ATi is ch**t**g in Filtering

Chade said:
If all it takes it to look at things at a certain angle to degrade the filtering that much (in comparison to the 9800) that would not be very 'intelligent' at all would it..
With trilinear filtering, it doesn't make much of an angle change to reduce the mipmap level used. Just load up Quake3 with trilinear filtering and jump while looking at different areas of the floor. You'll see blur/unblur very easily.

ATI's (and NVidia's) optimization can definately reduce quality without aniso, but once aniso is applied, it's very tough to see the difference. When ATI was justifying bilinear AF back in the 8500/9000 days, they were half right. Mip-map boundaries are far less visible with AF, so this optimization is pretty good.

The biggest problem with NVidia's optimizations in NV3x was they also reduced the degree of AF when they wanted to, like here or adjusting where the first mip-map transition is in "balanced" mode. I thought the idea of just blending the edges of mip-maps was pretty good, and was hoping for someone to do it since I learned the real cost of trilinear filtering in games while at ATI.
 
Re: :/

FUDie said:
Chade said:
Looks like no aniso enabled on the X800 at all. Real good "test case".

-FUDie

We are getting to the bottom of this at the moment. The idea is all the cards have no af so its bilinear vs ATIs new filtering vs trilinear. The X800 and 6800 ultra shots are from driver heaven and the X800 looks like its only doing bilinear

Basically currently we believe those the 9800 shots which were used for comparison are actually invalid as they have af enabled. Most of the thread is speculation rather than actual facts. Thug is hopefully going to get a shot of trilinear on his X800 soon, although it has already been shown that the 9600 doesnt produce such results so its probably a mistake on driver heavens part.
 
nAo said:
It does contain a little error in their intermediate log2 approximation step, I hope it's just a typo and it doesn't invalid their calculations, the final approximation step is correct though.
Good catch :) I usually don't bother to check their calculations... guess this should convince me to start again ;)

Interesting read though. Thanks for the link.
 
nAo said:
It does contain a little error in their intermediate log2 approximation step, I hope it's just a typo and it doesn't invalid their calculations, the final approximation step is correct though.
Fixed, thanks to eagle eye :)
 
Talking about sample in motion... I made a short video of a FarCry "scene", with a 9600Pro. As the same adaptive filtering seems to be used in the RV350, maybe this could be of some interest.

The video is an AVI about 6 seconds long; the codec used is the lossless Uffyuv. The file size is obviously uge, around 75MB.

But I'm trying to host it via BitTorrent, so... if you find that it download slowly, keep the faith and let your BitTorrent client downloadit. The more downloaders, the faster it should be.
And to those on a big fat pipe :) : let the BitTorrent client open / seed the torrent when you have finished, so others can download it faster.

Here are the relevant links:

Huffyuv codec:
http://www.avisynth.org/index.php?page=HuffYUV

BitTorrent clients:
http://bitconjurer.org/BitTorrent/
http://azureus.sourceforge.net/ (Java multiplatform)

Torrent for the video:
http://mark0.ngi.it/var/FarCry-RV350-AF16-MaxQ.avi.torrent

Bye!
 
The trouble is, that, as i see it, rv3x0 and r420 are using _different_ tlf optimisation scheme, at least in the software part of those. I came to this conclusion since i don't have any filtering problems on my r9600 in Enclave and Halo (well, besides the >0 stage problems).
 
Maybe this can serve as a "hurry up" for someone with an R420, to produce and share a meaningful video that can be evaluated.

If all continue taling about moving samples, and noone provide one... :D

Bye!
 
OK, now download should be A LOT faster (over 50KB/sec., there are 3-4 seeders).

Also, suggestions are accepted for more rappresentative scenes/video! :)

Bye!
 
3 seeders active in this moment, so bandwidth is not an issue.
This is a minishot to show the scene I choosed.
I'm sure someone could choose a better one so... do it! :)

FarCry-RV350-AF16-MaxQ.jpg


Bye!
 
What are the drivers used ? I have it seeded now too . But i am looking to reproduce it which i haven't been able to on both the 9600 and x800pro .
 
jvd said:
What are the drivers used?
Catalyst 4.5.

Within some minutes I'll shut down the tracker (have to turn off the PC, and the go to the mountains for the weekend! :D ).
10 peoples downloaded it. Near 1GB was served without using any webspace! :)
So, for anyone who have the possibility to make a video on a X800, bandwidth or limited webspace available are not an excuse: BitTorrent is the way.

If someone want to recreate the torrent and host on another tracker (if he think it's worth the effort), you are welcome.

Bye!
 
Mark0 said:
jvd said:
What are the drivers used?
Catalyst 4.5.

Within some minutes I'll shut down the tracker (have to turn off the PC, and the go to the mountains for the weekend! :D ).
10 peoples downloaded it. Near 1GB was served without using any webspace! :)
So, for anyone who have the possibility to make a video on a X800, bandwidth or limited webspace available are not an excuse: BitTorrent is the way.

If someone want to recreate the torrent and host on another tracker (if he think it's worth the effort), you are welcome.

Bye!

What map is this from . I want to make one with my x800pro and see if its the same problem .
 
Evildeus said:
jvd said:
What map is this from . I want to make one with my x800pro and see if its the same problem .
So there's a visible issue? At least on 9600s?
I'm only 3 boards into farcry and I have yet to see any issue. Except in the halo demo i have. But i shudder to think about what that game is doing and how the demo would look on any card.


I still have yet to see any problems. I'm trying to post a video but i'm having problems www.ooakalabeth.com/video and get the 5 meg video. Its currently being uploaded
 
jvd said:
Evildeus said:
jvd said:
What map is this from . I want to make one with my x800pro and see if its the same problem .
So there's a visible issue? At least on 9600s?
I'm only 3 boards into farcry and I have yet to see any issue. Except in the halo demo i have. But i shudder to think about what that game is doing and how the demo would look on any card.


I still have yet to see any problems. I'm trying to post a video but i'm having problems www.ooakalabeth.com/video and get the 5 meg video. Its currently being uploaded
My question is. Did you DL the video (i didn't not enough BW)? If yes, did you see what Marko points out as you seem to admit in your answer that i quote?

For the rest, i know what is your stance.
 
Back
Top