture units to standard fixed function multi-texturing. You can use up to 16 textures per rendering pass on NV30 hardware and up, but only in fragment programs (NV_fragment_program*, ARB_fragment_program, GLSL, etc...). This could be an explanation to the rendering differences though.

[spyke]

Old topic, but still not fixed... Indeed you are right though.

NVIDIA just needs to add 2 more texturing units to their fixed function pipeline.
Kinda sucks they havent made an effort to yet.

[skirst]

Originally Posted by spyke

Old topic, but still not fixed... Indeed you are right though.

NVIDIA just needs to add 2 more texturing units to their fixed function pipeline.
Kinda sucks they havent made an effort to yet.

Honestly spyke, I don't think they are going to. Nor should they IMO. This is more of a content problem. The 4 TMU limit for fixed-function texturing on nvidia HW in OpenGL has been known since the NV30 came out. Epic of all people knew this before the general public (members of nvidia's "The way it's meant to be played" program) . If epic wanted the power cores to look the same on all hardware, they should have found a way to multipass the extra textures in for nvidia on opengl. Maybe they tried and failed, so what you see is what you get. It's nothing new. This has always been the case for games supporting a broad spectrum of hardware. You either cater to the lowest common denomenator, or you give what you can where you can to appease the people with the HW capable of displaying it. It's unlikely that this 'problem' (if you can call it that) is going to go away. On the bright side, nvidia is the only IHV with linux drivers with EXT_framebuffer_object support, and Ryan Gordon is working on getting us all the render-to-texture effects in UT2K4, like real shadows I'll take shadows over power cores any day

[UDHA]

without the purdy power cores what will cast said shadows?

I'm just shitstiring I don't even own this game.