"I WANT TO BELIEVE": Nvidia's texture hardware X-file and Anantech's silence

[audreymi]

Label me a unbeliever due to this one glaring issue, but the Game Basement link has confirmed (for me) that this review
site has been, at times, handing out preferential treatment
to certain produts. There is definitely some sort of
conflict of interest taking place, for whatever unknown
reasons, even if no money is changing hands
(Dinner at COMDEX? Growing up with them over the years? Some of their engineers got their start at Anandtech?,???).

Maybe it is as simple as plain preference. The same reasons my dad drives a GM car even though all his sons and daughters drive imports.

I'm not saying that Anandtech is
a bad person or that I do not find his site useful, it is just that
given the opportunity to sway a review one way or another
on the basis of simple unwarranted personal choice, I inclined to think he might. Integrity
is somewhat like virginity, once broken, it is next to impossible
to recover.

Instead of going boxing day shopping, I decided to chase down some more
of the game basement links to give some more
background to the texture hardware bug currently
raising questions about Nvidia GeForce based graphics cards
having artifacking in skies under Quake 3 and first uncovered
by the Game Basement site.

As I dug, I found that the site that has quite a knack for sorting out complex issues, like this Quake3 sky banding texture issue, using a combination of clear thinking and experimentation. In the writing department,
it was quite refreshing to see articles published on the strength of this
basis instead of typical "read it here first!!!" and the
standardized easy mark of posting and ranking graphics cards
in the order of the framerate scores.

The series of review articles that ignore this problem form the smoking gun evidence. Changes in tone and content
from one article to
the next point are quite telling in
how "biased" certain popular review sites are.

The acts of accentuating the positives of a favoured son while
emphasizing the faults of a gifted daughter are common place in male dominated cultures. They wil doom her to a predictable
and undeserved role of poverty and untapped recognition.
In more liberal cultures, she
might run for president with her gifts. This is the dark secret
of silence and bias.

Whatever the reasons for bias? Here is a brief chain of articles
to help you draw your own conclusions on the nature of this
problem and the role reviewers have played to date.

1. May 31, 2000: 16 bit vs 32 bit article examiming banding in relation to rendering resolution and texture resolutions
2. August 15: Detonator drivers released. No metion of banding in reports from Anandtech and Tom's site.
   
3. August 31, 2000: Quake3 and sky textures with GeForce: part 1. Radeon and VooDoo 5500 look fine.
   
   As the article says:
   
   << &quot;The most important features on the card appear completely broken&quot; >>
   
   .
4. September 24, 2000: Quake 3 and sky textures with GeForce [ part 2. Pre-compressed textures look fine but Real time compression of textures and lightmaps are blotchy and broken.
5. December 21, 2000: Quake 3 and sky textures with GeForce: part 3. Hardware bug detonated with systematic experiments to explode
   3D coverup.
   

For Tom's part, the evidence is startling. 8 Geforce related articles
published between August 14 and October 19 where Nvidia cards were
reviewed and not a single mention of the 3D artififacking issue
found in many forums and raised 4 times in that period by Games Basement


For Anandtech's part the evidence is just as startling. 20+ GeForce related articles published between July 2000 and November 2000, again, not a single
mention of the banding problems.

Trust me, I still respect Anantech's site, but I am finding it more and more difficult to trust them implicitly, as their recommendations become increasingly different with those opinions of other review sites and retailers in my own city that carry
both Nvidia and competing produts. This X-file remains open the longer Anand and Tom remain silent about the current issue.

I want to believe.

 

[Compellor]

It's possible -- but unlikely -- that they weren't using texture compression when reviewing the cards. The sky issue in Quake 3 has been around since the first Q3A demo was released in 1999. My old TNT had problems with the sky. I'm sure if you look back at the actual GAME REVIEWS of the game, you'll see that many sites made mention on it if they were using a nVidia based card. Anandtech and other hardware sites are simply just benchmarking the cards, and appear not to be mentioning the image quality of Quake 3. Maybe they don't consider it that big of a deal, or maybe they are on the take.

 

[Cerb]

The hardware sites are reviewing the performance of the cards. They have less reason in those reviews to notice this problem. On top of that, they may not really care as much about this bug. I am getting a GF2 GTS...this whole thing doesn't bother me one little bit. We use sites for many different things, also. Games Basement is full of this kind of thing. Anandtech, sharkyextreme, and tomshardware aren't. It is good that there is someone around figuring this stuff out, but no one site will be omnipotent, nor should the staff be expected to try for that. We've known that w/ TC on, Q3 has looked terrible on the GTS cards. If NVidia really purposely did this, they have some real idiots planning stuff, as for many people, image quality means more than framerate...and even when it doesn't, when it looks THAT BAD, framerate takes a slight back seat.
Enough of my ramblings...

 

[BFG10K]

Compellor:

It's possible -- but unlikely -- that they weren't using texture compression when reviewing the cards.

Not a chance. S3TC was being used.

Anandtech and other hardware sites are simply just benchmarking the cards, and appear not to be mentioning the image quality of Quake 3.

Exactly. Most of these websites just run the benchmarks and don't seem think it's important enough to mention image quality.

I was especially surprised at Anadtech's review where they &quot;threw out&quot; the Radeon SDR and crowned the GF2 MX as the best value card. Um, hello? A lot of other sites have compared the image quality in Quake 3 and other games have found the Radeon to have the best.

Just because the GF2 MX scores 5 FPS more than the Radeon at 640 x 480, that's hardly a reason to crown it the king.

 

[audreymi]

Just got back from the Boxing day sales with my son an we
we got ourselves a HP 612C refurbished printer for $49 after the rebate.
UnReal Tournament for free after the $20 rebate.

My last video card was a Matrox G200 bought on the strength of
Tom's review that focussed on image quality.

Anandtech has been aware of this problem since July when he did
a comparison of FSAA image quality.
He has side by side shots of
three video cards to highlight the issue. Since then, no mention
of the resolution or importance of this issue has been raised.
The tone of all graphics reports has decidely changed from that
of the July article where he said



<< &quot;Let's take a quick look at how bad the sky looks first: Pretty bad, no?&quot; >>


.
.
.


<<
&quot;You can also see that both the 3dfx &amp; ATI solutions are using texture compression without the horrible effects on the sky that are present in the GeForce2 GTS image.&quot;
>>



For something that was horrible in July, it somehow became unimportant in all subsequent reviews on Nvidia GeForce products.
Why?



To me it is their choice as to whether they continue to
refrain further from following up
on this issue and continue to focus on preferencing the positives
of one brand of video card (for whatever reason(s)). It just begs
the question of why?

 

[audreymi]

restored post:

Date Posted: Dec/27/2000 12:42 PM

Starting today, I will begin to count the days until
a response is heard from someone at Anandtech.

Day 1: ALS, your move

 

[audreymi]

Day 2: MW, your move.

 

[BenSkywalker]

If you have any questions on bias on my post, check my sig

Thanks for the kind words on GB, but I feel that you are being rather harsh on the hardware review sites. Their &quot;duty&quot; is to review hardware, at the Basement we focus on gaming. The two are intertwined, I will give you that, but in this particular case you are talking about tweaked Quake3 having poor sky textures, that is the extent of the problem as it exists right now.

As of this moment, the most up to date Quake3 patch disables TC by default, clearing up any and all issues with any board. As of right now benchmarking is also disabled, but we all hope that will be quickly rectified.

Ignoring that, the problem is one with Quake3 and only Quake3 dealing with this particular issue, and one that is very likely to vanish shortly(you already can fix it yourself with the latest nV beta drivers). How much attention should be paid to an issue with one game that will soon be gone by a hardware review site? The GF2 exhibits this issue in one game, and even then only when tweaked as of this moment, what business does a hardware review site have dealing with it?

At the Basement, a great deal of research went in to the TC series of articles. The final point that prompted the latest article was the release of nVidia drivers that had a workaround built in to deal with this exact problem. The workaround is done by directing the hardware to use DXT3 instead of 1. Remapping the DXT calls is something that 3dfx already does, instead of using S3TC they use DXT1.

What about UT on the subject of texture compression? The V5 quite frankly looks very poor when compared to the Radeon or the V5 if we use the Loki S3TC patch. Should that also warrant special attention from AnandTech?

How about the Radeon's disabling of compression on textures that are sized 128x128? Should that be worthy of AT's focus?

They deal with hardware, and how it performs. We, at the Basement, deal with gaming, and issues that effect it. AnandTech gives you the raw numbers and breakdown on the hardware itself. To ask why they didn't report fully on this issue some time ago is to seriously underestimate how much work it took to find out exactly what was going on. The articles may be spread out over a lengthy period of time, but I can't think of any two week stretch since I have been at the Basement that I didn't get at least one email from &quot;wumpus&quot; on the subject. We have been working on this, both the good and the bad, for quite some time now.

We have contacted numerous people at many different companies, hardware and software, dug through every article posted on old S3 boards that we could find, searched through the S3 servers to dig up all domcumentation we could find, and worked with a couple of different programmers on exactly what was happening. Not to mention observing as many possible issues in games as we could.

Should a hardware review site be focusing on such a relatively minor issue? Seriously, if this was an issue with every game, or even a handful of games, I could understand a hardware site making an issue out of it. But one game? And even then, what exactly was the problem? We spent some time looking for the answer, 'wumpus' quite a bit more then I and I know how much time I have spent going over this.

This isn't an image quality problem on a broad basis, as the people at AT definately saw how it looked before TC was enabled, and the sky problem in Quake3, in the scheme of things, is quite minor.

Where is the reporting on how poorly the V5 looks in UT compared to the GF or Radeon, or Evolva? Where is the reporting on the problems with MBTR with the Radeon? Look to a site focused on gaming for articles like that, not hardware review sites.

 

[audreymi]

Bias might be termed the &quot;dark side&quot; of the Force wrought by
by the power of internet publishing. I'll just comment on
the general gist of your reply.

The merit of one's work in in the details. Saying that &quot;Anandtech is
just a hardware performance site to provide raw numbers&quot; is
just pure and simple rationalization to allow them to continue
their silence on the virture of what they deem &quot;minor&quot;.
Ethics govern the &quot;right&quot; actions to take and integrity
makes for an overall holy grail grade to strive for. I believe
that can do better on both fronts.

As in any comparison used to generate observations
and data for reporting, differences in the details need to be stated
if the conclusions and recommendations are to carry any weight
for the broad segment of consumers. Assuming that some details
are unimportant has led to the return of those products on
the basis of blocky textures (Riva), bad 2D (GeForce and GeForce2),
and now perhaps broken DXT1 (GeForce2).

Saying something that is &quot;broken&quot; on my part is oversimplifying
the matter and guilty of labelling an issue.
Discussion of the how broke it is, the consequences of it being broke,
and its effects on benchmark comparisons is a necessary step as
the benchmark comparisons are not &quot;apples to apples&quot;. If one
prioritizes quality (many do), then leaving off texture compression
during benchmark comparisons is the right thing to do until
GeForce chips fix their problem.

Using DXT3 is fine providing users understand the consequences
of remampping calls from DXT1 to DXT3 (less compression but
any other issues?). This is especially true before making wholesale
recommendations in roundup reviews and the like.

 

[oldfart]

FYI. There is a benchmark that works on Q3 1.27

From the console:

timedemo 1
demo demo127

 

[TMTCC]

Sheesh...it's a moot point trying to stir up trouble over such an insignificant issues. The issues over the sky issue in Q3 has been beaten, rolled, and beaten to death many times in this forum. But have people objected or raised concern over Anand not pointing or putting less emphasis on the issue? Not from what I've read. And I don't think Anand needs to bring up the issue because readers can access the forums and read of people's experiences on the card.

If you want to continue to gloat over such issues, go right ahead, but don't expect much change.

 

[BenSkywalker]

&quot;Using DXT3 is fine providing users understand the consequences
of remampping calls from DXT1 to DXT3 (less compression but
any other issues?).&quot;

Between one and two FPS loss on average during my rather extensive testing on the subject.

Other then that, do you expect AT to test every single game and application ever made and report every single issue with every single video card and then every single driver revision?

It sounds like that is what you are asking for. I gave examples of each of the three main competitors from this generation and problems they have with games, that is the way things are. Comparitively speaking I would say that the GeForce TC problem is quite minor, it doesn't exist in the default settings and will be fixed shortly in an official driver release.

Besides that, how much time should they spend investigating what exactly is causing every given issue with every given game? Would it be better for them to not post a review until six months after the card launches so they can look at every possible issue?

&quot;If one prioritizes quality (many do), then leaving off texture compression during benchmark comparisons is the right thing to do until GeForce chips fix their problem.&quot;

If you disable TC on the GF boards, then they have clearly superior visuals. None of the boards have &quot;lossless&quot; texture compression, they all suffer image degredation. Test them all with it off, or all with it on. Either way would be fair(both would be best IMHO), the issue with the sky has been noted on AT, what else is there to say?

Edit- Thanks oldfart Have to try that out now

 

[RoboTECH]

I personally hold Anandtech to a higher standard. I expect them to find issues like this and report on them. They certainly jumped at the chance to review the framerate increase in Q3 with Det3, yet ignored some of the &quot;issues&quot; that Det3 (6.18 specifically) brought to the table.

The TC issue is/was a MAJOR issue, since it had a profound effect on the visual quality of a game which is considered the &quot;cornerstone&quot; of any video card review.

At the same time, the Radeon was/should've been chastised for their &quot;16-bit texture&quot; issue, and the 5500 was/should've been chastised for their lack of trilinear filtering.

Is this because these issues have a profound impact on visual quality....to an extent. The point is, we're not comparing numbers here. There is a VERY valid concern that if you use &quot;the same settings&quot;, then you should get &quot;the same results&quot;. Unfortunately, this isn't true.

Imagine (most obvious example) if, for example, no one had &quot;settings&quot; for FSAA, they only had &quot;on&quot; and &quot;off&quot;

If card 1 uses 4 samples/pixel, and card 2 uses 2 samples per pixel, yet both are &quot;FSAA on&quot;, is it fair to compare them?

If card 1 uses trilinear filtering for &quot;High quality&quot; settings, and card 2 uses bilinear filtering, is it fair to compare them?

What if a card allows you to set &quot;4x4&quot; FSAA, but then will drop down to 2x2 or 1.5x1.5 without &quot;telling&quot; you? Is it fair to compare?

That's the problem I have with 99.9999% of all video card reviews nowadays. As ole' Roscoe might say, &quot;you can't validate these results&quot;, even if they are repeatable.

Or, to put it more plainly, video card reviewers are f$$$ing lazy these days. The venerable &quot;timedemo&quot; has become the bane of honest, unbiased graphics reviews everywhere. The focus is now on the #'s. Well guess what, there are 3 types of lies:

1) lies
2) bold faced lies
3) statistics

today's graphics card reviews are stuffed to the brim with lie #3.

<dismounts soapbox>

That being said, just disable TC on the GTS. It's fast enough without it.

P.S. Kudos to Jeff for pushing this issue so hard to get to the bottom of it. For someone who is so vehemently anti-3dfx, it's funny to see him rip nvidia a new a$$hole.

 

[bluemax]

You bought an HP612C? Ewww... Good luck with that junker when the 3-month warranty runs out. And ink is over 2x the cost of their 900-series. It's old, slow, poor quality, and prone to breakdowns (wanna' know how many we repaired? I lost count... hundreds... ecch.)
Take it back if you can- if not, I hope it was only $20.

 

[Spotch]

If you review a card for its performance and value, quality and quantity cannot be treated separately. I believe that this may not be &quot;persuasion&quot;, but instead just a guy rooting for his favorite team. I have an Asus V6800 Deluxe ($325) and a Radeon 32DDR ($149)and the Radeon's image quality and performance (in 32 bit... is there anything else?) is strikingly better to even the untrained eye (My wife's for instance).
I had the GeForce for a year when my Radeon arrived and showed me what I was missing.

 

[WetWilly]

I'm with Robo. For those of us from whom visual quality is a serious consideration - to the point of trading off FPS for it - pretty much ALL review sites ignore visual quality issues unless they're extreme. I find it rather pathetic that a one sentence comment on comparative visual quality - as subjective as that may be - is such a rarity in graphic card roundups. And yet, these reviews will spend absurd amounts of time benchmarking Quake, MDK2 and UT at multiple resolutions and bit depths on multiple processor platforms. But it's asking too much to have a reviewer make a comment like &quot;Maybe it's just us, but we found X card has better 2D quality than the rest and Y card has better color saturation.&quot;

I'm one of those V5500 owners who is still within the return period for the card, and I'm looking at the V5500, Elsa Gladiac GTS, and Radeon 32MB DDR cards. In fact, I've got all three on my desk right now, and I've just popped out the GeForce2 and put in the Radeon. The first, immediate observation I made was that the GeForce is DEFINITELY not as good as the Radeon or V5500 at 2D - not maybe, not kind of close - DEFINITELY. And the Elsa is supposed to be one of the better GeForce2 cards. This is on a NEC 17&quot; FD Trinitron at 1024x768. I'd give 2D to the Radeon by a hair over the V5500. And yes, I went through multiple refresh rates and Colorific calibration to see if that made any difference for the GeForce. No dice. Could I live with it? I'm thinking about it. Would I have liked to have this info without buying three cards and doing it myself? Well, it would have been convenient to have this info in a review, but that seems to be asking too much.

And another thing while I'm at it ... using FSAA screen shots to evaluate FSAA is useless, and reviewers do FSAA an injustice by posting them. One good example - I haven't seen a screen shot yet that has illustrated how FSAA removes texture shimmer. Has anyone? I ran Unreal at 1280x1024 on the V5500 with 2x FSAA, and used the FSAA toggle to compare FSAA/non-FSAA. While the jaggies were minimized enough to arguably justify turning off FSAA, texture shimmer was still there and even 2x FSAA minimized it significantly.

 

[Doomguy]

The S3TC issue is pretty minor, it really only affects Q3. You cant say the GF2 has bad image quality because of a problem with one game. The Voodoo 5 has bad image quality too then because you cant use S3TC textures in UT so it looks terrible compared to GF's/Radeon's.

What issues did the 6.18 drivers have that were serious?

 

[BFG10K]

Doomguy:

The Voodoo 5 has bad image quality too then because you cant use S3TC textures in UT so it looks terrible compared to GF's/Radeon's.

Comparing an official game to an unofficial backyard patch is hardly the same thing.

 

[Playitagainsam]

Spotch

Not to put you down or something, but you're comparing a GeForce 256 with a Radeon? A GeForce 2 is a completely different animal. Far above the 256, better than Radeon in 3D, weaker than Radeon in 2D.

 

[RoboTECH]

well, I think people get a bit overboard with the image quality of the GTS.

It's 32-bit 3d image quality kicks fuggin' ass. It's FSAA is below what I like (5500), and it's 2d is just plain weak.

as far as the OGL S3TC thing, it looks fantastic, no doubt. However, and again, I'm going from my experience at a recent LAN, the GTS users with the S3TC patch were having far more problems than I was, WRT framerate maintenance, in-game stuttering and ESPECIALLY stability.

Maybe it's a LAN vs. offline thing. there can be a tremendous difference in some games. UT's netcode is kinda...&quot;interesting&quot; compared to, say, Q3's.

 

[audreymi]

1)Regarding printer:
Impulse buy to fill in for my infamous Brother HL-630 printer which is well known for streaking problems. Great printer when it work but a design shortcoming in the paper feed is the root of our problems. I'm
told that once the rubber cam wears out, misfeeding of paper eventually leads to severe streaking. Should be a consumer recall on this product.
I agree after seeing some Canon printouts, there are better printers
than the HP 612C. I sure hope your are not right about costs of
replacement ink cartridges.

2)Regarding followup posts:
Some very well written material which I hope the moderators will read
and take into account in future reviews. I will continue to count
the days until a response is heard.

3)Regarding 3Dfx textures or FTX1:
An excellent article on texture compression from digit-life.

4)Regarding side-effects:
The multiplayer aspect of Quake3 is a major departure for John Carmack and a new deliberate focus of this
game. It seems as if the last post indicates some problems associated
with the patch that perhaps alter typical latencies associated
with decompression and might explain the lack of
synchronization of game play with many players interacting at once. Most posts to date
about the patch refer to single play mode. Any more comments out
there about syncrhonization issues if one player uses DXT1 and another uses DXT3?

Day 3 is 3 hours away.

 

[oldfart]

I was using the UT S3TC thing for awhile on my Radeon but had to give it up. The LOKI driver only worked with the Radeon up to UT ver 4.32. 4.36, its broken. Even worse, I had a weird bug that only would happen when I played online, but not single player. I'd respawn and be stuck in my tracks. Couldn't move. I did some searching on the net and I saw posts from other people who had the same thing happen. Very weird. Oh well, it was fun while it lasted. The high res textures are really nice, but just too much of a pain (for me).

 

[WetWilly]

well, I think people get a bit overboard with the image quality of the GTS.

It's 32-bit 3d image quality kicks fuggin' ass

I would have agreed with you until this afternoon when I pulled the Geforce2 out and put the Radeon in. The Radeon's 32-bit is noticeably better than the Geforce's - colors are much more saturated. After looking at the Radeon, the GeForce looks washed out. ATI's FSAA is pretty useless, though.

If it weren't for ATI's driver issues, the Radeon would be a no-brainer over the GeForce. Too bad some of those guys from 3dfx don't go to ATI to give them a hand.

 

[lsd]

The S3TC issue is gone with Team Arena.
Also if Team Arena is a sign of things to come..say good by to 32bit color in new FPS games.

 

[oldfart]

lsd, I have the TA demo, and the TA full game on the way. S3TC is disabled by default on the TA demo game, and I assume on the full version. What do you mean by &quot;say good by to 32bit color in new FPS games&quot;? I thought The demo played very smoothly. It only has that one map, so its kind of hard to say.