AM3 COMPARISON @ DH - lowered FX IQ confirmed

[Ostsol]

Apparently you didn't read my post. The fact AM3 does use PS 2.0 so it qualifies as DX9 if perhaps only minimally. Thus, it is "'possible'" that it could recieve a performance boost.

Same case as with 3dMark03's GT4, isn't it? At the very least, both are better qualified as DX9 than Gun Metal. . .

 

[vb]

has anybody checked if det's detect when a screenshot is taken?

if aquamark can take a screensot every 10 frames performance delta should be higher than with no screensots taken

my 2 cents

 

[eSa]

It would be rather.... hmmm... arrogant from the nVIdia to do the "screen capture polishing" with latest 50. series Detonators. Especially if they have been caught lately by Valve. But you never know...

While I'm writing I could just as same voice my opinion about the whole IQ thing. In a sense nVIdia has been really lucky that they are in GFX business. At the moment they can still get away with inferior PS 2.0 performance by falling back to the 1.4 shaders etc. Because the real-time shaders are relativly short and equations and data still quite low with the accurancy, they get away with it. Now imagine Intel doing the SPEC_FP tests with the lower accurancy/fixed point and this way claiming better performance than rivals.

"Our fixed point math is just as good as standard IEEE floating point".

 

[Veridian3]

I also raised the issue of screenshots with massive when i notified them the article was up. I asked that they allow the user to specify any single frame to capture ans it was easy to optimise for the frames already specified within aquamark.

From a personal standpoint i have to say that i noticed no difference from the general image quality to that of the captured screenshots. I havent however had a long time to look into it.

The other issue i raised with them was the opertunity for the benchmark to be paused and the camera moved off track (for obvious reasons). If i get anything back on either suggestion i'll let you all know.

Stu

 

[engall]

I've about this on the guru3d.com board. Someone has concluded that the whole Driverheaven and Det51 issue is due to a reduction from fp32 to fp16. Hence, the lighting disappears because it's some sort of satuation effect that can't be reproduced by fp16, and also there is the color banding. I don't think a real conclusion about what's going in Det51 can be made at this moment, though so far it looks like its only PS precision dropping, which may be used to increase performance it might be a bug since the Det51 may have something weird like an adaptive "optimizer," which may or may not be legit.

Engall, you can't treat the numbers like that. UT2003 is a DX8 game while AM3 is DX9. A performance increase is "possible."

What about TombRaider arkness of Angel and Half life2 test?
The poor performance on Dets45.23
If AM3 Dx9,why FX on Dets is so poor in these real Dx9 ---------------gamesTombRaider arkness of Angel and Half life2 ?

 

[Evildeus]

What about TombRaider arkness of Angel and Half life2 test?
The poor performance on Dets45.23
If AM3 Dx9,why FX on Dets is so poor in these real Dx9 ---------------gamesTombRaider arkness of Angel and Half life2 ?

If i was a Nv employee, i would say: yes, but the work on det 4* have been closed for a long time, and the det 5* are those which boost DX9 performance. Have you seen the performance of GFFX with det 5* on those games?

Of course, i'm not an Nv employee, and even if det 5* boost the performance, it really seems to be at the cost of quality so...

 

[g__day]

AM3 is built like a DX9 game engine. It targets a wide base of cards and assigns shaders it thinks your card can handle based upon its reported DX 7, 8 and 9 capabilities. But the game doesn't code thins as DX9 shaders for the fun of it - for speed and compatability across the widest possible user base the developers provide effects using the lowest possible level of DX APi they can provide. So 70% of the game is goiing to be DX 8 shaders.

This is confirmed by Massive today to me here

http://arc.aquamark3.com/forum/showthread.php?s=&threadid=118

If you only develop your drivers for the top end of cards functions both your top end and your middle and lower end cards all suffer, so this repsonse we were concentrating on the top end (for a universal driver) is not a brilliant response.

 

[991060]

Is there any functionality that can give us the reference pictures,such as that of 3dmark03?

the pictures from det.51.75 are different from radeon's pictures on Cat 3.7,but that doesn't gaurantee the radeon pictures are 100% identical to the supposed ones.

I mean radeon could follow the path 99% while it's only 70% on FX,but we want that 1% from radeon be pointed out.

 

[engall]

What about TombRaider arkness of Angel and Half life2 test?
The poor performance on Dets45.23
If AM3 Dx9,why FX on Dets is so poor in these real Dx9 ---------------gamesTombRaider arkness of Angel and Half life2 ?

If i was a Nv employee, i would say: yes, but the work on det 4* have been closed for a long time, and the det 5* are those which boost DX9 performance. Have you seen the performance of GFFX with det 5* on those games?

Of course, i'm not an Nv employee, and even if det 5* boost the performance, it really seems to be at the cost of quality so...

No, I dont understand.

45.23 on Am3 is so good and even Dets 5* cant help a lot.
But TombRaider arkness of Angel and Half life2 also tested on Dets 45.23.The result showed FX family so weak.
I bet that you cant explain it.

 

[mane]

the pictures from det.51.75 are different from radeon's pictures on Cat 3.7,but that doesn't gaurantee the radeon pictures are 100% identical to the supposed ones.



I don't know what the picture is supposed to look like, but i do know that Cat and det 4403 draw it the same. Actually with 4403 image is sharper but otherwise nearly identical. (lighting etc.)

You can get pictures with different drivers from
http://www.driverheaven.net/articles/aquamark3/index3.htm

Big downloads but you can definetly see the differences
45.23 are missing shadows etc etc

 

[Florin]

It's not too dramatic, but if you for instance look at the upper right part of the picture, there's this wing or tail thing standing up straight, in the 45.23 picture you can make out the yellow/black stripe pattern, in the 51.75 pic it's a blur..

Your right I do see it Thanks now I don't feel left out I can harangue Nvidia with the best

In all fairness, if you look at Zardon's flash animation in the actual article on DriverHeaven's site, it's not just the 51.75 shots that look worse. The ATI screenshots also show blur instead of the yellow/black striping from the 45.23 screenshot.

 

[mboeller]

What about TombRaider arkness of Angel and Half life2 test?
The poor performance on Dets45.23
If AM3 Dx9,why FX on Dets is so poor in these real Dx9 ---------------gamesTombRaider arkness of Angel and Half life2 ?

c't tested TRAoD with the new Deto's 51.75 already. They got an whooping 5% performance increase. So now the FX-cards are fast enough to play the game.

Link (german): http://www.heise.de/newsticker/data/ad-13.09.03-004/

Messungen im c´t-Labor bestätigen zudem, dass sich die Tomb-Raider-Leistung mit dem neuen Treiber nur um rund 5 Prozent verbessert.

 

[Ostsol]

It's not too dramatic, but if you for instance look at the upper right part of the picture, there's this wing or tail thing standing up straight, in the 45.23 picture you can make out the yellow/black stripe pattern, in the 51.75 pic it's a blur..

Your right I do see it Thanks now I don't feel left out I can harangue Nvidia with the best

In all fairness, if you look at Zardon's flash animation in the actual article on DriverHeaven's site, it's not just the 51.75 shots that look worse. The ATI screenshots also show blur instead of the yellow/black striping from the 45.23 screenshot.

That could be just a difference in default LOD settings in the drivers, you know. . .

 

[Josiah]

it's been said before, but if I have to flip through screenshots 100 times to notice a difference then I don't really care. some of these IQ comparisons are just crazy, "look at the small rock to the right of the crosshairs in frame 1679342, the 3432.535435b drivers produce a funny smell when compared to the 1545.46464r drivers, nvidia is teh sux"

 

[vrecan]

well in my opinion its a lot harder to see IQ issues in screenshots then it is in real life watching it on your monitor. So something that is marginaly diferent in a screenshot can be extremely obvious in motion. Since every review so far seems to have noticed the fx's being down right ugly with the newest drivers I am more likely to believe them.

 

[jvd]

I'll tell you this much. If your an ati user you know the last two cat drivers have a horrible opengl bug in them. It makes banding effects . Well i rather play quake 3 on my radeon 9800pro with the 3.7s than on my geforce fx 5800ultra with the new dets that i have. The image quality is still better.

 

[sonix666]

Not to mention that I can't play the old Unreal Tournament on my Radeon 9700 with Catalyst 3.7. Problems I have noticed when playing at 1280x968 6xFSAA and 16xanisotropic filtering (performance):

- Some objects (players, items, etc.) are invisible at certain angles and distances.

- Mipmaps are messed up bigtime with animated textures. Until I get close, they are completely wrong.

- The game is shown too dark at default settings. My Voodoo2, GF1 and GF3 never exhibited this problem. I have to max out gamma in game to make it playable.

Might ATi be cheating or do you suppose it are bugs?

 

[Star_Hunter]

Opengl gl bug that has 16 bit color on highest qualitygoing to high performance and 32 bit comes back......This has been fixed in the 3.8(they fixed it right away like 2 days after the 3.7's came out). Also if the game has a force 32-bit I do think its in 32 bit then.

 

[Iceman]

In all fairness, if you look at Zardon's flash animation in the actual article on DriverHeaven's site, it's not just the 51.75 shots that look worse. The ATI screenshots also show blur instead of the yellow/black striping from the 45.23 screenshot.

Yes in DHs screenies this is the case. But today, after the release of Aquamark 3, I took my own screenshot.

If you want the big picture(.png, 1 MB): http://staff.withingames.net/iceman/Aquamark3.png

You can clearly see, that the texture quality is equal to the 44.23 Detonator's. I think DH made a mistake here.

 

[Veridian3]

Very interesting Iceman...

...can you let me know the exact settings you have, both in the control panel and in aquamark. Thanks.

Also your system specs...

...i ran just about every driver build you can imagine whilst looking into that texture, and every one i tried had the same issue.

Stu