PDA

View Full Version : AquaMark 3 image quality comparison


Pages : [1] 2

MikeC
05-22-04, 05:28 PM
The following information is based on various image quality tests that I ran last night. I really didn't have a plan of attack when I started, but I knew it was going to be related to the Radeon X800 Pro and the stories we've been reading in regards to texture filtering. This series of tests may not lead to anything, but I have some free time this weekend and I felt this would be an interesting experiment.

The weapon I ended up choosing was AquaMark3 (http://www.ati.com/developer/tools.html). Last year Alexander Jorias of Massive Development provided nV News with a key that unlocks all of the features in AquaMark3. One test in the licensed version is a screen capture test. The test accepts a user-supplied frame interval setting and screenshots are taken at the interval provided (for example, take a screenshot every 10, 50, 100, 500, or 1000 frames). AquaMark3 also has application control over antialiasing and anisotropic texture filtering.

All of the graphics options in AquaMark3 were configured to their maximum setting and I adjusted the level of texture filtering for each test. I selected a number of frames that are good examples to use for conducting image quality comparisons.

I then used the The Compressionator (http://www.ati.com/developer/tools.html) tool to compare the screenshots from AquaMark3, which I converted from TGA to PNG format. The screenshots were not altered in any other way. The Compressionator can be used to identify the differences between two images. In fact, you can download and install The Compressionator and use the images from my tests to run comparisons.

The example below is a comparison of trilinear filtering on the left and 2X anisotropic filtering on the right. The middle section shows the differences, which are often subtle and requires the difference brightness level to be adjusted before the differences become visible. In this example, the difference brightness was increased to 400%. The Compressionator also has a zoom in/out feature.

http://www.nvnews.net/images/aquamark3/compressionator.jpg

The Compressionator is also able to quantify the differences as it provides the following statistics for a comparison.

http://www.nvnews.net/images/aquamark3/error_statistics.png

Although I'm not familiar with the measurements that take place, I do know that similar images have a lower error deviation than images that are vastly different. Taking all of the information, I developed the following chart, which lists the mean square error for each image quality comparison.

The mean square error of 2.52 from the example above was based on a comparison of trilinear filtering and 2X anisotropic filtering from frame 300.

http://www.nvnews.net/images/aquamark3/gf6800u_60.72_filtering.png

Below are the links to the various AquaMark3 screenshots, which range in size from 700 KB to 1.2 MB. Increase the brightness on your monitor if they are too dark. I hope to install the Radeon X800 Pro in one of my other PCs this evening and conduct the same tests.

Cheers,
MikeC


Frame 200:

http://www.nvnews.net/images/aquamark3/aquamark3_fr0200.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr0200.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr0200.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr0200.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr0200.png)

Frame 300:

http://www.nvnews.net/images/aquamark3/aquamark3_fr0300.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr0300.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr0300.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr0300.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr0300.png)

Frame 900:

http://www.nvnews.net/images/aquamark3/aquamark3_fr0900.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr0900.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr0900.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr0900.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr0900.png)

Frame 1700:

http://www.nvnews.net/images/aquamark3/aquamark3_fr1700.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr1700.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr1700.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr1700.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr1700.png)

Frame 2400:

http://www.nvnews.net/images/aquamark3/aquamark3_fr2400.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr2400.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr2400.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr2400.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr2400.png)

Frame 2500:

http://www.nvnews.net/images/aquamark3/aquamark3_fr2500.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr2500.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr2500.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr2500.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr2500.png)

Frame 2600:

http://www.nvnews.net/images/aquamark3/aquamark3_fr2600.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr2600.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr2600.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr2600.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr2600.png)

Frame 3500:

http://www.nvnews.net/images/aquamark3/aquamark3_fr3500.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr3500.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr3500.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr3500.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr3500.png)

Frame 3900:

http://www.nvnews.net/images/aquamark3/aquamark3_fr3900.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr3900.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr3900.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr3900.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr3900.png)

Frame 4000:

http://www.nvnews.net/images/aquamark3/aquamark3_fr4000.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr4000.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr4000.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr4000.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr4000.png)

Frame 4300:

http://www.nvnews.net/images/aquamark3/aquamark3_fr4300.jpg

Trilinear (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_trilinear/fr4300.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_2xaf/fr4300.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_4xaf/fr4300.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/gf6800u_60.72_noaa_8xaf/fr4300.png)

Cota
05-22-04, 06:04 PM
can you include 16x aniso?

Jarred
05-22-04, 11:44 PM
What card was used in this test?

Clay
05-23-04, 12:02 AM
What card was used in this test?
Check out his last chart dood. :p

This is cool Mike. Once you have the X800Pro resulsts we should learn a few things.

MikeC
05-23-04, 12:16 AM
What card was used in this test?

The GeForce 6800 Ultra with the 60.72 beta drivers.

I've made good progress and have finished the Radeon X800 Pro screen captures and compiled the results from The Compressonator comparison. I ran across some significant differences in the Compressionator results between both cards that I need to study further.

I'm also going to run a few additional tests with the 60.72 drivers before I test the 61.11 drivers. I'm getting similar results from the 60.72 drivers in AquaMark3 that I was getting in our GeForce 6800 Ultra preview. Changing the driver image setting from quality to high quality and/or enabling trilinear optimizations has no affect on image quality.

MikeC
05-23-04, 12:24 AM
can you include 16x aniso?

Maybe in a follow-up, but 16X aniso had literally no effect on image quality in these tests. If you examine The Compressonator chart results, you'll find that even moving from 4X to 8X doesn't significantly enhance image quality.

The largest improvement always occurs when moving from trilinear to 2X anisotropic. This particular experiment is beginning to show that 8X and 16X AF are overkill based on the texture filtering scheme each GPU employs.

ChrisRay
05-23-04, 12:53 AM
I like it alot MikeC, you never cease to impress me at times.

Jarred
05-23-04, 01:21 AM
word!, do you plan on testing some unreal with Ati's trilinear filtering? I'm really curious to see what the results would be...

Jarred
05-23-04, 01:21 AM
Check out his last chart dood. :p

This is cool Mike. Once you have the X800Pro resulsts we should learn a few things.

did you REALLY just spell dude "dood" !!? ? :)

mikechai
05-23-04, 04:10 AM
This is a interesting test. Can't wait for the update.

MikeC
05-23-04, 03:19 PM
I've run into a slight glitch :)

The frame rate is displayed in the upper right corner of the AM3 screen capture, which can change from frame to frame. Not sure if this feature can be disabled, but The Compressonator is picking up the differences in the frame rate, which is skewing the results. I'm going to crop the screen captures and remove the frame rate using IfranViews's batch conversion routine and adjust the results accordingly.

Blacklash
05-23-04, 03:49 PM
Most of the reviews I have seen like Ati with no AF and x2, at x4 they consider them a dead heat, and some say x8 as well, others dead heat there too, now x16 I have seen most say slight edge to Nvidia. This will be an interesting test. Good post.

Razor04
05-23-04, 09:37 PM
Umm I am not trying to be rude or unappreciative of the work MikeC has put into this but what exactly is the point of all this? The chart shows the difference values between the different modes...this means nothing in my eyes. Of course Trilinear Filtering will be different than Anisotropic Filtering and so on. The number just shows how different the two screens are.

I don't see how these numbers will be of any significance when the X800 is added as the two cards use different algorithms which will produce different results. There is no correlation between the two different cards with those numbers unless you are comparing to a refrast image there is no true baseline for comparison. The only way those numbers would be valid for comparing is if you get a number by comparing refrast to a 6800 running Trilinear and then you get a number by comparing refrast to a X800 running Trilinear. This type of comparison would clearly illustrate the differences between the various IHV implementations.

The biggest difference seen is between Trilinear and any Anisotropic mode and I don't think any of you will argue that the Anisotropic modes are much sharper and overall more pleasing to the eye. I appreciate the work that MikeC has done especially with regards to the pictures as they show the differences between the modes quite well.

Ruined
05-23-04, 09:44 PM
I don't see how these numbers will be of any significance when the X800 is added as the two cards use different algorithms which will produce different results. There is no correlation between the two different cards with those numbers unless you are comparing to a refrast image there is no true baseline for comparison. The only way those numbers would be valid for comparing is if you get a number by comparing refrast to a 6800 running Trilinear and then you get a number by comparing refrast to a X800 running Trilinear. This type of comparison would clearly illustrate the differences between the various IHV implementations.

You're supposed to wait for the damage before you do damage control. :)

Clay
05-23-04, 09:52 PM
Umm I am not trying to be rude or unappreciative of the work MikeC has put into this but what exactly is the point of all this? The chart shows the difference values between the different modes...this means nothing in my eyes. Of course Trilinear Filtering will be different than Anisotropic Filtering and so on. The number just shows how different the two screens are.Did you read the first paragraph of Mike's initial post? (see below) :confused:

The following information is based on various image quality tests that I ran last night. I really didn't have a plan of attack when I started, but I knew it was going to be related to the Radeon X800 Pro and the stories we've been reading in regards to texture filtering. This series of tests may not lead to anything, but I have some free time this weekend and I felt this would be an interesting experiment.Your opinion is fine and I'm sure others will echo it. However, many others will find this to be very interesting and potetially telling. Mike has a history of being very thorough and professional regarding things of this nature. So, let's just all sit back and wait for the X800 Pro results before we start questioning anything.

Rytr
05-23-04, 09:58 PM
I will second that Max! Should be interesting if nothing else. Kudos to MikeC for taking the time to do some experimenting and sharing the results.

noko
05-23-04, 10:04 PM
These tests are a valid way to determine how much difference a frame has going from one level of filtering to another. Now if the difference is very little going from 8x to 16x but yet the performance hit is rather large then you could say 16x is going overboard in a quatitative way for a very little real difference on the screen. I am wondering if the LOD changes with the different levels of anisotropic? Doesn't look that way with the 6800 but the X800 Pro may adjust LOD much more skewing the results. Something to consider.

Razor04
05-24-04, 08:18 AM
You're supposed to wait for the damage before you do damage control. :)
Ok you know what...attitudes like this around here are seriously starting to tick me off. I wasn't trying to do any sort of damage control but instead bring up the issue of how those numbers relate to anything. I have no problem whatsoever with the pics and think they are great but the minute a second set of numbers goes up for the X800 it will become a numbers war between the fanboy clans. Some will be shouting that higher numbers are better while others will be shouting that lower numbers are better when the numbers are only representative for comparison between different modes on one card.


Did you read the first paragraph of Mike's initial post? (see below) :confused:
So I am not allowed to comment on anything around here anymore? I can't add my opinion or view with regards to one aspect of a post? Is that wrong?


Your opinion is fine and I'm sure others will echo it. However, many others will find this to be very interesting and potetially telling. Mike has a history of being very thorough and professional regarding things of this nature. So, let's just all sit back and wait for the X800 Pro results before we start questioning anything.
I know he has a history of being very thorough and professional but I just think this is going to lead to misconceptions and all for both companies. Something about that whole numbers portion of the post just doesn't seem right to me and I wish you guys wouldn't be so harsh when dealing with constructive criticism. I believe that I was quite objective with regards to my first post in this thread and tried to avoid limiting my post to any one IHV in this post as those damn numbers are going to cause both to get nipped in the butt.

Clay
05-24-04, 08:56 AM
Ok you know what...attitudes like this around here are seriously starting to tick me off. I wasn't trying to do any sort of damage control but instead bring up the issue of how those numbers relate to anything. I have no problem whatsoever with the pics and think they are great but the minute a second set of numbers goes up for the X800 it will become a numbers war between the fanboy clans. Some will be shouting that higher numbers are better while others will be shouting that lower numbers are better when the numbers are only representative for comparison between different modes on one card.If we worried about how fanboys would react then we'd never do anything around here. :) We cannot control how people may perceive/twist things towards their own bias. I have faith in the overwhelming majority of our readership to be able to read and comprehend correctly though.

So I am not allowed to comment on anything around here anymore? I can't add my opinion or view with regards to one aspect of a post? Is that wrong?Err, no...noone ever suggested that you are not allowed to comment around here anymore or that you cannot add your opinion. You've gotta realize that your post was very premature as we don't even have the X800 Pro results as of yet. That's all I was saying, just sit back a wait a bit before passing judgement.


I know he has a history of being very thorough and professional but I just think this is going to lead to misconceptions and all for both companies. Something about that whole numbers portion of the post just doesn't seem right to me and I wish you guys wouldn't be so harsh when dealing with constructive criticism.Again, why not give it a chance? Seems to me that you're just jumping the gun a bit. Harsh? Oh come on now, no one is being harsh. I even said that your opinion was fine, just seems premature. It is very, very tame here compared to many other forums.


I believe that I was quite objective with regards to my first post in this thread and tried to avoid limiting my post to any one IHV in this post as those damn numbers are going to cause both to get nipped in the butt.Sure, I thought you were fairly objective too. What does "nipped in the butt" mean though? I've heard the cliche "nipped in the bud"..but that doesn't apply in the context of your sentence.

Razor04
05-24-04, 09:15 AM
If we worried about how fanboys would react then we'd never do anything around here. :) We cannot control how people may perceive/twist things towards their own bias. I have faith in the overwhelming majority of our readership to be able to read and comprehend correctly though.
I agree that if all you worried about was fanboys then nothing would get done around here. I think a lot of people though will have difficulty understanding what is going on with those numbers though. Perhaps a better way of showing the results would have been as percent differences. Instead of .02 difference you have a 2% difference and so on. I just picked some arbitrary values but you get the idea. Something like this will probably be easier for a layman to understand.


Err, no...noone ever suggested that you are not allowed to comment around here anymore or that you cannot add your opinion. You've gotta realize that your post was very premature as we don't even have the X800 Pro results as of yet. That's all I was saying, just sit back a wait a bit before passing judgement.
There is nothing wrong with being forward looking and determining possible problems. There has already been at least one post in this thread that has inferred that there will be some sort of correlation between the 6800 numbers and the X800 numbers. This is the only real thing that worries me about the whole post. Like I said I love the idea of showing the differences and showing the pictures but having the results in the same thread could lead to disaster...but I will hold off on that for now until they are out.


Sure, I thought you were fairly objective too. What does "nipped in the butt" mean though? I've heard the cliche "nipped in the bud"..but that doesn't apply in the context of your sentence.
I have heard both sayings being used. I think a lot of it depends on where you are in the country. There are a lot of other phrases that are slightly different based on where you are or where you grew up too.

Clay
05-24-04, 09:43 AM
I agree that if all you worried about was fanboys then nothing would get done around here. I think a lot of people though will have difficulty understanding what is going on with those numbers though. Perhaps a better way of showing the results would have been as percent differences. Instead of .02 difference you have a 2% difference and so on. I just picked some arbitrary values but you get the idea. Something like this will probably be easier for a layman to understand.Percentages is a good idea.


I have heard both sayings being used. I think a lot of it depends on where you are in the country. There are a lot of other phrases that are slightly different based on where you are or where you grew up too.Yeah, I know of some colloquialisms that are real doozies from around where I grew up. :)

MikeC
05-24-04, 07:15 PM
Sorry about taking so long to get the comparative images posted, but we lost power last night after a thunderstorm hit the area. I appreciate the feedback and I think jumped the gun a bit with the TheCompressionator comparisons.

Using percentages to show the relative differences in texture filtering is a great idea. Unfortunately, the full size images from AquaMark 3 are not good candidates for comparisons since most of the frames contain vegetation and other objects that are dynamically generated from run-to-run.

Before the power died, I was copying sections from each frame that showed noticeable difference in the amount of texture filtering that was being applied. This his was accomplished by a visual inspection with my eyeballs positioned about 9 inches from the monitor :wtf:

I'll eventually clean up my original post, but I wanted to give you a look at what I've got at this point. One of the reasons that AquaMark is effective for this type of comparison is due to the terrain - we all know how irregular shaped objects can cause havoc on texture filtering algorithms :)

Each image is 520x520 pixels and around 350-400 KB.

Frame 300:

GeForce 6800 Ultra - Bilinear (http://www.nvnews.net/images/aquamark3/frame_0300/6800u_60.72_0xaa_bi.png) -Trilinear (http://www.nvnews.net/images/aquamark3/frame_0300/6800u_60.72_0xaa_tri.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0300/6800u_60.72_0xaa_2xaf.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0300/6800u_60.72_0xaa_4xaf.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0300/6800u_60.72_0xaa_8xaf.png)

Radeon X800 Pro - Bilinear (http://www.nvnews.net/images/aquamark3/frame_0300/x800pro_4.5_0xaa_bi.png) - Trilinear (http://www.nvnews.net/images/aquamark3/frame_0300/x800pro_4.5_0xaa_tri.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0300/x800pro_4.5_0xaa_2xaf.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0300/x800pro_4.5_0xaa_4xaf.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0300/x800pro_4.5_0xaa_8xaf.png)

Frame 900:

GeForce 6800 Ultra - Trilinear (http://www.nvnews.net/images/aquamark3/frame_0900/6800u_60.72_0xaa_tri.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0900/6800u_60.72_0xaa_2xaf.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0900/6800u_60.72_0xaa_4xaf.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0900/6800u_60.72_0xaa_8xaf.png)

Radeon X800 Pro - Trilinear (http://www.nvnews.net/images/aquamark3/frame_0900/x800pro_4.5_0xaa_tri.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0900/x800pro_4.5_0xaa_2xaf.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0900/x800pro_4.5_0xaa_4xaf.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_0900/x800pro_4.5_0xaa_8xaf.png)

Frame 2400:

GeForce 6800 Ultra - Trilinear (http://www.nvnews.net/images/aquamark3/frame_2400/6800u_60.72_0xaa_tri.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_2400/6800u_60.72_0xaa_2xaf.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_2400/6800u_60.72_0xaa_4xaf.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_2400/6800u_60.72_0xaa_8xaf.png)

Radeon X800 Pro - Trilinear (http://www.nvnews.net/images/aquamark3/frame_2400/x800pro_4.5_0xaa_tri.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_2400/x800pro_4.5_0xaa_2xaf.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_2400/x800pro_4.5_0xaa_4xaf.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_2400/x800pro_4.5_0xaa_8xaf.png)

Frame 3500:

GeForce 6800 Ultra - Trilinear (http://www.nvnews.net/images/aquamark3/frame_3500/6800u_60.72_0xaa_tri.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_3500/6800u_60.72_0xaa_2xaf.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_3500/6800u_60.72_0xaa_4xaf.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_3500/6800u_60.72_0xaa_8xaf.png)

Radeon X800 Pro - Trilinear (http://www.nvnews.net/images/aquamark3/frame_3500/x800pro_4.5_0xaa_tri.png) - 2X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_3500/x800pro_4.5_0xaa_2xaf.png) - 4X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_3500/x800pro_4.5_0xaa_4xaf.png) - 8X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_3500/x800pro_4.5_0xaa_8xaf.png)

Frame 4300:

GeForce 6800 Ultra - Bilinear (http://www.nvnews.net/images/aquamark3/frame_4300/6800u_60.72_0xaa_bi.png) -Trilinear (http://www.nvnews.net/images/aquamark3/frame_4300/6800u_60.72_0xaa_tri.png) -2X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_4300/6800u_60.72_0xaa_2xaf.png) -4X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_4300/6800u_60.72_0xaa_4xaf.png) -8X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_4300/6800u_60.72_0xaa_8xaf.png)

Radeon X800 Pro - Bilinear (http://www.nvnews.net/images/aquamark3/frame_4300/x800pro_4.5_0xaa_bi.png) -Trilinear (http://www.nvnews.net/images/aquamark3/frame_4300/x800pro_4.5_0xaa_tri.png) -2X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_4300/x800pro_4.5_0xaa_2xaf.png) -4X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_4300/x800pro_4.5_0xaa_4xaf.png) -8X Anisotropic (http://www.nvnews.net/images/aquamark3/frame_4300/x800pro_4.5_0xaa_8xaf.png)


Keep in mind that NVIDIA has yet to debut the offical driver release for the GeForce 6 series.

Blacklash
05-24-04, 07:37 PM
Might want to go see this thread at Rage3d. As I have said before, nothing is free...

http://www.rage3d.com/board/showthread.php?s=&threadid=33760559

What I got out of it:

Ati/Nvidia optimized trilinear give almost the same results.
Ati optimizations can not be switched off, Nvidia can.
AF "dithering" and "moire" can occur in the Ati images because of "underfiltering". They use the DX rec of 5bit weight while Nvidia uses the OGL of 8bit which is higher quality/sharpness.

My added thoughts :To be fair to Ati, in AF you can use a third party application like Rtool which will force trilinear filtering through all stages of AF and defeat these artifacts, however if you are getting 45FPS in game, it will drop to something like 39. < for a rough example. IQ and performance will always have a positive negative correlation.

Nothing is free, this engineering, not magic.

ntxawg
05-25-04, 12:55 PM
aside from tri for frame 300 and 600 there isnt much of a difference from i can see. are the optimiztion on for the 6800 cuz they look kinda blurry vs x800's tri in those 2 frames?

MikeC
05-25-04, 09:48 PM
aside from tri for frame 300 and 600 there isnt much of a difference from i can see. are the optimiztion on for the 6800 cuz they look kinda blurry vs x800's tri in those 2 frames?

All the GeForce shots are with trilinear optimizations off. I ran a test of frame 300 with trilinear optimizations on and couldn't spot any differences.

With The Compressionator, the difference in trilinear filtering with optimizations on and off was 0.58. With 2X AF and 4X AF the difference was 0.04 and there were no differences with 8X AF.