Cat 3.6 and 3.7 render 32bit textures in 16bit under OpenGL AND D3D.

[Red Dog]

Cat 3.6 and 3.7 render 32bit textures in 16bit under OpenGL and D3D when the texture quality slider is set to high quality. 3.5 doesnt have this bug, which is why I stuck with 3.5. (CM say's its tested, fixed in 3.8)

Many people picked up on the OGL bug, but I dont think anyone noticed it affected D3d as well.

I know I didnt until Catmaker stated it in a post at rage3d's forum.

My question, has the rendering bug skewed recent benchmarks to some extent? Imo, rendering textures in 16 bit would be a nice speed advantage.

Your opinions?

CM's Post follows below this line,

Quote

CATALYST maker
ATI Guru

Registered: Oct 2002
Location: ATI Canada
Posts: 823


quote:
--------------------------------------------------------------------------------
Originally posted by Genius
You do if you're using the Catalyst 3.6 or 3.7 drivers, you just don't realize it.
--------------------------------------------------------------------------------



Genius is right. The bug exists and is on all RADEON cards. You need to have the D3D or OGL main slider set to high quality to see this.
In high quality mode it should be 32 bits but we mistakingly set it to 16 bits.

7.94 will fix this, I have verified this myself as have the beta testers

End Quote.

http://www.rage3d.com/board/showthread.php?threadid=33711048

[lunix]

I cant wait for this to be fixed. I read endlessly about how great the image quality with ATI was. When I got my 98pro and fired up some sweet GL demos my jaw hit the floor. I was like WTF? It looked horrible, coulor aliasing and banding everywhere. Alot of these demos use render to texture so the output looks really bad, like I was runing in 16bit which is much worse. Then I thought they were cheating. Eventually I figured out the problem and I am patiently waiting for the 3.8's so I can show off some cool stuff with my card without being embarassed.

[bkswaney]

I did not notice any speed gain going from the 3.5 to the newer drivers.
They will run the same. :)

As a matter of fact the 3.5's run my benchmarks faster. So it seems 32bit is faster than 16 on ati's cards. ;)

[Red Dog]

Yeah I cant wait for 3.8 either.

Many people are just happy with the performance increase afforded by 3.6 and 3.7, but Ive yet to see anyone realize the reason for the performance increase may be 16bit textures.

[The Baron]

Interesting.

I'll play with it.

[deejaya]

I can't wait to replay the CoD demo (or full game if it's out by then) with the 3.8's and actually see a sky that looks like a sky rather than what my neice can do in mspaint :D

[-=DVS=-]

No matter how many bugs catalyst have , they still work and look better then any detonator :angel2:

[lunix]

Originally posted by -=DVS=-
No matter how many bugs catalyst have , they still work and look better then any detonator :angel2:

I will repeat. I was shocked at how terrible the IQ was comapred to my GF4.

[sbp]

Originally posted by -=DVS=-
No matter how many bugs catalyst have , they still work and look better then any detonator :angel2: Careful now, your fanaticism is showing. http://home.earthlink.net/~sbp777/smilies/tongueout.gif

[Dazz]

Originally posted by Red Dog
My question, has the rendering bug skewed recent benchmarks to some extent? Imo, rendering textures in 16 bit would be a nice speed advantage.

Your opinions?
Not likely as the review would of ripped them a new one as there tends to be a bid diffrence in 16 and 32bit colour most notably is the shading which is VERY easy to see.

[lunix]

Originally posted by Red Dog
My question, has the rendering bug skewed recent benchmarks to some extent? Imo, rendering textures in 16 bit would be a nice speed advantage.

Your opinions?

It has been demonstrated previously that the R300 architecture gains no performance increase, sometimes even a decrease, by redering textures in 16 bit quality rather than 32 bit quality. Forgive my lazyness in not researching to find a link to this conclusion. It would appear that it is a honest bug.
Unlike 16bit rendering versus 32bit rendering, texturing differences are far more subtle.
What is truly impresive is the fact that there is not a army of nvidiots screaming bloody murder while splatering the front pages of hardware websites everywhere.
For a vivid example of the terrible IQ is with these drivers watch the following http://scene.org/file.php?id=191527
Sorry all, I love my Raddy but such a terrible bug cannot go un-razed. Unless your Microsoft ;)

[Sotos]

Originally posted by lunix
I will repeat. I was shocked at how terrible the IQ was comapred to my GF4.

:eek: :eek: :eek:

Are you freaking kidding me????!!!!

I had a Gf4 and the textures in 3d games was like blurry S*** compared to my Radeon 9800pro

Do you want to say how blurry and not so sharp was my Gf4 Ti 4400 with 43.xx (with digital vibrance enabled) and with 3d games textures (example C&C generals)???

Just to give a Hint again:
My Geforce 4 WAS like BIG sh**t in IQ with 3d games

I do not understand what are you talking about???
maybe your monitor is problematic or OLD or you have not installed the drivers properly etc...There is no way Gf4 have better IQ from a ATI card ask whatever you want

[CATALYST MAKER]

Originally posted by Red Dog
Cat 3.6 and 3.7 render 32bit textures in 16bit under OpenGL and D3D when the texture quality slider is set to high quality. 3.5 doesnt have this bug, which is why I stuck with 3.5. (CM say's its tested, fixed in 3.8)

Many people picked up on the OGL bug, but I dont think anyone noticed it affected D3d as well.

I know I didnt until Catmaker stated it in a post at rage3d's forum.

My question, has the rendering bug skewed recent benchmarks to some extent? Imo, rendering textures in 16 bit would be a nice speed advantage.

Your opinions?

CM's Post follows below this line,

Quote

CATALYST maker
ATI Guru

Registered: Oct 2002
Location: ATI Canada
Posts: 823


quote:
--------------------------------------------------------------------------------
Originally posted by Genius
You do if you're using the Catalyst 3.6 or 3.7 drivers, you just don't realize it.
--------------------------------------------------------------------------------



Genius is right. The bug exists and is on all RADEON cards. You need to have the OGL main slider set to high quality to see this.
In high quality mode it should be 32 bits but we mistakingly set it to 16 bits.

7.94 will fix this, I have verified this myself as have the beta testers

End Quote.

http://www.rage3d.com/board/showthread.php?threadid=33711048

Well you could just ask me direct :D

There are no performance hits on the tests we are doing so far. The only thing that seems to be impacted slightly is GLExcess which loses 2-3% (but in the margin of error), because of this.

Note these are not final results, but hey I like to share information!

Later

EDIT: NOTE I JUST NOTICED THAT THE TITLE INCLUDES D3D AND MY ORIGINAL RESPONSE INCLUDED D3D. THIS IS AN OPENGL ONLY PROBLEM

MY BUDDY OGLGUY JUST NOTICED AND LET ME KNOW I MENTIONED D3D. (which was wrong)

[The Baron]

So wait. When you have it set to high quality and it's displaying 16-bit textures rather than 32, there's no performance difference?

[bkswaney]

Originally posted by The Baron
So wait. When you have it set to high quality and it's displaying 16-bit textures rather than 32, there's no performance difference?

Not that I've noticed.

[DMA]

Originally posted by deejaya
I can't wait to replay the CoD demo (or full game if it's out by then) with the 3.8's and actually see a sky that looks like a sky rather than what my neice can do in mspaint :D

Yeah, i wan't a nicer sky in CoD too, so hopefully it will be fixed :)

(Actually i'm more excited about that game than HL-2 :p )

[Kruno]

Originally posted by The Baron
So wait. When you have it set to high quality and it's displaying 16-bit textures rather than 32, there's no performance difference?

Performance is worse. I always keep it set to just high.
Very high = 16bit textures and texture aliasing.

[NickSpolec]

So, you can get 32 bit texutres in OpenGL/DirectX if you set both the OpenGL and DirectX texture sliders to High and not Very High?

What's the overall quality difference between High and Very? Just a less aggressive LOD bias for High?

[Kruno]

Originally posted by NickSpolec
So, you can get 32 bit texutres in OpenGL/DirectX if you set both the OpenGL and DirectX texture sliders to High and not Very High?

What's the overall quality difference between High and Very? Just a less aggressive LOD bias for High?

I believe so, I always use High and never very high so I'm just taking the word from others that there is a texture problem when setting it to very high. I play at high and I don't see any banding. Set textures to 60b then set to 32bit in past I did and noticed increas 5-8fps at 32b textures from 16bit. Got to go watching stargate.
Could be just me though.

Yes, that is the LOD bias setting.

[Outkaster]

The radeon 9500 and up do internal 24fp rendering. As I understand it they actually emulate 16bit color mode. That is why there is a decrease in performance.

[Kruno]

Originally posted by Outkaster
The radeon 9500 and up do internal 24fp rendering. As I understand it they actually emulate 16bit color mode. That is why there is a decrease in performance.

We are talking about textures here, not colour. :)

[eesa]

Maybe it's just me, but for some reason I'm getting the same problem with the "high" setting in jk2. And jk2 looks horrendous in 16bit.