View Full Version : Is Hexus.net Blind ?
Look at this page and compare the two screen shots. Notice the difference in details on the back wall and elsewhere.
http://www.hexus.net/content/reviews/review.php?dXJsX3Jldmlld19JRD02NDkmdXJsX3BhZ2U9MTI =
Look at these comments with regards to the screen shots.
If there's an obvious difference between the two, I can't see it. There's more final texture detail in the wall textures in the ATI shot, due to the viewer being slightly closer.
In fact, it is the opposite, the ATI screen shot is from a further distance and it still has noiticeably better detail.
Agreed, a particularly egregious oversight.
FYI, as nelg forgot the link, he was talking about the SS2 screenshots (in the IQ Comparison section, IIRC).
:oops: :oops: Added the link.
nonamer
24-Oct-2003, 19:45
That's hard to believe he could've missed that. Though it could be that he's got the wrong images to compare (they seem to be at difference AA levels, look at cross hair).
EDIT: Wow, I've notice something weird about the stairs: they're not the same number of stairs in both images. Then I realized that one's more zoomed in than the other. Hexus.com is either more blind than it first seems or got something mixed up or they're really sloppy here.
Why are you surprised? Hexus was always the most ridiculously biased site, sometimes even against the real world realities. Business as usual: I'm pretty sure it's worth it for them, in terms of income...
EDIT: grammar
Look at the big head on the left.
And he can see no differences :roll:
AlphaWolf
24-Oct-2003, 21:19
Think he means specifically in regards to the texture detail of the floor?
What you are looking for is obvious stepped banding between the textures on the floor, as the view goes away from you. If the texture stages don't blend together well, you'll notice it, indicating a poor quality trilinear filter.
/shrug the ATI shot is definitely a more detailed clearer image overall.
THe_KELRaTH
24-Oct-2003, 21:21
Hell.. just look at the plants!
There is a problem with this screenshot because if you take them exactly from the same place with the same aniso setting you get exactly the same level of detail on the back wall with NVIDIA and ATI.
nonamer
24-Oct-2003, 23:30
Hell.. just look at the plants!
The plant is off the screen in the nVidia, which brings us to another problem: the pictures are not taken in the same place.
Chris123234
25-Oct-2003, 00:58
lol he justifies ati detail being better because its closer...but the nvidia is actually closer. :lol: :roll: .
Or did i mix them up?
Ok, I'm probably going to start a war with this but...
I think the first image looks better. Although it looks more blurry than the second image, I think this makes it smoother in a positive way. After all, we're looking at a still image, and you can't see temporal aliasing on that.
What I'm trying to say is: you can totally turn off mipmapping and use the highest detail texture, and the still image will look fabulously sharp (unless you're looking at a chess board where you see spacial aliasing). But when you start moving the camera, the pixel colors start jumping around which is very ugly. I'm not claiming this is happening in the second image, I just think you can't base an opinion on still images. And for the sake of gameplay, I'd rather have some overblur than temporal aliasing. But in the end, actually everybody should test both cards themselves and decide for themselves.
Don't forget, Nyquist's limit is still 2...
OpenGL guy
25-Oct-2003, 02:00
Ok, I'm probably going to start a war with this but...
I think the first image looks better. Although it looks more blurry than the second image, I think this makes it smoother in a positive way. After all, we're looking at a still image, and you can't see temporal aliasing on that.
And you shouldn't assume that there is aliasing based on a still image.
But in the end, actually everybody should test both cards themselves and decide for themselves.
Shouldn't the reviewer be doing this? The average consumer can't go out and buy one of each board to test this sort of stuff. In fact, the average consumer has no idea what any of this means.
And you shouldn't assume that there is aliasing based on a still image.
I didn't assume that. I just wanted to point out that sharpness of a still image isn't necessarily a good thing.
Shouldn't the reviewer be doing this? The average consumer can't go out and buy one of each board to test this sort of stuff. In fact, the average consumer has no idea what any of this means.
Unfortunately, often even the reviewer doesn't have an idea what this means. You can compare sharpness of a still image and you can show it on the internet with an uncompressed image format, but you can't say much about temporal aliasing. Actually... we could theoretically do some flickering analysis. Compare every frame with a software generated image with extreme supersampling. A constant, converging difference with the reference images (like with overblur) is not as bad as when it's contantly alternating. Of course practically there would be some problems...
Anyway, you have two kinds of consumers: those who really care, and those who don't. I think the "average consumer" doesn't care much and will buy either card as long as the first impression isn't horrible. For those people who do care, I'm afraid there is no other method yet than to compare both cards themselves and pay attention to what really matters to them.
Shouldn't the reviewer be doing this? The average consumer can't go out and buy one of each board to test this sort of stuff. In fact, the average consumer has no idea what any of this means.
Yes but as Nick alluded to not many can or will.
Please talk to your marketing department about something along the lines of this Andy, why don’t you talk to your marketing department about putting together a video demonstrating the issues that you raised? I would assume that a large number of people would be interested in having as much information as possible about it. Maybe a whole presentation on image quality. Show examples of different filtering, precision, AA etc. include it on the driver cd that comes with ATI’s cards and host it on the ATI web site. Arm people with enough information about image quality issues and then maybe the slippery slope that you alluded to could be avoided.
(my post to Andypski)
The funny thing is that the average consumer would very likely have the impression the nV is the leading IHV with respect to I.Q. In this area I think ATI's marketing really should be putting its best foot forward.
The review states that viewer distance is different, so the images wont be identical. I didn't have time to make sure I was standing in exactly the same spot in each shot, apologies for that, but the review states that's the case, on the IQ page.
I was also specifically talking about the trilinear filter being applied to the ground texture to show up mip level banding as the texture is drawn further and further away. Again, the review text states that.
I was trying to highlight the fact that the trilinear filter isn't too bad, despite being wrong.
Anything else in the images, peripheral to the ground texture, I don't care about for the purposes of the screenshots. Again, if you read the review I state that clearly.
The texture detail on the wall and stairs is different due to viewer distance. *sigh*, I state all that in the article text. Yeah I get it wrong by stating the ATI picture is closer than the NV one, that's incorrect, but the reason for differences in the shots are still explained the same way, viewer distance is different.
I'll try and take a 2nd load of shots for examination, see what you think. Or maybe Dave can, I know he's got FX5700 Ultra, I'd be interested in his take on the trilinear issue too.
Rys
EDIT:
Here's a pair of images, uncompressed .tga, a selection mask from frame 350 of the Nature test in 3DMark03. Forced (not application set) control panel 4xAA and 8xAF with the driver 'Quality' slider set to maximum.
I chose frame 350 since it shows mip levels, some aliased edges on the stones to compare AA with, and the sand texture shows up nicely when aniso filtered. Beware, these images are 1.2MB each, due to being uncompressed. I concede the SS2 review images could have been more accurate (viewer distance) but I just simply didn't have time to do the IQ part of the review with any detail.
CAT3.8, 9800XT (http://img.hexus.net/nv36nv38/iq/ATI_CAT38_Nature350_4AA8AF.tga)
52.16, FX5950 Ultra (http://img.hexus.net/nv36nv38/iq/NV_5216_Nature350_4AA8AF.tga)
I'd post my D3D AF tester shots for 52.16, but they are identical to what 3DCenter has for 52.14. The driver is definitely doing wrong trilinear (what I'm most interested in). Please look at the two .tga's if you have the time and bandwidth.
To me, using those screenshots (and the UT2003 shots I took for the review), along my own experience testing both boards using the settings above, that while trilinear is being performed incorrectly and IQ is indeed worse, compared to CAT3.8 at equivalent settings, it's really hard to spot. Anti aliasing quality (at 4x) appears to be equivalent to my eyes, and NVIDIA's aniso filter looks good to me too.
Maybe I've just spent too long staring at the screen to see anything now, but the reason I've not completed my IQ article is that I'm honestly struggling to find differences to talk about, other than what 3DCenter has covered already with their 52.14 article.
Why are you surprised? Hexus was always the most ridiculously biased site, sometimes even against the real world realities. Business as usual: I'm pretty sure it's worth it for them, in terms of income...
EDIT: grammar
Please, show me our rediculous bias, because I'd love to see examples of it. I've been writing there for 3 years and I've yet to be biased towards anything, at least not conciously. So be my guest, show me our bias, mine in particular (I'm not the only one who writes there) :)
As for income, I've personally yet to make a penny from HEXUS. I currently don't get paid for writing there. I get to play with nice hardware when it comes in for review, but I'm definitely not earning a living there. Thanks for your completely unfounded speculation, it's utterly entertaining.
Rys
Althornin
26-Oct-2003, 06:40
The review states that viewer distance is different, so the images wont be identical. I didn't have time to make sure I was standing in exactly the same spot in each shot, apologies for that, but the review states that's the case, on the IQ page.
I was also specifically talking about the trilinear filter being applied to the ground texture to show up mip level banding as the texture is drawn further and further away. Again, the review text states that.
I was trying to highlight the fact that the trilinear filter isn't too bad, despite being wrong.
Anything else in the images, peripheral to the ground texture, I don't care about for the purposes of the screenshots. Again, if you read the review I state that clearly.
The texture detail on the wall and stairs is different due to viewer distance. *sigh*, I state all that in the article text. Yeah I get it wrong by stating the ATI picture is closer than the NV one, that's incorrect, but the reason for differences in the shots are still explained the same way, viewer distance is different.
I'll try and take a 2nd load of shots for examination, see what you think. Or maybe Dave can, I know he's got FX5700 Ultra, I'd be interested in his take on the trilinear issue too.
Rys
EDIT:
Here's a pair of images, uncompressed .tga, a selection mask from frame 350 of the Nature test in 3DMark03. Forced (not application set) control panel 4xAA and 8xAF with the driver 'Quality' slider set to maximum.
I chose frame 350 since it shows mip levels, some aliased edges on the stones to compare AA with, and the sand texture shows up nicely when aniso filtered. Beware, these images are 1.2MB each, due to being uncompressed. I concede the SS2 review images could have been more accurate (viewer distance) but I just simply didn't have time to do the IQ part of the review with any detail.
CAT3.8, 9800XT (http://img.hexus.net/nv36nv38/iq/ATI_CAT38_Nature350_4AA8AF.tga)
52.16, FX5950 Ultra (http://img.hexus.net/nv36nv38/iq/NV_5216_Nature350_4AA8AF.tga)
I'd post my D3D AF tester shots for 52.16, but they are identical to what 3DCenter has for 52.14. The driver is definitely doing wrong trilinear (what I'm most interested in). Please look at the two .tga's if you have the time and bandwidth.
To me, using those screenshots (and the UT2003 shots I took for the review), along my own experience testing both boards using the settings above, that while trilinear is being performed incorrectly and IQ is indeed worse, compared to CAT3.8 at equivalent settings, it's really hard to spot. Anti aliasing quality (at 4x) appears to be equivalent to my eyes, and NVIDIA's aniso filter looks good to me too.
Maybe I've just spent too long staring at the screen to see anything now, but the reason I've not completed my IQ article is that I'm honestly struggling to find differences to talk about, other than what 3DCenter has covered already with their 52.14 article.
can you cut the melodrama and sighs and stop trying to make your mistake look like less of a mistake?
If there's an obvious difference between the two, I can't see it. There's more final texture detail in the wall textures in the ATI shot, due to the viewer being slightly closer.
That is incorrect, and was the only point of this thread.
*sigh*
The thing is, its becoming more and more apparent that many reviewers just cannot seem to understand things. For instance, you compare framerates between FX cards and Radeons in Serious Sam, but make no mention that ATI offers more detail...i honestly wonder if the loss of detail in the FX screencaps gives them an unfair performance advantage. I'd say yes.
Just as a quick note, it is pretty easy to make a saved game place in SS2 to take screenshots from the same place at. I always made one in the Technology Demo map to use for that, the long walkway between both ends of the 3d demo buildings made a good filtering test.
Doomtrooper
26-Oct-2003, 12:08
Please, show me our rediculous bias, because I'd love to see examples of it. I've been writing there for 3 years and I've yet to be biased towards anything, at least not conciously. So be my guest, show me our bias, mine in particular (I'm not the only one who writes there) :)
As for income, I've personally yet to make a penny from HEXUS. I currently don't get paid for writing there. I get to play with nice hardware when it comes in for review, but I'm definitely not earning a living there. Thanks for your completely unfounded speculation, it's utterly entertaining.
Rys
Ok..
1. There can be no question that the GeForce FX out performs the Radeon. We saw the demos of the Ogre, Dawn and the truck - these are incredible. We will have these on the media archive on the end of this review. There is more to the FX than just gaming, and benchmarks. We believe that this card will be the future of games - with the CineFX engine, High-Precision Graphics, Intellisample Technology all needing to have support from the developers to take full advantage of this card.
Card was cancelled BTW :lol:
http://www.hexus.net/content/reviews/review.php?dXJsX3Jldmlld19JRD00OTcmdXJsX3BhZ2U9MTQ =
2.The recent 9600 XT review where the reviewer (you) includes benchmarks from a unreleased card.....5700. http://www.handykult.de/plaudersmilies.de/eek7.gif
I made a thread about your journalistic integrity..way back along with the other pawns of the industry...the sad state of it rather.
http://www.rage3d.com/board/showthread.php?s=&threadid=33670711&highlight=hexu s.net+5800
can you cut the melodrama and sighs and stop trying to make your mistake look like less of a mistake?
If there's an obvious difference between the two, I can't see it. There's more final texture detail in the wall textures in the ATI shot, due to the viewer being slightly closer.
That is incorrect, and was the only point of this thread.
*sigh*
The thing is, its becoming more and more apparent that many reviewers just cannot seem to understand things. For instance, you compare framerates between FX cards and Radeons in Serious Sam, but make no mention that ATI offers more detail...i honestly wonder if the loss of detail in the FX screencaps gives them an unfair performance advantage. I'd say yes.
There's only loss of detail between them due to viewer distance. Now that I have a quick and easy way of creating identical screenshots in SS2, I'll redo them.
The review only talks about the floor texture (the easiest with which to see mip transitions and the filter in question). I mention wall detail to explain viewer distance again, since it's obvious they are different.
As for the melodrama, people latched on to things I tried to explain in the review text, apparently ignoring the review text altogether. It annoyed me, I apologise.
Doomtrooper: I've explained both things in other threads. The 5800 article should never have been published and my reasons for using the 5700 in the 9600XT article are well stated in other threads. If that's all there is to show bias, I'm pretty happy.
Rys
Doomtrooper
26-Oct-2003, 12:49
Then why was it ?? Journalists are nothing without integrity....
Dave Baumann
26-Oct-2003, 13:18
Rys,
You might want to check out this for comparing IQ:
http://www.beyond3d.com/forum/viewtopic.php?t=8612
digitalwanderer
26-Oct-2003, 13:35
Then why was it ?? Journalists are nothing without integrity....
You obviously never read the Inquirer... ;)
The review only talks about the floor texture (the easiest with which to see mip transitions and the filter in question). I mention wall detail to explain viewer distance again, since it's obvious they are different.
That is kind of like beind run over by an elephant while looking for a needle in a hay stack.
Rys,
You might want to check out this for comparing IQ:
http://www.beyond3d.com/forum/viewtopic.php?t=8612
Thanks for the link. I prepared two new SS2 screenshots and amended the review with them. They are of equal size and viewer distance this time :oops:
Anyway, my comments about the filter quality are still valid, using The Compressonator to compare both new images (getting a Difference output from the tool), shows precious little difference between them, especially in the floor texture. I've stated that in my opinion, the ATI shot does look better, but that at the same time, the NVIDIA shot doesn't look bad. Just ever so slightly worse.
I'll keep using the tool, I've been hunting for something that showed an XOR style difference between two images for a while.
What's your opinion on 52.16 IQ?
Rys
Althornin
26-Oct-2003, 18:22
The thing is, its becoming more and more apparent that many reviewers just cannot seem to understand things. For instance, you compare framerates between FX cards and Radeons in Serious Sam, but make no mention that ATI offers more detail...i honestly wonder if the loss of detail in the FX screencaps gives them an unfair performance advantage. I'd say yes.
There's only loss of detail between them due to viewer distance.
That makes zero sense, considering the ATI shot was MORE detailed and was FARTHER away.
Explain how getting closer makes one lose detail?
But, regardless, you fixed that.
One other question:
Analysing a static screenshot is a different beast to watching a moving picture.
you use this to imply that its even harder to see a difference in-game, while i find its exactly the oppposite. Crappy moving mip-lines are easier to see in motion.
gokickrocks
26-Oct-2003, 18:35
That makes zero sense, considering the ATI shot was MORE detailed and was FARTHER away.
Explain how getting closer makes one lose detail?
the camera is far sighted with poor near sightedness :wink:
you use this to imply that its even harder to see a difference in-game, while i find its exactly the oppposite. Crappy moving mip-lines are easier to see in motion.
i would have to agree
Rys, why don't you use PNG instead of TGA? PNG uses lossless compression and can be viewed in any modern browser.
martrox
26-Oct-2003, 19:11
Hexus isn't blind....they just need glasses! :roll:
Didn't your mothers tell you to stop doing that or you would go blind? Just decided to take a pause, huh? :wink:
Explain how getting closer makes one lose detail?Consider it a free DoF effect. ;)
Why are you surprised? Hexus was always the most ridiculously biased site, sometimes even against the real world realities. Business as usual: I'm pretty sure it's worth it for them, in terms of income...
EDIT: grammar
Please, show me our rediculous bias, because I'd love to see examples of it. I've been writing there for 3 years and I've yet to be biased towards anything, at least not conciously. So be my guest, show me our bias, mine in particular (I'm not the only one who writes there) :)
As for income, I've personally yet to make a penny from HEXUS. I currently don't get paid for writing there. I get to play with nice hardware when it comes in for review, but I'm definitely not earning a living there. Thanks for your completely unfounded speculation, it's utterly entertaining.
Rys
My friend, take a look on EVERY FX-related 'article' you/your colleagues did. Almost each of them has something interesting approach, towards NV cards...
Unfortunately (err, lucky me ;)) I was off for this weekend - I could be only second to Doom... he posted our favourite quote from last year...:D
Ahem... you're saying you do this for free? :shock:
Doomtrooper: I've explained both things in other threads. The 5800 article should never have been published
:shock:
And you guys have no time to remove that since 27th January 2003, huh? Actually you started but this action finished very soon: only graphs been removed since then... nice try. Now NOBODY can judge anything from them.
? :shock: Talking about conspiracy theories...
and my reasons for using the 5700 in the 9600XT article are well stated in other threads.
May I have a link, pls?
If that's all there is to show bias, I'm pretty happy.
Rys
I don't like to turn this thread to some journalist (PRE)school ...
PS: I always been pretty harsh on unbalanced writings or mispresentation - especially if they tried to act like a 'very independent' piece of something.
As per melodrama: sorry to say this but you should have been learned one of the main rules: you write articles - and I judge them. Even if I'm blind, bad and evil...
Doomtrooper: I've explained both things in other threads. The 5800 article should never have been published
:shock:
And you guys have no time to remove that since 27th January 2003, huh? Actually you started but this action finished very soon: only graphs been removed since then... nice try. Now NOBODY can judge anything from them.
? :shock: Talking about conspiracy theories...
and my reasons for using the 5700 in the 9600XT article are well stated in other threads.
May I have a link, pls?
If that's all there is to show bias, I'm pretty happy.
Rys
I don't like to turn this thread to some journalist (PRE)school ...
PS: I always been pretty harsh on unbalanced writings or mispresentation - especially if they tried to act like a 'very independent' piece of something.
As per melodrama: sorry to say this but you should have been learned one of the main rules: you write articles - and I judge them. Even if I'm blind, bad and evil...
The 5800 article isn't my doing, I don't have control over whether it's removed or not. As for the missing images, there's a load of reviews we have, that don't have pictures any more. Our image server was hacked, we lost all of our image content. Only stuff that we had backups for were restored. Randomly choose an article from our database, sooner or later you'll find another with no pictures, there's over 100 like that I think.
And yes, I do this for free. I write for my own enjoyment, to learn about hardware. I often get to hang on to hardware for an extended period for my own use though, that counts for something I guess.
Search for posts by me to find the thread on 5700. I used the card because I had nothing else to hand (yes, my fault, but that's the reason nonetheless).
We learn from our mistakes :) I'll convert the .tga's to .png later today, I deemed .tga more suitable since it's the native capture format and was conscious of losing image data due to conversion.
Rys
Genghis Presley
28-Oct-2003, 02:06
Anyway, my comments about the filter quality are still valid, using The Compressonator to compare both new images (getting a Difference output from the tool), shows precious little difference between them, especially in the floor texture. I've stated that in my opinion, the ATI shot does look better, but that at the same time, the NVIDIA shot doesn't look bad. Just ever so slightly worse.
I'll keep using the tool, I've been hunting for something that showed an XOR style difference between two images for a while.
When using The Compressonator for investigating the differences between two images it often pays to bump up the diff image gamma (using the lightbulb icons). Small differences are difficult to spot as diff images are inherently dark. Obviously the differences still need to interpreted with great care but bumping up the gamma allows them to be more clearly seen and often allows patterns within the diff to emerge.
GP.
OpenGL guy
28-Oct-2003, 02:19
I'll keep using the tool, I've been hunting for something that showed an XOR style difference between two images for a while.
XOR is about the worst possible way to compare two images.
Imagine: 0x80 XOR 0x7F. Result is 0xFF!
I'll keep using the tool, I've been hunting for something that showed an XOR style difference between two images for a while.
XOR is about the worst possible way to compare two images.
Imagine: 0x80 XOR 0x7F. Result is 0xFF!
Wrong expression, but you get what I mean :) Thanks for the tip about gamma GP. Is the Difference output of Compressonator what I should be looking at, where non black pixels show the difference between the image pair?
If so, I'll start and use some commentary on it in the future. I remember 3DMark2001 Result Browser doing something similar, when it compared IQ shots to the refrast output.
Rys
Genghis Presley
28-Oct-2003, 04:57
I'll keep using the tool, I've been hunting for something that showed an XOR style difference between two images for a while.
XOR is about the worst possible way to compare two images.
Imagine: 0x80 XOR 0x7F. Result is 0xFF!
Wrong expression, but you get what I mean :) Thanks for the tip about gamma GP. Is the Difference output of Compressonator what I should be looking at, where non black pixels show the difference between the image pair?
Rys
Yes. The Compressonator calculates the diff image as the absolute difference between the two images for each channel, so where pixels are identical between two images the diff image is black.
GP.
Mintmaster
01-Nov-2003, 05:01
Rys, I see a clear difference between the (updated) images. Just flip back and forth, and you can clearly see the diff. I'm not sure how you came to that conclusion. However, this probably has to do with fudging with LOD or AF level by the drivers rather than the mip-map issue we see.
I find the best place to see AF and filtering problems are in racing games with a low camera like a bumper cam or hood cam. When the road has lane stripes, then you can really see little things.
Consider a road lane divider stripe created by a texture that's one texel wide and say 20 texels long.
Now, if you had bilinear filtering, within a mipmap a road strip will get narrower in the distance due to perspective. When you cross a mip-map, then suddenly the stripe will get wider, blurrier, and lighter. Then it will get narrower with distance due to perspective again, and after the next mipmap boundary widen and lighten again, and so on.
Not only are you seeing a mipmap boundary in the stripe, but the edge of the stripe has a discontinuity as well. With true trilinear filtering, however, you wipe out this discontinuity. With NVidia's pseudo-tri filtering, you'll get a bumpy edge on the road line.
I'm not sure if anyone can envision this as well as I can. If you guys really want, I can make a small app to demonstrate this, but it'll have to be worth my while (maybe some screenshots in a review or even in some threads). Just PM me and I'll get it done within a week.
Good filtering does more than just make mipmap boundaries go away.
ElMoIsEviL
03-Nov-2003, 17:59
Rys
I downloaded both the newer image (the ones currently on the website as of November the 3rd 2003).
I used a program called ACDSee.
I used to Shrink zoom option when an item is viewed full screen.
I looked at both pictures and the Trilinear Filtering IS CLEARLY EVIDENT...so damn evident I was gonna to &*^* my pants with excitement.
Also the texture details on the Radeon card are soo superior over the nVidia card that you'd think you were wearing glasses (when you don't need them) by looking at the blurryness of the nVidia shot.
Peace.
vBulletin® v3.8.6, Copyright ©2000-2013, Jelsoft Enterprises Ltd.