Adaptive - Anti-Alising in cat 5.9 ? [Archive]

tEd

29-Sep-2005, 13:13

set ASTT_NA to 0 in the registry:

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Contro l\Video\{5A61609A-036A-4D32-B447-F0E42DA9714D}\0000

5A61609A-036A-4D32-B447-F0E42DA9714D this number is different for every one , you may have multiple numbers at the location because of multiple driver installation but only one is active. To be sure just change the value in every one.

If you use CCC try icon you will now have a new option in the AA menu

I made a litte mistake. To enable adaptive AA you have to create a new entry ASTT and set it to 1. The ASTT_NA entry i mentioned above is only there to make the option visible in the CCC tray icon.

it AA alpha test textures and it works :)

Geo

29-Sep-2005, 13:15

And does it actually do anything on X8xx-class hardware?

Also, would this suggest that 5.9 are the "shipping" drivers for X1xxx?

tEd

29-Sep-2005, 13:16

And does it actually do anything on X8xx-class hardware?

Also, would this suggest that 5.9 are the "shipping" drivers for X1xxx?

it works

i tried ss2 demo and hl2 , it's awesome and works on x800 too

Colourless

29-Sep-2005, 13:46

That's pretty nice... makes me want to download those drivers now

*downloads drivers*

Jawed

29-Sep-2005, 13:50

So, the implication is that the patent relating to, e.g., a maximum of 4 triangles meeting within one pixel is not what's referred to as "adaptive AA".

Hmm.

Jawed

tEd

29-Sep-2005, 13:51

It can be pretty slow too. If you have a game with alot of alpha test textures you may encounter a nice little performance hit.

tEd

29-Sep-2005, 13:52

So, the implication is that the patent relating to, e.g., a maximum of 4 triangles meeting within one pixel is not what's referred to as "adaptive AA".

Hmm.

Jawed

nope

Geo

29-Sep-2005, 13:59

It can be pretty slow too. If you have a game with alot of alpha test textures you may encounter a nice little performance hit.

That'll be interesting to check on R5xx hardware, whether it is hardware-acelerated there beyond what X8xx class hardware has. Might be a case of it will work in both, but with R5xx you get hardware acelleration. . . Which would explain why you had to go registry hacking for it. . .

tEd

29-Sep-2005, 14:15

I made a litte mistake. To enable adaptive AA you have to create a new entry ASTT and set it to 1. The ASTT_NA entry i mentioned above is only there to make the option visible in the CCC tray icon.

tEd

29-Sep-2005, 14:24

adaptive AA(fps in this picture 64):

http://www.gaeugf.ch/ted/ad_AA.png

no adaptive AA(fps in this picture 74):

http://www.gaeugf.ch/ted/noad_AA.png

x800pro cat5.9 / 1024*768 6xAA 16xAF AI off

Colourless

29-Sep-2005, 14:33

Ok now, this was unexpected... this is what FSAA Viewer said was happening:

http://www.users.on.net/%7Etriforce/atiadaptive.png

That looks like supersampling...

tEd

29-Sep-2005, 14:37

it doesn't show up like that in FSAA viewer for me

29-Sep-2005, 14:37

So I wonder are they using Humus tweak now? Remmeber he released those DX/GLOveride files to allow for this a month or so ago...

Colourless

29-Sep-2005, 14:39

Probably need my new version of FSAA Viewer that i made for the GF TAA

Download it here: http://www.users.on.net/~triforce/d3d_fsaaviewer/DX9FSAAViewer52.zip

tEd

29-Sep-2005, 14:47

could be that the adaptive part only works on r5xx

Colourless

29-Sep-2005, 14:51

Anyway the difference between the old versions of FSAA viewer and the new one was how the texture sample display code was done. The new one uses an Alpha test, the old ones didn't.

tEd

29-Sep-2005, 14:54

it looks that way

http://www.gaeugf.ch/ted/fsaatest.png

tEd

29-Sep-2005, 14:55

Anyway the difference between the old versions of FSAA viewer and the new one was how the texture sample display code was done. The new one uses an Alpha test, the old ones didn't.

ah ok that explains it

Geo

29-Sep-2005, 15:07

could be that the adaptive part only works on r5xx

Which would be sort of interesting on its own, as it might also mean waiting for the other shoe to drop on making regular SS, or even mixed-MS/SS (ala NV 8xs) modes available for full-screen. If I'm remembering correctly, there has been a small but vocal subset of FanATIcs yelling for SS since R300 came out, and ATI has always said "Meh. Given the performance hit, not high on the priority list." If they had to do it anyway in order to support this, then I can't see why they wouldn't make it available on its own as well. . .

wireframe

29-Sep-2005, 15:08

adaptive AA(fps in this picture 64):

http://www.gaeugf.ch/ted/ad_AA.png

no adaptive AA(fps in this picture 74):

http://www.gaeugf.ch/ted/noad_AA.png

x800pro cat5.9 / 1024*768 6xAA 16xAF AI off

That looks very nice. Could you run a timedemo to present some average fps numbers? Perhaps you should test this with lower AA, etc to determine the relative performance impact with other IQ enhancing options enabled/disabled.

tEd

29-Sep-2005, 15:10

Which would be sort of interesting on its own, as it might also mean waiting for the other shoe to drop on making regular SS, or even mixed-MS/SS (ala NV 8xs) modes available for full-screen. If I'm remembering correctly, there has been a small but vocal subset of FanATIcs yelling for SS since R300 came out, and ATI has always said "Meh. Given the performance hit, not high on the priority list." If they had to do it anyway in order to support this, then I can't see why they wouldn't make it available on its own as well. . .

well it seems they do SS only on alpha test that is where the adpative part comes in apparently.

what's interesting though they offer up to 6xAA for alpha test and it seems to work on all cards(r300 class). It makes the 5xx even more interesting with the HDR AA and alpha test AA and from what i know definately not worse AF (maybe even alittle better)

Geo

29-Sep-2005, 15:14

well it seems they do SS only on alpha test that is where the adpative part comes in apparently.

I hear you. I'm just saying it seems to me that is a *harder* case than doing regular fs ss, and as such they've probably already done the lion's share of the work required to make fs ss available, en passant as it were. . if they want to. They've always presented it as a priority issue for making the resources available for doing the work in the past. . .

tEd

29-Sep-2005, 15:16

I hear you. I'm just saying it seems to me that is a *harder* case than doing regular fs ss, and as such they've probably already done the lion's share of the work required to make fs ss available, en passant as it were. . if they want to. They've always presented it as a priority issue for making the resources available for doing the work in the past. . .

Frankly i'm surprised as well. What would have been if they offered that feature 2 years ago? Questions , questions :)

digitalwanderer

29-Sep-2005, 15:21

OMG! OMG! OMG! :shock:

Thank you tEd, I was just playing around with the new Tray Tools which claim to support this...but I have no clue how to enable it with it yet. (Still pre-coffee/dealing with kids though...gimme a bit. ;) )

Is there anyway to enable it without CCC, I still prefer the old control panel. :oops:

BTW-Those screenshots made me drool!!! :D

Geo

29-Sep-2005, 15:22

Frankly i'm surprised as well. What would have been if they offered that feature 2 years ago? Questions , questions :)

Well, I think two years ago it was pretty clear that they (with justification, in my view) felt pretty smug about their AA superiority over the competition. . .but NV has been pushing them harder on that front of late, so I think mayhaps they are beginning to "look to their laurels" as it were. . .

digitalwanderer

29-Sep-2005, 15:29

Damn it, I'm about to bite the bullet and install CCC...I WANT TO SEE THIS!!!!

Imagine GTA without jaggies.... :shock:

tEd

29-Sep-2005, 15:35

Damn it, I'm about to bite the bullet and install CCC...I WANT TO SEE THIS!!!!

Imagine GTA without jaggies.... :shock:

you don't need CCC to enable it. Either use ati tray tools which seems to have it integrated or manually edit the registry as i explained

RingWraith

29-Sep-2005, 15:53

Even with ATT you need to enable it manually. Easiest way to do it is to open the registry and do a 'search' for 'ASTT' and change the value to '1'.

HL2 seems to be the only DX9 game that is still playable with it, and DX7 and DX8 games are all fine. If you really want to see this in all its glory load up Black & White, it seems to use quite a bit of transparant textures and it looks gorgeous!!!!

tEd

29-Sep-2005, 15:54

Could it be that they mix alpha to coverage and SS dependent on whether the texture is minified or magnified? I get the impression that on some alpha textures it looks like alpha to coverage and on others like SS

trinibwoy

29-Sep-2005, 15:56

gybe

29-Sep-2005, 16:48

Is it suppose to work in OpenGL? I tried with a testapp using transparent textures and it look like it's not working.

digitalwanderer

29-Sep-2005, 16:48

I searched me registry and changed all instances of ASTT_NA to 0 and a new option showed up in TrayTools. :)

Now I'm going a bit nuts running games and seeing what if any differences I can see. AA works better in Bloodrayne, but it's weird....fantastic in spots and still bad in others. :?

Ultimate Spiderman it didn't help with, but that's more of a shader issue methinks. (But I was hopeful. :oops: )

tEd

29-Sep-2005, 16:50

Is it suppose to work in OpenGL? I tried with a testapp using transparent textures and it look like it's not working.

it seems not to work in opengl (yet?)

CMAN

29-Sep-2005, 16:53

Anyone able to test it aside a 7800 and see the quality and speed differences? (obviously the 7800 will be quicker, but it will be interesting to see the %drop by adding the adaptive AA compared to TAA).

Chalnoth

29-Sep-2005, 16:54

So, the implication is that the patent relating to, e.g., a maximum of 4 triangles meeting within one pixel is not what's referred to as "adaptive AA".
No, I don't see how one could call that algorithm "adaptive AA."

On the subject of the thread, I am rather impressed that ATI was able to get adaptive supersampling working on the R4xx.

29-Sep-2005, 17:24

Just tried it...works pretty well (only looked at CS Sorce for now)... Thanks Ted!!

tEd

29-Sep-2005, 17:35

:)

http://www.gaeugf.ch/ted/1.jpg

http://www.gaeugf.ch/ted/2.jpg

different edge and alpha test sample numbers

martrox

29-Sep-2005, 20:07

WOW........I'm impressed, and very happy, too!

MatiasZ

29-Sep-2005, 20:09

How come isn't this enabled by default as an option on the CP? Isn't it something ATI should be 'proud' of?

29-Sep-2005, 20:18

How come isn't this enabled by default as an option on the CP? Isn't it something ATI should be 'proud' of?

Could be a working in progress type of thing as there maybe some bugs left to iron out. IIRC "Temporal AA" first showed up as a hidden reg setting first, then later as an offical Cat feature...

RejZoR

29-Sep-2005, 20:27

Ok,so if you get those dots under Texture squre that means you have Adaptive AA (Supersampling) ?

Coz it looks like this on my Radeon 9600 XTR (6xFAA with 2xTAA,yeah it was flashing a bit hehe):
http://img175.imageshack.us/img175/8741/adaar9600xtr3od.png

RejZoR

29-Sep-2005, 21:00

Why are my green dots brighter than on other screenshots?

Xmas

29-Sep-2005, 21:22

On the subject of the thread, I am rather impressed that ATI was able to get adaptive supersampling working on the R4xx.
What's so impressive about it? I expected this right from the moment NVidia presented their transparency AA.

Why are my green dots brighter than on other screenshots?
The more samples, the less weight each one has.

RejZoR

29-Sep-2005, 21:28

Aha,thanks for explanation. I can confirm Adaptive AA for Radeon 9600 Pro.

Here are two samples from CS:S using 1024x768 with 4xFSAA and Adaptive AA disabled and enabled.
You can notice the difference on balcony railing and tree leafs.

NO Adaptive AA
[img=http://img117.imageshack.us/img117/9333/noadaa7vk.th.jpg] (http://img117.imageshack.us/my.php?image=noadaa7vk.jpg)

WITH Adaptive AA
[img=http://img117.imageshack.us/img117/510/adaa7na.th.jpg] (http://img117.imageshack.us/my.php?image=adaa7na.jpg)

radeonic2

29-Sep-2005, 21:39

So how long untill nvidia enables taa with NV4X cards :lol:

CMAN

29-Sep-2005, 22:01

So how long untill nvidia enables taa with NV4X cards :lol:

Not long hopefully! That would make my day.

radeonic2

29-Sep-2005, 22:04

Not long hopefully! That would make my day.
Seeing as they rained on ati's superaa parade I suspect they will try to get it asap.

Chalnoth

29-Sep-2005, 22:13

What's so impressive about it? I expected this right from the moment NVidia presented their transparency AA.
Well, it seemed very likely to me that it would require a hardware change to be able to switch between supersampling and multisampling on a per-drawcall level.

Sharkfood

29-Sep-2005, 22:15

I posted these on Rage3D.. I'm a very, very happy camper right now. One of the stabbing temptations to picking up a 7800GTX has now been peacefully calmed... heh.

I'm playing most of my games at 1280x1024, or in the case of BF2, 1280x960.. so the performance hit for me really isn't that big. I'm glad ATI has this for all their cards now.

Adaptive off:
http://img60.imageshack.us/img60/5029/adaptiveoff2bj.th.jpg (http://img60.imageshack.us/my.php?image=adaptiveoff2bj.jpg)
Adaptive on:
http://img56.imageshack.us/img56/873/adaptiveon2kr.th.jpg (http://img56.imageshack.us/my.php?image=adaptiveon2kr.jpg)

grecco_julio

29-Sep-2005, 23:11

Awesome. When I saw this thread my first tought was GTA San Andreas without shimmering. Can anyone take screenshots from the game with Adaptive AA Enabled? I wonder if Catalyst will have this option in the next release. And how did you discover that hidden in the register ? You're THE GUY.

uwackme

29-Sep-2005, 23:14

PLEASE...

Explain EXACTLY how to do this.

The #1 post instruction clashes with other posts....set ASTT_NA to 0 in the registry

OR set it to 1? What is it.

tEd

29-Sep-2005, 23:17

PLEASE...

Explain EXACTLY how to do this.

The #1 post instruction clashes with other posts....set ASTT_NA to 0 in the registry

OR set it to 1? What is it.

ASTT_NA to 0 which gives a new adaptive AA option in the CCC tray icon menu.

Make a new entry ASTT(string works) and set it to 1 , then it should be enabled. To make it active you need to make driver refresh you can do that with changing AA or AF in the cp.

If you have CCC you can now enable and disable it through the tray icon

Arty

29-Sep-2005, 23:22

Nice find ted. http://www.beyond3d.com/forum/images/icons/icon14.gif

Deathlike2

29-Sep-2005, 23:25

This is awesome stuff...

I guess that AA programmability is getting more play than it ever did before (well, since TAA).

Nice find. ATI did a brilliant job investing in some transisters/space for AA....

DegustatoR

29-Sep-2005, 23:37

Don't know about ATI's Adaptive AA, but on G70 enabling SS TAA for GTASA isn't doing much to reduce translucent textures aliasing in this game...

If you want screenshots i can make some :)

no-X

30-Sep-2005, 00:01

TrAA works fine on my R9500 :-)

http://img262.imageshack.us/img262/4932/r9500traaon7zw.png

BrynS

30-Sep-2005, 00:08

Thanks tEd!

Although I wasn't aware of this discussion at the time, I believe that this feature (or at least a very similar implementation) was raised (http://www.beyond3d.com/forum/showthread.php?t=1563&highlight=R300+Alpha+textures) as far back as mid-2002, when it became known that R300 could perform AA on Alpha textures in OpenGL via a specific extension, but wasn't forced due to it inducing conflicts on older titles (Half Life was cited) because the extension was used for other purposes.

Anyone care to surmise if software/driver implementation issues were the reason for this feature only be available now or was it likely resolved some time ago and flagged for release along with R5xx?

Cheers,

BrynS

Chalnoth

30-Sep-2005, 00:14

Well, there's probably some overhead in switching between using and not using AA on a per-surface basis. Alternatively, ATI may be implementing this by sending each draw call multiple times with different pixel centers.

As such, it would be interesting to compare the performance of this adaptive AA vs. the implementation on the G70.

Sharkfood

30-Sep-2005, 00:32

As such, it would be interesting to compare the performance of this adaptive AA vs. the implementation on the G70.
It seems to only work with Direct3D for me, albeit I could be gacking something myself here (user error).

I made a quick "SharkBench" in BF2, firing up a 16-player Single Player and blowing up a bunch of stuff, running around, etc.etc. X800XT 1280x960, all game settings maxed, except lighting at Medium.

4xAA, Non-Adaptive "SharkBench"
Total Frame Time: 99.0029 seconds
Number of Frames: 12380
Average Frame Rate: 125.047
Average Frame Time: 0.007997
Shortest Frame Time: 0.00301361 (frame #1847)
Longest Frame Time: 0.292343 (frame #7017)

4xAA, Adaptive, "SharkBench"
Total Frame Time: 98.9651 seconds
Number of Frames: 12237
Average Frame Rate: 123.65
Average Frame Time: 0.00808737
Shortest Frame Time: 0.00307455 (frame #3791)
Longest Frame Time: 0.291626 (frame #6874)

Now at 1280x960 and on a 16-player benchmark, it's obviously not taxing an X800 XT. I'd love more "cruncher" demos, but alas- all the Guru3d ones don't work with the latest patch.

Moreover, Valve's latest patch has rendered HLBench useless as none of THEIR demos will play either.

If anyone has any cruncher BF2 or HL2 benchmarks, I'd be more than happy to put them through the paces. If someone has some storage, I'd be happy to zip up Sharkbench + def-cam files too.

gybe

30-Sep-2005, 00:37

Is it really super sampling or it's multisampling with multiple texture fetch on transparent textures?

Sharkfood

30-Sep-2005, 00:39

gybe- It's not supersampling. It appears to be totally isolated to alpha-test textures.

CS:S test @ 1280x1024, 4xAA
101.98 fps

CS:S test @1280x1024, 4xAA + Adaptive
100.46 fps

The "drop" seems to be particularly isolated to how much alpha-test texturing is in the scene. As the stress-test has next to zippo (maybe even zippo), performance really doesn't change much. I'm also semi-CPU bound for an X800 XT, but I'd expect even 2xSS to have a bigger impact or flip the tables.

Colourless

30-Sep-2005, 00:55

Supersampling only when doing Alpha Tests, hence adaptive.

I should make another version of the FSAA Tester that lets you choose the texturing mode, alpha tests or not.

Geo

30-Sep-2005, 00:58

This thread is now officially famous.

http://www.bit-tech.net/news/2005/09/29/cat_59_taa/

And [H] linked the bit-tech page from their front page.

Colourless

30-Sep-2005, 02:25

Ok, new version of D3DFSAATester again. When you press the "T" key it changes the rendering mode for the Texture samples. Either the Alpha Test mode or an Alpha Blended mode.

http://www.users.on.net/~triforce/d3d_fsaaviewer/DX9FSAAViewer54.zip
(http://www.users.on.net/%7Etriforce/d3d_fsaaviewer/DX9FSAAViewer54.zip)
and this if needed:

http://www.users.on.net/~triforce/d3d_fsaaviewer/D3DXDLLAugust27.zip (http://www.users.on.net/%7Etriforce/d3d_fsaaviewer/D3DXDLLAugust27.zip)

Richard

30-Sep-2005, 02:49

This is from the changelog of Dungeon Siege 2's latest patch:

Changed the ATI vendor setting of swapeffect_discard to false to support future card anti-aliasing routines.

ChrisRay

30-Sep-2005, 02:52

Transparency SS effect looks exactly the same as whats shown in these screenshots with the new FSAA viewer. It does not however show anything for transparency multisampling.

tEd

30-Sep-2005, 02:58

With setting a another entry ASD to 2 you can have only 50% of the MSAA samples apply to alpha test textures.

http://www.gaeugf.ch/ted/1.jpg

http://www.gaeugf.ch/ted/2.jpg

Often having a 6x super-sampling on games with many alpha test textures can be a pretty hard hit and with this you can have some alpha test AA and still maintain 6xMSAA for the edges

Colourless

30-Sep-2005, 03:05

But that is non optimal sample placement...

tEd

30-Sep-2005, 03:09

But that is non optimal sample placement...

It's something. You can also have 6xMSAA with 2x SSAA alpha test

http://www.gaeugf.ch/ted/3.jpg

yeah i know not optimal ;)

teraeric

30-Sep-2005, 05:25

tEd

30-Sep-2005, 06:10

i made the reg change, but cannot see the option in ccc... even after reboot... and creating ASTT and setting it to 1 doesn't seem to have any effect...

am i doing something wrong? the omegas are supposed to be the cat 5.9 version...

You can use this program to exactly locate the registry key where you have to make the changes in case you did it in the wrong place:
http://www.gaeugf.ch/ted/RegKey.exe

tEd

30-Sep-2005, 06:18

I was just thinking with crossfire you should be able to get some really nice sample patterns.

12xMSAA with 6x/8x AT SSAA that should look gorgeous

ballero

30-Sep-2005, 06:24

7800GTX

4x + SS Transparency AA http://paintedover.com/uploads/show.php?loc=26&f=hl2_4xtrss.png

8xs + SS Transparency AA http://paintedover.com/uploads/show.php?loc=26&f=hl2_8xstrss.png

Comparison screens

ATI (thanks to tEd) http://img254.imageshack.us/img254/2905/adaa3505cu.png

4x + SS Transparency AA http://img141.imageshack.us/img141/8075/hl24xtrss3506pw.png

8xs + SS Transparency AA http://img141.imageshack.us/img141/7658/hl28xstrss3502qv.png

RejZoR

30-Sep-2005, 08:45

Actually you can even apply Temporal Anti-Aliasing feature for Alpha textures.
This means you can basically get up to 18x Adaptive AA for price of 6xFSAA (using 3x TAA). Thats maximum. Most of us will probably use 2xTAA resulting 12x ADAA.
Not bad eh? :D

Chalnoth

30-Sep-2005, 09:32

Blaaaaarghmeshubashalachukaruuf!!

2x temporal dithering + 6x MSAA != 12x MSAA!!

Luminair

30-Sep-2005, 10:22

Make a new entry ASTT(string works) and set it to 1 , then it should be enabled. To make it active you need to make driver refresh you can do that with changing AA or AF in the cp.

I'd like to test this and make sure I know when it is on and when it is off; lets confirm exactly how to do it.

After changing the registry setting, you must either a) restart or b) change any AA or AF setting in the control panel, just to refresh the driver settings. So for testing back and forth, we can just move the slider in the control applet and press Apply. Is this correct?

Do we have to have Temporal AA enabled for this to work? I didn't have it on before, but I'm guessing that everyone here keeps it on all the time.

Great find!

mistan

30-Sep-2005, 11:56

RejZoR

30-Sep-2005, 12:16

No,Temporal AA is not required for Adaptive AA, but helps to increase quality because it's changing the AA grid every frame.

FnF

30-Sep-2005, 13:23

Just a question, is it possible to do this with Cats 5.8? Because when I use Regkey.exe it is telling me where to put the key. Then I open up the new ATT and then I enable it. But does it work then :?: Or is it only possible on the Cats 5.9 (And up maybe) :?:

klacik

30-Sep-2005, 13:46

FnF

30-Sep-2005, 14:02

i think it's only for Cats 5.9+
Hm, did anyone tried this out? I can use ATT to enable the setting on the 5.8's, but I dont know if is really is enabled or not...

PS: Why cant I edit my posts, this really sucks...

digitalwanderer

30-Sep-2005, 14:28

Snap!
It works in BF2.
Alls you gots to do is...
Set AA and/or AF to Application Preference. Then turn on A-AA (adaptive) and then turn on AA in the game's menu. No more ugly grass/trees/fences for me. This is on a 9800pro @ 1280x960 with everything on medium and lighting on high. Its just beautiful...brings a tear to my eye.
w00t!
Thanks! I was wondering if you could still use in-game AA and find a way to use the adaptive stuff with it, now I not only know you can but I know how. :)

Really, thank you.

Why cant I edit my posts, this really sucks...
You're new. You have to be here a certain length of time before you gain board privelages...there's a thread in feedback about it.

Geo

30-Sep-2005, 14:32

You're new. You have to be here a certain length of time before you gain board privelages...there's a thread in feedback about it.

Exec Summary: 10 days, 10 posts in tech forums, good behavior (no net-negative rep changes).

digitalwanderer

30-Sep-2005, 14:34

You're an executive now here? When did that happen!?! :shock:

;)

EDITED BITS: I'd like to think it goes without saying, but I think tEd deserves a bit of a rep bump for finding this for us...I bumped him.

klacik

30-Sep-2005, 14:41

[QUOTE=FnF]Hm, did anyone tried this out? I can use ATT to enable the setting on the 5.8's, but I dont know if is really is enabled or not...

i haven't tried this, but i don't think it works with older drivers..

quote from BitTech: "...to get what ATI are calling Adaptive Anti-Aliasing working on the current generation of Radeon X800 series video cards with the recently released Catalyst 5.9 display driver."

more here: http://www.bit-tech.net/news/2005/09/29/cat_59_taa/

digitalwanderer

30-Sep-2005, 14:55

more here: http://www.bit-tech.net/news/2005/09/29/cat_59_taa/
No offense at all intended to you Klacik, and welcome to B3D; but I find it really funny so many people are linking to the Bit-tech article, (and no offense against Bit-tech, I like 'em..an excellent site!), when that article is based pretty much only on info from this thread.

Does no one else see the irony? :lol:

THe_KELRaTH

30-Sep-2005, 15:08

Heh well they're "extremely overwhelmed to see it work on ATI's current generation".
I wonder what level of overwhelmed they'll be when someone tells them it works even with the R300 series then... Ultra extremely?

klacik

30-Sep-2005, 15:13

No offense at all intended to you Klacik, and welcome to B3D; but I find it really funny so many people are linking to the Bit-tech article, (and no offense against Bit-tech, I like 'em..an excellent site!), when that article is based pretty much only on info from this thread.

Does no one else see the irony? :lol:

lol.. now i see you're right

digitalwanderer

30-Sep-2005, 15:13

It works on 9700 & 9500s too? I missed that, is it playable? (And please don't bash Bit-tech, I really do like 'em.)

klacik

30-Sep-2005, 15:14

Geo

30-Sep-2005, 15:22

digitalwanderer

30-Sep-2005, 15:25

..and thanks for inviting me here Digi
Pfffft! Being a gracious forum butterfly is what I'm all about. ;)

I re-downloaded HL2 & CS:S last night just to see the chainlink fences in the beginning and they were just fracking incredibly crisp...it was beautiful.

And GTA... :D

This still doesn't help in shader busted AA games though, Spider-man Ultimate and BloodRayne2 still got funkiness going down...but I'm very enthused!

I'm trying like hell to get hold of Terry Makedon about this, but am having no luck...I think he's hiding from me so he doesn't have to tell me he can't tell me yet. I thnk I might have to do a sneak attack on him and actually call him or Patti and catch 'em with their guard down... ;)

digitalwanderer

30-Sep-2005, 15:26

Well, you see, bit-tech took that extra step of cut n' pasting the two halves of the image together. :lol:

The important thing is they were very upfront about where they got the stuff, and linked B3D twice (once to the forums in general, and once to this thread). And end o' the day, everybody linking into the bit-tech article will probably punch thru to here anyway. So it's all good.

I just think of the bit-tech page as the "exec summary". :wink:
Oh no, I totally agree! I can't even count the times I've made forum posts into stories like that, it's good to get the information on the frontpages of a few places so more people know.

As long as they link and credit it's all good IMHO, and they did a very good job of linking and creditting. :)

digitalwanderer

30-Sep-2005, 15:32

klacik

30-Sep-2005, 15:37

Pfffft! Being a gracious forum butterfly is what I'm all about. ;)

I re-downloaded HL2 & CS:S last night just to see the chainlink fences in the beginning and they were just fracking incredibly crisp...it was beautiful.

And GTA... :D

hah.. you reminded me of one Dexter's Laboratory (cartoon) part, with D&D figures in it.. 'think there was some fairies ;)

i wanted to try this too, but i'm stuck with gf6600gt now.. hope those new ATI cards will be out soon! i'm deninetly going to buy one of x1600xt's or the lowest x1800's :-D

this 6600gt is my first and last nV card.. can't help it.. ATI's cards were always my favourite.. performance/quality-wise...

Geo

30-Sep-2005, 15:38

:oops:

I can't believe it, but I actually got stage fright leaving a mesage on Terry's voicemail. When the beep toned I realized it was the first time I'd ever actually heard his voice and the first time he'd hear mine, and I was a goofball. :oops:

I tried. Next is Patti.....

Don't forget Ibiza (and Oakland? Who knows) apparently started today. . .probably hit 'n miss until Monday on who is reachable. . .

sonix666

30-Sep-2005, 15:54

The improvements are great. I am surely going to try this in World of Warcraft, since that game does sport quite alot of alpha test transparent textures. And I am curious what the performance hit will be for my Radeon X800 Pro.

I do hope that in the future the sampling patterns will be better when you use less samples for transparent textures.

Sxotty

30-Sep-2005, 15:56

this 6600gt is my first and last nV card.. can't help it.. ATI's cards were always my favourite.. performance/quality-wise...

You should hope that ATI doesn't go under then :p

Anyway, I dunno personally still why people are so particular, I have had many of both and they seem good to me with the exceptions of a few bad apples. Anyway heres hoping the r520 is awesome, and all the derivatives too.

digitalwanderer

30-Sep-2005, 15:58

Don't forget Ibiza (and Oakland? Who knows) apparently started today. . .probably hit 'n miss until Monday on who is reachable. . .
Hubba-wa? What's Ibiza, what's up in Oakland? (I'm still waking up :oops: )

RejZoR

30-Sep-2005, 16:00

Trees and bushes in CS:S look so damn nice i got killed several times while i was watching them. Aaaaaah :D

uwackme

30-Sep-2005, 16:00

digitalwanderer

30-Sep-2005, 16:07

This is great BUT.

You need to be EXACT in registry editing.

So again, what PRECISELY ( as defined in the dictionary under precisely) do you have to do to enable this.

Ex"...on ATI, tyou have TWO entries in registry.... one for PRIMARY, and one for SECONDARY. What ever registry change this needs should be done to BOTH.

But is it a REG_DWORD? or a REG_STRING... or WHAT.

There isnt room to fk around, what PRECISELY needs to be done to activate this. Think instructions for DUMMIES and advise please.
If you're a thicky, do what Digi does....

Open up your registry, go to your "find" and enter "ASTT_NA" (without the quotes) and set the value to zero. Repeat until you don't find "ASTT_NA" anymore. (I had to replace 4 or 5 of them)

Close your registry, reboot. (I don't know if that is necessary or not, but I did)

Open up ATi Tray Tools by the ever wonderful Ray Adam's and check the "adaptive AA" box and hit "ok", you're good to go.

(Oh, make sure you have the Cat 5.9s installed first. ;) )

Geo

30-Sep-2005, 16:15

Hubba-wa? What's Ibiza, what's up in Oakland? (I'm still waking up :oops: )

Well, if I read the tea leaves correctly, the "Tech Day at Ibiza" that Andrzej talked about in his long post re Sassen is getting underway today, and is actually a full 3 day weekend. I linked a post from some French site about it somewhere else. And there's Wavey telling Rys to "Go, Drink!" yesterday. :lol: And Faud did a piece this morning where he said "Off to Ibiza!" at the end.

Supposedly there was going to be one in Oakland too, for Norte Americanos. Don't know if it was siimultaneous or not with Ibiza, but may be. Or maybe they did them in turns so as to provide all the heavyweights at both.

At any rate, if I haven't completely lost my tea-leaves reading ability, I would think it would be very difficult to get ahold of "the usual suspects" from ATI from today thru Sunday.

Mariner

30-Sep-2005, 16:26

Ibiza? The party capital of Europe (some would say the world)? Why would anyone from tech sites want to attend a product launch there, I wonder? :smile:

uwackme

30-Sep-2005, 16:33

KillerSneak

30-Sep-2005, 16:41

Did anny of you guys look into the AdaptNPatch Key ? And why TrueFor doesn't show in the Normal ATI CP with the Catalyst 5.9 drivers ?

tEd

30-Sep-2005, 16:44

You can use this program:
http://www.gaeugf.ch/ted/RegKey.exe

It will show you the exact location in the registry where to add things.

Change ASTT_NA to 0 to get the option in CCC.

Add ASTT entry as string(REG_SZ) and set it to 1(REG_BINARY works too) to enable adaptive AA.

http://www.gaeugf.ch/ted/reg.jpg

Restart or make driver refresh by changing AA or AF in CCC or CP to make the changes active

THe_KELRaTH

30-Sep-2005, 16:46

Did anny of you guys look into the AdaptNPatch Key ? And why TrueFor doesn't show in the Normal ATI CP with the Catalyst 5.9 drivers ?

I understand that this bit of code has now been removed from the drivers, if you still want to use it you will have to use earlier drivers.

FnF

30-Sep-2005, 17:37

Thanks! I was wondering if you could still use in-game AA and find a way to use the adaptive stuff with it, now I not only know you can but I know how. :smile:

Really, thank you.

You're new. You have to be here a certain length of time before you gain board privelages...there's a thread in feedback about it.
Thnx! I couldnt find anything about it in the threads here :smile: But I always visited this site, I know I have another account here, but I dont know it anymore:sad:

Exec Summary: 10 days, 10 posts in tech forums, good behavior (no net-negative rep changes).
Thnx geo!

It works on 9700 & 9500s too? I missed that, is it playable? (And please don't bash Bit-tech, I really do like 'em.)
Yes, it does work on that cards as well :smile:

FnF

30-Sep-2005, 17:38

Don't forget Ibiza (and Oakland? Who knows) apparently started today. . .probably hit 'n miss until Monday on who is reachable. . .
I believe Terry was visiting Ibiza :smile:

Hubba-wa? What's Ibiza, what's up in Oakland? (I'm still waking up :oops: )
The official launch of the X1x00 series :smile:

Did anny of you guys look into the AdaptNPatch Key ? And why TrueFor doesn't show in the Normal ATI CP with the Catalyst 5.9 drivers ?
ATi skipped the support of TrueForm, today's games arent using it anymore, so they skipped the support of it.

Still strange, I cant quote 6 things in one post, now I need to make two posts (Double Post). They really should do something about these settings...

digitalwanderer

30-Sep-2005, 17:59

good zinger, but misses the point. I dont have ATI Tray Tools, and want to edit the registry myself.

I get how to find the WHERE, what I need is PRECISELY what key to ADD.... what is it called and is it a REG_DWORD or REG_STRING?

Some of us dont use ATITrayTool.
Ah, sorry. I wasn't trying to be sarcastic, I thought you were honestly confused. :oops:

Thanks Geo for the info, I did talk to Patti but she can't tell me anything quite yet... ;)

Geo

30-Sep-2005, 18:02

Still strange, I cant quote 6 things in one post, now I need to make two posts (Double Post). They really should do something about these settings...

Well, I hate to mention this in polite company, for the same reason I don't take a drink around known alcoholics, but, umm, err. . .

It was probably the number of emoticons that was messing you up. You can't have more than four in any one post, both your own and any in the quoted material you may have included. Try not to talk about it around you-know-who.

KillerSneak

30-Sep-2005, 19:07

I believe Terry was visiting Ibiza :smile:
ATi skipped the support of TrueForm, today's games arent using it anymore, so they skipped the support of it.

I know but the AdaptNPatch KEY hasn't been arround for to long if I'm not mistaken..

teraeric

30-Sep-2005, 19:51

kinda strange... i tried a different version of atitray than came with the omega driver and the option shows up...

but i'm a little befuddled about enabling it...

i tried a couple of different combinations of aa an af, and if i turn af DX9FSAAViewer.exe doesn't show the multiple images in the boxes...

and i have to have 2x aa enabled, also to get multiple images in the boxes... but if i go 4x i get the single image...

sorry if this is noobville, i thought i had a grasp of the nature of these features but it seems that it might be a bit more superficial than i thought...

Purple Stuff

30-Sep-2005, 21:10

Excellent work indeed.

Dwarden

30-Sep-2005, 21:41

You can also use updated ATT

http://www.radeon2.ru/atitray/attsetup.exe

version 1.03.755

* Updated core and product names database. Included X1300, X1800 information
* Fixed bug in Display Detection procedure in Windows 64
* Fixed error with ScreenShot file name generation
* Fixed bug with Custom Game profiles.
+ Added option in Advanced Direct3D Tweaks database to force support for Adaptive AA

have fun

uwackme

30-Sep-2005, 22:46

Thanks guys, got it working, things seem to look better, but since it's adaptive its also hard to notice clearly while being shot at, etc ;-)

AArmy, FarCry, HL2, CS-S all stay quite playable.... A64 2.75Ghz w/X850XT 560/560 2G ram on DFI NF4. Sitting at 4xAA, 8xAF with A-AA and TAA enabled.

Chain link fences look better in HLife2 for example.

Vadi

30-Sep-2005, 22:52

I love you! :D

MuFu

01-Oct-2005, 02:15

Seems to work fine on a 9700 Pro.

Kanyamagufa

01-Oct-2005, 02:23

Brilliant discovery tEd. :grin:

You da man.

SlowJoeBlow

01-Oct-2005, 02:33

I'm a noob,let's get that out of the way :oops: I have a 9700 Pro..Got the New ATT and 5.9's..but before i got the new ATT I did the regedit thing, that I'm really a noob at, ASTT_NA = 0, created ASTT=1 strings, I have the AAA tab on, and what i want to know..from 9XXX user's, is how do I verify, that AAA is truely enabled?

Thanks :grin:

SlowJoeBlow

01-Oct-2005, 02:36

Seems to work fine on a 9700 Pro.

MuFu If ya got the time can ya e-nail me at wkirby20032003@yahoo.com and let me know how ya set up your 9700Pro?
Thanks

Killer-Kris

01-Oct-2005, 07:23

I can report that it also works on a 9500Pro, though there's really no reason it shouldn't.

digitalwanderer

01-Oct-2005, 07:30

I can report that it also works on a 9500Pro, though there's really no reason it shouldn't.
Well, except for the fact that I've had a friend of mine tell me point blank that R300 hardware would NEVER support aa on transparent textures. (True story, I was bugging 'em about GTA:VC at the time. :oops: )

Colourless

01-Oct-2005, 08:18

Just noting I updated FSAAViewer again with a bug fix... Get from the link in my sig

Killer-Kris

01-Oct-2005, 08:45

Well, except for the fact that I've had a friend of mine tell me point blank that R300 hardware would NEVER support aa on transparent textures. (True story, I was bugging 'em about GTA:VC at the time. :oops: )

I mainly meant that if it was working on 9700s and 9600s it should obviously work on 9500s.

From what Humus has shown, and what others said I always thought it was a little weird that Ati never allowed us to force any sort of transparency AA before now.

Chalnoth

01-Oct-2005, 08:49

Well, except for the fact that I've had a friend of mine tell me point blank that R300 hardware would NEVER support aa on transparent textures. (True story, I was bugging 'em about GTA:VC at the time. :oops: )
Well, the R300 has been out for what, four years now?

If the hardware was designed to be capable of AA on alpha test surfaces, you'd think they would have done it by now.

I'm willing to bet that instead all that we're seeing is ATI re-sending drawcalls on alpha test surfaces with different pixel centers. This could be tested easily by drawing a few alpha test surfaces with monstrous vertex programs (enough to make the benchmark completely vertex limited), and comparing between using and not using dynamic supersampling.

If ATI is indeed using this method, then I expect it to have performance issues in places where lots of draw calls are used, such as with foliage, especially grass.

Colourless

01-Oct-2005, 09:20

martrox

01-Oct-2005, 09:49

Absolutely kills performance in WOW....at 6X FSAA/16xAF everything on at 1680X1050 on Dell 2005....... It's the damn trees! They must be all Alpha textures, runs great when few trees are onscreen. Running AMD64 939 3000+@ 2650, X850XT @540/575.
However, it look incredible.........dropped to 4XFSAA/8XAF and it still kills pefromance, hate to drop resolution but will give it a try.......;)

ChrisRay

01-Oct-2005, 09:57

In World of WarcraftTransparency Super Sampling causes performance to drop into the 20 FPS from 90 FPS on 7800GTX cards in SLI. Ironically. Transparency Multisampling suffers no hit worth mentioning in WoW((not even -1%)) and really looks almost as good. Honestly aside from HL2 too I have still preferred transparency multisampling for most games which are still GPU limited on my setup because the performance tradeoff is pretty much non existence.

Chalnoth

01-Oct-2005, 09:58

Most of the performance problems with lots of Draw call is due to direct3d itself and the way the data is transfered from the runtime to the drivers. Chances are the drivers would be resending 'decoded' instruction so the hit would be much lower than doing multiple DrawPrimitive calls.
That's true, but there's still going to be a hit from rendering more geometry and having more bus traffic. Depending on the application, this may or may not be an issue.

Fences, for example, are typically just a few large quads, and so wouldn't be affected by this at all. But take a game like FarCry that has huge numbers of alpha tested foliage, and it may start to become an issue.

It is always conceivable, of course, that the extra vertex shader processing and bus bandwidth usage will be overshadowed by the extra fillrate usage, but this won't be the case for occluded pixels.

martrox

01-Oct-2005, 10:13

Bouncing Zabaglione Bros.

01-Oct-2005, 10:16

I'm willing to bet that instead all that we're seeing is ATI re-sending drawcalls on alpha test surfaces with different pixel centers. This could be tested easily by drawing a few alpha test surfaces with monstrous vertex programs (enough to make the benchmark completely vertex limited), and comparing between using and not using dynamic supersampling.

That may be the case for older ATI hardware, but we don't yet really know if the R5x0 has hardware support for this feature that minimises any framerate impact.

Regardless, it's still pretty nice for customers with older cards to have this feature as an option.

ChrisRay

01-Oct-2005, 10:18

Well...tried lowering the resolution and it still kills performance at 1280x720...and looks absolutely fuzzy on the 2005 LCD. So it's a no go on WOW for me, I'm afraid. Looks like I need a new Videocard (or 2!*) ......;)

*especially if I get a Dell 2405........

Well my 7800GTX SLI setup crashes((Performance wise)) with transparency SS. It seems like a game issue. Though if you had an Nvidia card transparency multisampling is a good option.

*Edit* Made a correction to make sure people didnt think my computer crashed with Transparency SS. :P

martrox

01-Oct-2005, 10:29

I plan on getting a 7800GT/GTX and a X1800XT.......I have a SLi MB, and will get a Crossfire MB when they are available. I'll play with both, read the comparison reviews and then decide which way to go.......

Xmas

01-Oct-2005, 13:36

Well, the R300 has been out for what, four years now?

If the hardware was designed to be capable of AA on alpha test surfaces, you'd think they would have done it by now.
Transparency multisampling is possible since NV20 (and a somewhat flawed variant of transparency supersampling, too). You'd think it takes less than four years...

I'm willing to bet that instead all that we're seeing is ATI re-sending drawcalls on alpha test surfaces with different pixel centers.
And why do you think NVidia is doing anything different? Because it's not available on NV40 and earlier? Well, NVidia doesn't have programmable sample positions, so they had to add some changes in hardware.

RejZoR

01-Oct-2005, 13:39

Too bad we can't use this feature in OpenGL games :(
There is so many Quake 3 based games that render outdoor scenes (FAKK2 for example).

martrox

01-Oct-2005, 14:07

Works in World of Warcraft.......and that's OpenGL........!!!!!!

bloodbob

01-Oct-2005, 14:10

Well, the R300 has been out for what, four years now?

If the hardware was designed to be capable of AA on alpha test surfaces, you'd think they would have done it by now.
Ahh so ATI was lying when they said they could do SSAA ( and well advertised )?

Kynes

01-Oct-2005, 14:31

Works in World of Warcraft.......and that's OpenGL........!!!!!!
WoW is DirectX9

martrox

01-Oct-2005, 15:14

WoW is DirectX9
It's both.......In checking ATT, says it's OpenGL. However, you are right in that I am using D3d....

Chalnoth

01-Oct-2005, 16:31

Transparency multisampling is possible since NV20 (and a somewhat flawed variant of transparency supersampling, too). You'd think it takes less than four years...
Not necessarily. For all we know you can't change the pixel center relative to the z-sample centers in NV hardware at all, or at least not on a per-drawcall basis.

And why do you think NVidia is doing anything different? Because it's not available on NV40 and earlier? Well, NVidia doesn't have programmable sample positions, so they had to add some changes in hardware.
That's entirely possible. I never said that nVidia was doing anything different. There was just no evidence one way or another what nVidia was doing.

Chalnoth

01-Oct-2005, 16:32

Ahh so ATI was lying when they said they could do SSAA ( and well advertised )?
Huh? Supersampling's been working in their Mac drivers for ages. The ability to do supersampling is completely unrelated to the ability to do adaptive supersampling.

FnF

01-Oct-2005, 17:33

Well, I hate to mention this in polite company, for the same reason I don't take a drink around known alcoholics, but, umm, err. . .

It was probably the number of emoticons that was messing you up. You can't have more than four in any one post, both your own and any in the quoted material you may have included. Try not to talk about it around you-know-who.
'They' are the admin(s)/owner(s) of this wonderful site :) Why im saying this is because this is the only site who has such restrictions, and I dont see why the maximum allowed quotes should be 4, this only makes double, tripple and even more posts, this is making the topics go faster then it should be...
I know but the AdaptNPatch KEY hasn't been arround for to long if I'm not mistaken..
Its there already since december 2004 I believe... 4.12 had it anyway...

sonix666

01-Oct-2005, 17:37

After enabling this in WoW I am certainly impressed. If this has been possible all along on an RX800, why o why didn't they do it before. :(

5150 Joker

01-Oct-2005, 20:25

After enabling this in WoW I am certainly impressed. If this has been possible all along on an RX800, why o why didn't they do it before. :(

They claimed SSAA wasn't economically feasible on the PC (b.s. excuse). Adapative supersampling..well I'm sure they've had it down for quite awhile and have been waiting on R520 to finish up so they could intro. it as a new feature.

Chalnoth

01-Oct-2005, 20:45

They claimed SSAA wasn't economically feasible on the PC (b.s. excuse). Adapative supersampling..well I'm sure they've had it down for quite awhile and have been waiting on R520 to finish up so they could intro. it as a new feature.
If they can enable it in software now, you can be certain there was no great amount of work required to get it operating. This is a response to nVidia's transparency AA.

BRiT

02-Oct-2005, 00:46

This is a response to nVidia's transparency AA.

Which was in response to CrossFire SuperAA modes, no?

Seems like it's all a series of cat and mouse moves anyways...

EasyRaider

02-Oct-2005, 00:55

Better late than never...

This feature should have been introduced alongside multisampling. MSAA has always seemed useless to me in all the games with alpha tests (and now, in games with bad shader aliasing, but I guess we'll have to rely on devs to fix that).

Xenus

02-Oct-2005, 00:55

Which was in response to CrossFire SuperAA modes, no?

Seems like it's all a series of cat and mouse moves anyways...

No SLI AA or watever its official name is was the response to Xfire SuperAA. Transparency AA came out with the G70.

EasyRaider

02-Oct-2005, 00:56

Which was in response to CrossFire SuperAA modes, no?
I don't think so, NVidia has offered mixed modes for a long time.

bloodbob

02-Oct-2005, 01:04

Huh? Supersampling's been working in their Mac drivers for ages. The ability to do supersampling is completely unrelated to the ability to do adaptive supersampling.
Yes its completely unrelated to adaptive supersampling buts its not completely unrelated to what you said which is they can't do AA on alpha tested surfaces ( which is what you said and I quoted) clearly SS does AA on alpha tested surfaces.

Xmas

02-Oct-2005, 11:30

Not necessarily. For all we know you can't change the pixel center relative to the z-sample centers in NV hardware at all, or at least not on a per-drawcall basis.
That's why I wroteTransparency multisampling is possible since NV20 (and a somewhat flawed variant of transparency supersampling, too)

Geo

02-Oct-2005, 13:17

They claimed SSAA wasn't economically feasible on the PC (b.s. excuse).

Did they? Not my memory that their statements were that bald. I seem to recall it was more like they didn't like the performance hit considering they also felt that their AA was clearly superior to NV's anyway, even with NV using it --which then translated into a resource argument; "why spend the time on it?". Which might make one wonder if there was something in the two architectures that made it marginally more efficient for NV to do, so ATI didn't want to do it given they already felt superior in AA in general.

tEd

02-Oct-2005, 17:09

Did they? Not my memory that their statements were that bald. I seem to recall it was more like they didn't like the performance hit considering they also felt that their AA was clearly superior to NV's anyway, even with NV using it --which then translated into a resource argument; "why spend the time on it?". Which might make one wonder if there was something in the two architectures that made it marginally more efficient for NV to do, so ATI didn't want to do it given they already felt superior in AA in general.

Yes somethng like that. I think they tried to implement it once and decided that it takes to long and to many recources after implementation to support it. Plus considerng the implication of the big performance hit they didn't feel it was good choice.

Dave Baumann

02-Oct-2005, 22:52

FYI, AFAIK this is intended to be fully supported by R300 onwards, however they haven't got round to Q/A'ing it on these parts yet.

tEd

02-Oct-2005, 23:15

FYI, AFAIK this is intended to be fully supported by R300 onwards, however they haven't got round to Q/A'ing it on these parts yet.

what about opengl support?

Dave Baumann

02-Oct-2005, 23:29

Intended, but prioritised relative to the level of title support...

Geo

02-Oct-2005, 23:42

Intended, but prioritised relative to the level of title support...

The two questions I really want answered I would imagine can't be until Wednesday.

1). Is there any special hardware accel in R5xx beyond R4xx for this feature?
2). Would we have seen it in a product announced at, say, Computex? The fact that it will be supported in R3xx suggests one way (i.e. pure "Hey, great idea, NV. . .let's do it too." Not that there is anything wrong with that --see NV jumping on the SuperAA bandwagon). But if there is special accel in R5xx it would suggest the other way.

Not that it matters all that much in the end, just curious minds and all that wondering who's "innovation" bucket to drop this little dollop of credit in, or whether to call it a tie. . .

mjtdevries

03-Oct-2005, 08:08

1) If they are really doing supersampling than there is very little that you can accelerate in hardware, afaik. And then it makes sense that this new feature would be available to all cards. (and of course they will only introduce it with R520 so that it's introduction will look even better)

2) Is it a great idea? In the case of supersampling it's not new. I discussed smarter methods to AA alpha textures with opengl guy on this (or Rage3D's) forum a few years ago (R300 timeframe). But the main problem was that in every case you ended up doing supersampling and that was seen as a unacceptable.
This feature is nice in games where only a very small part of the screen has alpha textures. But you could end up with your entire screen getting supersampling AA, and then you get a 75% performance hit with 4xAA. Is that really a great idea? Would people then use it?

I think they thought: "Damn, Nvidia added this useless feature, but the reviewers like it????? Guess that means we will have to implement it too...."

This is not innovation. It's a return to technologies we turned away from (with good reasons) in the past, because nobody can find a proper solution.

And talking about a proper solution: How does Humus's alpha to coverage technique compare to this? Would that not be preferable to these solutions?
Time to get Microsoft to put it in D3D?

Chalnoth

03-Oct-2005, 08:15

1) If they are really doing supersampling than there is very little that you can accelerate in hardware, afaik. And then it makes sense that this new feature would be available to all cards. (and of course they will only introduce it with R520 so that it's introduction will look even better)
It depends upon how the supersampling is done, and where the performance bottleneck is. If the supersampling is done by sending geometry multiple times, then either the vertex traffic or the vertex processing can possibly be a performance bottleneck instead of the pixel processing.

This feature is nice in games where only a very small part of the screen has alpha textures. But you could end up with your entire screen getting supersampling AA, and then you get a 75% performance hit with 4xAA. Is that really a great idea?
Not so important that it's a small part of the screen, but rather that it's a small portion of the total pixel processing that is done on the screen. It's rare that all surfaces need an alpha test.

tEd

03-Oct-2005, 08:20

And talking about a proper solution: How does Humus's alpha to coverage technique compare to this? Would that not be preferable to these solutions?

Well so far quality wise alpha to coverage just can't stack up. There are cases where it works good and looks good and there are case where it looks not good. The main problem are the dithering problem which can occur and objects can get partly transparent when they shouldn't .

Just comapring it in HL2 for instance. With super-sampling you get overall the suprior picture. Better AA on all objects with no artefacts. With alpha to coverage only close range objects get AA'd and every here and there you see the dithering problem.

But i wouldn't throw alpha to coverage just out the window. It's still there and you can still use it if you prefer as it may have a performance advantage.

mjtdevries

03-Oct-2005, 08:33

I guess I get a bit upset when I see Nvidia and ATI spending time on making these kinds of workarounds, while they could also work together with Microsoft in making a proper solution.

As Chalnoth said, it is rare that all surfaces need an alpha test, but even so, there are too many games (imho) where people get a huge performance hit with these solutions.

A while ago people thought that alpha tests wouldn't be used anymore in the future. But it seems that there isn't really any alternative?

ChrisRay

03-Oct-2005, 09:08

At least with the supersampling solution. There are many games which produces huge performance hits with transparency SuperSampling. ((IE WoW) That do fine with transparency multisampling. And transparency multisampling provides a pretty healthy benefit to these games. And its performance hit is negligable at best.

*Edit* I apologise for bringing this up again. But IMO its an often overlooked alternative to transparency Super Sampling and its significantly more usable than people give it credit for. The only game I've really encountered issues with it working all the time was Half Life 2.

tEd

03-Oct-2005, 09:31

That's why i think a having a option for differnt MSAA/SSAA ratio is a good thing for mid/lower end hardware

Sobek

03-Oct-2005, 10:55

It's odd, nothing I do can get it to work on my x850, however, my server with my old 9700 pro enables it first go without a problem. I've tried reinstalling drivers on mine along with ATT and such, no go. (And i'm definitely using the 5.9's on my system, same as my server).

Anyway, while it does work quite nicely, the performance hit is pretty damn big, especially in Vietcong, which uses excessive amounts of foliage :P Went from 62fps to 15fps in one area.

I tried it on Far Cry at 6xAA (what does it equal in Adaptive AA numbers? :P) And it looks absolutely stunning. I can seriously say that the overall quality of the game looked better. That does sound strange but i've always ran Far Cry with just normal 6xAA, and with Adaptive on, everything looks so much crisper and vibrant, albeit at 17fps most of the time :)

It also has this wierd glow on the side of my screen which is always present, but of which i've only noticed in Far Cry :

http://img369.imageshack.us/img369/6900/farcry00108uz.jpg

uwackme

03-Oct-2005, 15:09

I concur on your Farcry observation, including the weird "glow" on the left/top of my 930B LCD.

Most of the veggy's scene's in FCry wasnt getting AA'd with normal 6xAA, now it is and WOW wicked cool looking, but a healthy performance hit.

On multiplayer maps its a huge improvement too...they were always artificially "de-cooled" and alot of the maps were not AA'd... until this tweak. Now the veggy's etf look incredible but it does take a hit.

So I compromised, went with 2xAA, A-AA enabled, and T-AA enabled. This combo gives a great improvement in multiplayer, since normal AA wasn't even applied to most of the maps content ingame, I assume over concerns about feedback lag due to client low performance if they had AA running completely. Somehow this A-AA setting overrides FCry's forced NO-AA in multiplayer mode, and the results are quite awesome.

Richard

03-Oct-2005, 15:45

Can anyone corroborate these results?

http://img359.imageshack.us/img359/1671/cod2aaa4fh.th.png (http://img359.imageshack.us/my.php?image=cod2aaa4fh.png)

I get a 8fps increase to the average by enabling adaptive AA. X850XTPE, Catalyst 5.9.

Geo

03-Oct-2005, 16:07

:shock:

And explain how that could be?

Richard

03-Oct-2005, 16:19

Well, me and wireframe both get an increase in framerate (by 5fps) by enabling dynamic shadows in this game too. This may or maynot be related but if so I can think of two things:

1) The Timedemo output report is completely flawed (which would somewhat surprise me considering the devs are probably using it to profile bottlenecks).
2) There's something very wrong going on with the game's renderer (again, this would surprise me considering there are no rendering problems that I can spot).

Hopefully, none of these are correct but that's why I'm asking for corroboration.

wireframe

03-Oct-2005, 17:30

Well, me and wireframe both get an increase in framerate (by 5fps) by enabling dynamic shadows in this game too.
Did you confirm that this is not the case if you go by the in-engine console average? That is, using the log output in Excel shows an increase in fps when shadows are enabled, but a decrease is reported in the console summary average. (They seem almost exactly reversed for me, which is very peculiar).

I wrote about this is one of the two threads concerning COD 2 demo, but I never saw any reaction to it.

Paul Oblomov

03-Oct-2005, 20:15

Screenshot was taken form CS:S de_port
http://i5.photobucket.com/albums/y162/XOBOT/Snap211.png

But it's quite invisible from huge distance ;)

digitalwanderer

03-Oct-2005, 20:31

Screenshot was taken form CS:S de_port
http://i5.photobucket.com/albums/y162/XOBOT/Snap211.png

But it's quite invisible from huge distance ;)
Is that a zoomed image? :shock:

Paul Oblomov

03-Oct-2005, 22:03

Nope ;)
Just try to crouch on top of tanks when you close to it. . .

Richard

04-Oct-2005, 11:52

Did you confirm that this is not the case if you go by the in-engine console average? That is, using the log output in Excel shows an increase in fps when shadows are enabled, but a decrease is reported in the console summary average. (They seem almost exactly reversed for me, which is very peculiar).

I wrote about this is one of the two threads concerning COD 2 demo, but I never saw any reaction to it.

Yes I can confirm this for AAA too. I get ~2fps drop in the average reported in the console by enabling adaptive AA.