Something wrong with 9700?

[Ailuros]

I don't think there's actually anything wrong in terms of IQ on the 9700 and I can't be sure what exact settings have been used; that's why I never let any screenshots of a card out until I'm 100% certain that I'm affiliated with it.

Doomtrooper,

While the framecounter can get influenced with screenshot capturing the average score usually doesn't. Let's not forget that UT2003 is still an unreleased game and I preserve judgement for it's real behaviour until release (ie performance should tune up by a lot across vga's), as you pointed out.

Don't forget its only a XP 1800 on a DDR 266a board, so a KT333 with PC3200 ram and a Athlon XP 2600 would help alot...

With 4xAA and 16x aniso? Maybe....

Overall there isn't another sollution out there that could be considered playable with those settings in UT2003 at the moment (in all fairness).

 

[Doomtrooper]

True but looking at these charts (as much as I hate too but Anand is the only one with the newest build)

http://www.anandtech.com/video/showdoc.html?i=1686&p=9

It shows the 9700 scoring 60-105 fps with the same settings...of course he is using a faster platform too.

I never posted those pictures for the frame counter, and I'm only stating what he stated..

For some reason when i hit the screenshot button my FPS dips by 5-10...prolly 'cause I'm takin' a screenshot

 

[Doomtrooper]

Please put the pics in a link, and not inline.

All those big pics screw up the forum formatting and make it a bitch to read.

But, to comment on this thread--its this guy's opinion, and he seems to offer reasoning for it. If you disagree, more power to ya, but you don't need to resort to name calling (e.g. calling it FUD). Polarizing comments like that will only create a flashpoint for YAST (yet another stupid thread)

Name calling ??

1)It was meant as sarcasm towards Opengl_guys post since he got labelled for posting FUD in another thread
2)Even if it was towards the poster, its not name calling...

As for the PIC's...I will make the change to make our resident chip expert happy.

 

[nAo]

As for the PIC's...I will make the change to make our resident chip expert happy.

And a lot of people will be happy too. You don't know how many ways there are to make us happy

 

[rubank]

OK, it looks like some of you take it as a personal insult that I have the nerve to question the IQ of the 9700. Sorry, but I have no cure fo you.

I referred to Daves pics as â€for startersâ€, there is another thread where someone (misae?) posted screenshots that look considerably worse even, and looking at the pics in iXBT:s review doesn´t really help, does it?

If you read my first post you´ll see that I didn´t make any claim that there´s something wrong with 9700, I put a question to you based on the looks of what I consider to be lacking IQ: software or hardware problems?
If you can´t see the anomalies that I point to, or if you maintain they are only limitations of the software then there is not much we can debate, but somehow I find it hard to beleive (that you can´t see my points). Can´t you really see the blurry spots and ugly shadows in Nascar and GP, can´t you see the â€funny†look of â€Davids†car, don´t you see the blocky shadows in the outdoor winter scene and so on? I´m amazed.

There surely must be some useful game out there, with stunningly sharp textures and shadows that don´t look like they´ve been cut out of black cardboard and glued to the scene.
After all, where I live (Sweden) this is a $550 card and I think I have the right to expect something better than what can be had with a $50 card in terms of visuals.
If it´s all due to software limitations, this business is in trouble.

 

[jvd]

I'm really happy with my card , so far the only game that looks bad is nerver winter nights as you can see in my pics , like i said in the other thread send me anything you want and i will take pics of it for you at any setting

 

[DemoCoder]

This is the problem with the idea of $500 cards improving the IQ of "older games". I mean, CounterStrike with 4XFSAA and anisotropic filtering looks nice, but removing aliasing artifacts is only one part of IQ. Pixar wouldn't be happy if their movies were just free of artifacts. Part of their image quality comes from their artwork, and the other comes from the enhanced rendering techniques they are using. For years on these B3D, people have been laboring under this idea that buying new cards that "make all your old games play better!" is a great excuse for high end cards. If aliasing is what turns you off the most, then maybe. It comes down to personal taste. For me, it's the geometry, lighting, and texture detail.

I hate blocky geometry, little dynamic lighting and shadow, and blurry textures. One of the reasons I love DoA3 on the X-Box, or Metal Gear Solid 2, is the superb attention paid to character model detail, or the hyper texture detail of Halo.


Basically, if you're buying a $500 card to play counter-strike of quake3, I basically think you're wasting your time, since previous generation cards like a Gf3 or Radeon85000 will do just fine. Now, buying a $500 card to enhance UT2003 and Doom3, is a better buy.


Now on to what I *think* Rubank was talking about. I've attached an image highlighting one of his complaints. Dave, can you retake this with a GeForce3/4? I'm curious as to whether or not this is just a bad low-res hull being used for the shadow projection, or whether it's something to do with R300 specific features.

I'm amazed most of the IQ afficianados can't spot this.

If it is not a flaw in Nascar's rendering engine (The Most Likely Scenario!, piss poor boxy car models, or low res ones used for shadowing), then the only possible R300 specific interaction I could think of is TruForm. Are these cards low-polygon models that have been "truformed" into high-res models, but the shadow extrusions aren't effected?

The result sure does look ugly. Extremely sharp shadows that almost look like a quad painted on the road. The silhoutette of the car and trunk look like they taper smoothly.

There appears to be another problem with the bumper/trunk area I highlighted. The car models on the back look wacked out, or stretched in the bumper area.


Dungeon Siege shows other problems. It looks like the shadows were down with shadow maps and Z-buffer precision issues are causing "blocking" artifacts in the shadows.


In any case, Rubank, I think you need to demand better game software, not better hardware.

 

[DemoCoder]

BTW, compare the look of Nascar to *any* recent console racer on Xbox/PS2/GC. It looks like complete crap. Crap backgrounds, crap models, crap textures, crap lighting. What ever happened to need-for-speed on the PC? Is it still being enhanced?

 

[Saem]

Those blocking articifacts I believe are due to the rendering engine if I'm not mistaken.

 

[jvd]

http://www.beyond3d.com/forum/viewtopic.php?t=2215&start=40

go there if you want to see whats going on with my neverwinter night shots.

 

[Galilee]

â€￾for startersâ€￾, there is another thread where someone (misae?) posted screenshots that look considerably worse even,

If you are talking about the picture from Grand Prix 4 they were actually taken on a GF4, and he had problems with the imagequality.

 

[DemoCoder]

BTW,
I found a shot of Nascar 2003 on the PS2. Soft shadows and maybe penumbras

Also, Nascar 2002 on the X-Box looks like it has blocky shadows like the R300 shot, however, they are not "sharp" but look like they are blended at the edges.

 

[Nappe1]

BTW, compare the look of Nascar to *any* recent console racer on Xbox/PS2/GC. It looks like complete crap. Crap backgrounds, crap models, crap textures, crap lighting. What ever happened to need-for-speed on the PC? Is it still being enhanced?

Need For Speed 6 Hot Pursuit 2 for PC is going to hit the streets on september. (if I recal right. nope, I didn't... now the release date for all platforms is 24th of October. So little bit more waiting I think. )

screenshots definately look good. Depth of Field, Motion Blur and perpixel lightning with real time cube (??) enviroment mapping at last.

check out the www.needforspeed.com for official info.

I have been editing NFS games since 1998 and at least now we are getting brand new engine. (every game from NFS3 to NFS PU used same basic engine. NFS PU had a lots of tweaks but the same basic engine is same. even NFS2 SE had some form of rash driver based engine.)

and my own website?? well, check the http://rp-design.totalnfs.net

this was bit off topic, but hopefully moderators understand.

EDIT: added real known release date.

 

[rubank]

...If you are talking about the picture from Grand Prix 4 they were actually taken on a GF4, and he had problems with the imagequality.

No Galilee, I was referring to misae´s pics in jvd´s thread.

Lasse, thx for the info. I´m looking forward to NFSHP2.

 

[jvd]

rubank - they are my pics that he is posting it seems to be a bug in that game with 6fsaa cause every other setting looks sweet

 

[hax]

After all, where I live (Sweden) this is a $550 card and I think I have the right to expect something better than what can be had with a $50 card in terms of visuals.
If it´s all due to software limitations, this business is in trouble.

Yes, you do. However, if you are making unsubstantiated claims regarding IQ based on what you think the image is supposed to look like, then the argument is faulty. You claimed that the images on a Radeon32DDR looked better than a 9700. You then commented on some screenshots from Dave. Like I had stated, please post some pics from your Radeon32DDR. If you did, you would clearly see that some of your arguments would easily be proven to be the fault of the application. If you are arguing based on hypotheticals, please realize that the images generated are based on instructions from the application. Even if one were to have some holographic imaging hardware from Star Trek, there are limitations based on what is fed in. In other words, Tetris won't look better just because it's running on better hardware.

Rest assured, if there are driver bugs, they will be delt with accordingly. A report of findings wouldn't hurt. If your arguments are based on some tangible data, then you may have found a bug and it will be dealt with, and with kudos to you. The only question is, where is your tangible data for the comments on Dave's pics?

Regarding jvd's pics, they do look a bit odd to me as well, and will be investigated.

 

[demalion]

...

Now on to what I *think* Rubank was talking about. I've attached an image highlighting one of his complaints. Dave, can you retake this with a GeForce3/4? I'm curious as to whether or not this is just a bad low-res hull being used for the shadow projection, or whether it's something to do with R300 specific features.

I'm amazed most of the IQ afficianados can't spot this.

As Dave mentioned, we need a reference. Looking at those shots, I had assumed the issues you highlighted were game engine issues, as I wouldn't expect a graphics card to have anything to do with how blocky the shadows are when the edges are that sharp. I went looking for some screen shots and found a few at the Papyrus web site, though they are pretty low resolution. There are shadows, but they are minimized in all the shots I saw. As far as I can make out, they are identical in the blockiness and sharpness to these screenshots.

Your link didn't work (as an img tag, atleast), btw. Here is the page with some images, with the most evident shadows I noticed being in the 2nd row, in the middle:

http://www.sierra.com/games/nascar2002/mediascreenshots.html

If it is not a flaw in Nascar's rendering engine (The Most Likely Scenario!, piss poor boxy car models, or low res ones used for shadowing), then the only possible R300 specific interaction I could think of is TruForm. Are these cards low-polygon models that have been "truformed" into high-res models, but the shadow extrusions aren't effected?

The problem is with the "Most Likely Scenario" being dismissed out of hand, and a thread being started and called "Something wrong with the 9700?" without any frame of reference provided. Not a high crime, , but sensationalist and I think warranting the criticism for the way it was introduced. "every screenshot look bad!" is a quote from the starting post, I believe, and the way it reads it is like rubank is comparing some arbitrary sharp texture rendering in some game he has in mind to some arbitrary impression of the screenshots carried over in what seems a bit of hyperbole.

The result sure does look ugly. Extremely sharp shadows that almost look like a quad painted on the road. The silhoutette of the car and trunk look like they taper smoothly.

The dispute was how this was a 9700 problem, and not a game problem. I haven't found a set of good shadows in any Nascar screenshots in my searching yet (except for the REALLY good shadows in the screenshot you provided of that PS2 game, which is hardly applicable I think).

There appears to be another problem with the bumper/trunk area I highlighted. The car models on the back look wacked out, or stretched in the bumper area.

I see exactly what you mean, but it feels "expected" to me. I presumed it was an attempt to simulate the lens effects of actual Nascar cameras, because I have the impression that is how it is supposed to look. Couldn't find a similar screenshot, though...just the recollection of other screenshots looking exactly like that (say, probably from some other revies from Dave...should be easy to find on this site, maybe I'll go look *duh* should have done that for the screenshots above). I fail to see how it would be a graphics card problem, however.

Dungeon Siege shows other problems. It looks like the shadows were down with shadow maps and Z-buffer precision issues are causing "blocking" artifacts in the shadows.

Heh, I recognize the problem exactly. Dungeon Siege has a recognition mechanism for setting up it settings. You can hack around it, and I have myself...for instance, it decided I needed low detailed textures and shadows, and my system (my Radeon 32 DDR) at the time of installation could handle higher ("64MB" comes to mind when I think of setting the shadows...I think I copied and pasted from that section for the shadow settings or something...too lazy to go check). It failed to recognize the capabilities of the 9700 here...that is exactly how the shadows look when the settings are lower (it isn't too hard to hack around the issue...and there are even higher setting for things like the shadows than the autoconfiguring will set, and I'm pretty sure the R300 should handle easily).

In any case, Rubank, I think you need to demand better game software, not better hardware.

That's all everyone was saying I think.

EDIT: oh, your images work now...strange.

 

[Dave Baumann]

It failed to recognize the capabilities of the 9700 here...that is exactly how the shadows look when the settings are lower (it isn't too hard to hack around the issue...and there are even higher setting for things like the shadows than the autoconfiguring will set, and I'm pretty sure the R300 should handle easily).

I'm using an updated confiruation file with the 9700 in it.

(Why the hell developers use a config file in the first place is beyond me though - isn't this what DXCaps are for???)

 

[demalion]

It failed to recognize the capabilities of the 9700 here...that is exactly how the shadows look when the settings are lower (it isn't too hard to hack around the issue...and there are even higher setting for things like the shadows than the autoconfiguring will set, and I'm pretty sure the R300 should handle easily).

I'm using an updated confiruation file with the 9700 in it.

(Why the hell developers use a config file in the first place is beyond me though - isn't this what DXCaps are for???)

OK, I'll actually go BROWSE into my Dungeon Siege directory and OPEN the file and READ my setting...the things you ask of me, I swear...

What are your "shadow_tex_size" and "detail" in your system_detail.gas file under the heading you think it is using for the R300? I'd expect it bo be "[128M32B]". For my system, I'm pretty sure I have "detail=4" and "shadow_tex_size=256" and I'm sure the shadows look much better than that. Depending on whether it is capped, and whether a shadow_tex_size increase for the shadows is data that has transferred over the AGP bus (I'd expect it is calculated by the CPU and would be), maybe even 512 would work for that.

EDIT: By "For my system, I'm pretty sure" I mean that I'm pretty sure the heading where I got these value I'm reading were being used. It also appears that a detail of "5" is supported (and I'm going ahead and moving it to the 64M32B since I don't think I've edited this file since upgrading to my 8500).

 

[Dave Baumann]

There doesn't appear to be a 128MB profile. Here's the config for the memory sizes:

[system_detail]
{
	[8M32B]
	{
		detail			= 0;
		shadow_tex_size	= 64;
		[resolutions]
		{
			[640x480]	{ max_back_buffers = 1; }
			[800x600]	{ max_back_buffers = 1; }
		}
	}
	[8M16B]
	{
		detail			= 1;
		shadow_tex_size	= 64;
		[resolutions]
		{
			[640x480]	{ max_back_buffers = 1; }
			[800x600]	{ max_back_buffers = 1; }
		}
	}
	[12M32B]
	{
		detail			= 2;
		shadow_tex_size	= 64;
		[resolutions]
		{
			[640x480]	{ max_back_buffers = 2; }
			[800x600]	{ max_back_buffers = 1; }
			[1024x768]	{ max_back_buffers = 1; }
		}
	}
	[12M16B]
	{
		detail			= 3;
		shadow_tex_size	= 64;
		[resolutions]
		{
			[640x480]	{ max_back_buffers = 2; }
			[800x600]	{ max_back_buffers = 2; }
			[1024x768]	{ max_back_buffers = 1; }
		}
	}
	[16M32B]
	{
		detail			= 2;
		shadow_tex_size	= 64;
		[resolutions]
		{
			[640x480]	{ max_back_buffers = 2; }
			[800x600]	{ max_back_buffers = 2; }
			[1024x768]	{ max_back_buffers = 1; }
		}
	}
	[16M16B]
	{
		detail			= 3;
		shadow_tex_size	= 64;
		[resolutions]
		{
			[640x480]	{ max_back_buffers = 2; }
			[800x600]	{ max_back_buffers = 2; }
			[1024x768]	{ max_back_buffers = 2; }
		}
	}
	[32M32B]
	{
		detail			= 2;
		shadow_tex_size	= 64;
		[resolutions]
		{
			[640x480]	{ max_back_buffers = 2; }
			[800x600]	{ max_back_buffers = 2; }
			[1024x768]	{ max_back_buffers = 2; }
		}
	}
	[32M16B]
	{
		detail			= 4;
		shadow_tex_size	= 64;
		[resolutions]
		{
			[640x480]	{ max_back_buffers = 2; }
			[800x600]	{ max_back_buffers = 2; }
			[1024x768]	{ max_back_buffers = 2; }
		}
	}
	[64M32B]
	{
		detail			= 4;
		shadow_tex_size	= 64;
		[resolutions]
		{
			[640x480]	{ max_back_buffers = 2; }
			[800x600]	{ max_back_buffers = 2; }
			[1024x768]	{ max_back_buffers = 2; }
		}
	}
	[64M16B]
	{
		detail			= 5;
		shadow_tex_size	= 64;
		[resolutions]
		{
			[640x480]	{ max_back_buffers = 2; }
			[800x600]	{ max_back_buffers = 2; }
			[1024x768]	{ max_back_buffers = 2; }
		}
	}
}

Heres the default ATI GFX settings:

	[vendor*]
	{
		vendor						= "ATI";
		
		b no_color_control			= false;
		b simple_render				= true;
		b no_complex_shadows		= true;
		b no_flip					= false;
		b below_min_spec			= true;
		b only_16_bit				= false;
		b full_tex_restore			= false;
		b trilinear_filt			= false;
		b shadow_render_target		= false;
		b buffers_frames			= false;
		b scene_fade				= true;
		b tex_state_reset			= false;
		b manual_mouse_copy			= false;
		b modulate_only				= false;

Here's the 9700 specific settings (and 8500 for reference):

		[device*]
		{
			name					= "Radeon 8500";
			x vendorid				= 0x1002;
			x deviceid				= 0x514f;
			x driver_product		= 0;
			x driver_version		= 0;
			b simple_render			= false;
			b no_complex_shadows	= false;
			b below_min_spec		= false;
			b trilinear_filt		= true;
			b tex_state_reset		= true;
			b shadow_render_target	= true;
		}
		[device*]
		{
			name					= "Radeon 9700";
			x vendorid				= 0x1002;
			x deviceid				= 0x4e44;
			x driver_product		= 0;
			x driver_version		= 0;
			b simple_render			= false;
			b no_complex_shadows	= false;
			b below_min_spec		= false;
			b trilinear_filt		= true;
			b tex_state_reset		= false;
			b shadow_render_target	= true;
		}

I hadn't noticed the 'tex_state_reset' difference before; it doesn't appear to be something that alters IQ (and along with 9700, 9000 has it set to falese as do all NV boards).