My take
Blackwind, several of your comment strike me as inconsistent. Perhaps we could discuss them a bit.
Blackwind said:
The write up was not a justification for "almost trilinear." It was a review of IQ between cards if anything."
The problem with this statement is related to, as one example, the distinction between hardware capability that is the common case and can be highlighted by specific tools, and a driver decision made for a specific application and therefore a specific representation of the hardware. The description "a review of IQ between cards if anything" excludes that fps was even mentioned, and that the IQ "review" was being used to represent hardware performance. Your commentary seems to consistently depend on isolating the opinion on IQ from any concern about fairness in hardware descriptions.
Let us start with an attempt at a more complete representation of the anisotropic filtering issue than [H] presents, and, as far as I understand at the moment, you seem to as well when yuu've commented here:
- [*]There are aniso tools that highlight the behavior of aniso and magnify issues with it in a way that aids in objective evaluation by giving information on what the aniso implementation is doing. Unless you spend your time just using such tools, this is just a foundation for relevance to the next points. Extensive and well thought out scene selection is a more complex, and less illustrative, alternative, but objective factors need to be established somewhere as the criteria for their selection to demonstrate accuracy.
[*]There are benchmarks and games used as benchmarks chosen to aid in providing the user information to evaluate, where the user can observe the performance cost of the aniso implementations in the context of achieving the last point. Unless you spend your time watching benchmarks of either stripe, and maybe bragging about scores in them, this is just a foundation for that last point, which is the primary intended use for the cards (I'm focusing on games).
[*]There are games where the user is using the aniso, and the image quality the user is getting matters for the user's experience. For the "tech junkie", the above foundations for comparing this are of primary interest, because they have interest in evaluating their relevance for themselves. For the "gamer" concerned only with this final issue, the foundations are important if they actually want an accurate picture of it for more than the specific games and positions represented in screenshots.
The last point doesn't stand by itself, it depends on the prior issues to establish its accuracy. This concerns there being a distinction between informed and uninformed opinions outside of saying the person who has the opinion matters more than the support for it, and brings to mind the saying "opinions are like a...". My answer (to the saying, which might not represent what you intend): "yes, everyone has one, but speak for yourself when you say they have to stink".
The focus of a
hardware review or article can
emphasize whichever it wants, but it can't beg off recognizing the first elements as being the foundation of the last while putting forth something built upon them, when the question of how well that structure will stand is being asked.
Now, the article is not a review of UT 2k3 image quality, it is a response to a question concerning [H]'s review of
hardware while using UT2k3 and comparing fps as well.
Let us compare some example situations for UT 2k3 without omitting consideration of these particulars. For this discussion, we'll consider the relevant example cases of: the Radeon 8500, the Radeon 9500-9800, and the GeForce FX family cards and driver versions exhibiting this behavior.
Let's start by discussing these example cases with regard to the first point about aniso tools:
- 8500: The hardware can't do trilinear while doing aniso (up to 16x) and has fairly prevelant issues with applying aniso at some orientations. Aniso tools show this, but they have to be used and there was a time they weren't prominent and some cases where what they show aren't represented at all. This is a straightforward issue of knowing about it, and standards of reviewing evolved to provide that knowledge.
- >9500: The hardware has less prevelant issues with applying aniso (up to 16x) at some orientations, and can do full trilinear at the same time. Also, there is a driver behavior when forcing aniso that requires tools to represent more than the initial texture stage. Both are straightforward as above (once known), and the latter is consistently and logically resolved by using Application Preference.
- FX: The hardware is capable of doing aniso (up to 8x) with full trilinear at the same time. Also, there are driver behaviors that control how this is applied on a per application basis. One of these driver behaviors is defined as trilinear filtering, others are defined as bilinear and some type of intermediate. These are straightforward (once known, and if application detection isn't abused).
You maintain that evaluating image quality equivalency is "simply be a matter of opinion in regard to appearance". The aniso tools are part of the picture for making that opinion an informed one and providing a guideline for objective factors regarding it.
Next, let's consider benchmarks and games used as benchmarks:
- 8500: All benchmarks and games used as benchmarks accurately represent the performance and image quality of the aniso implementation when showing their scenes. At the time of the 8500's introduction, this emphasized a need for screenshots and analysis of what is represented in them.
- >9500: All benchmarks and games used as benchmarks accurately represent the performance and image quality of the aniso implementation when showing the scenes. When aniso is overriden and "Quality" selected, there is an issue of scenes being represented in the specific case of colored mip level usage if a lack of understanding of what colored mip levels don't show is present.
- FX: Benchmarks and games are specifically targetted such that performance evaluations are blatantly misrepresented. "Coincidentally", this is made evident for both 3dmark 03 and UT 2k3 specifically (by name, even), the leading applications used for benchmarking. Showing scenes where the "opinion" of the impact of this on image quality is disputable does not change that this action is a blatant misrepresentation. Doing so while providing and defending fps comparisons for this misrepresentation is participating in that misrepresentation yourself.
You're treating the occurrence of this stage as the same between the FX and the 8500 ("not apples to apples") because you're viewing it as if it is only a matter of the user playing the game in the specific scenes shown and not what is being represented about the card playing other games or other objective factors that are obvious in different scenes. UT 2k3 is
not the only tool available, and knowing it specifically is problematic indicates that another is called for or that the issues with it need to be discussed.
OK, when discussing the results in game:
- 8500: Shouldn't mention fps results and represent equivalency without taking steps to indicate what the screenshot selection might not emphasize or might exaggerate about image quality. For the 8500, it is what might not be emphasized that is an issue with regards to typical screen shots, and screen shots of an aniso tool and some evaluation of the apperance of game scenes in general tends to accomplish this briefly at what is established as "minimum" reviewing at the moment.
- >9500: A similar discussion suffices for "minimum" review detail. Thanks to exploration by "tech junkies", the issue of control panel "Quality" behavior is also determined to be a further requirement (though colored mip levels which would tend to misrepresent the issue as they are currently implemented and interpreted) for the "new minimum", in the vein of ongoing evolution of reviewing.
- FX: A similar discussion suffices when using software where Quality behaves as it is represented. This can include aniso tools and some games, along with screenshots. UT2k3 is not one of those games. The [H] rebuttal to the concerns of ATI (among others) focuses on answering this from the perspective of screen shots and game play, while ignoring that the question is related to the foundation of that representation and the factors discussed above concerning how they are applied to hardware reviewing.
Note: As I've said before, I think another card that can be done similar disservice of this type is the 5200 when compared to the 8500 without bilinear filtering being selected. Given the performance hit of AF and the performance of the card (as long as it compares price wise to the 8500-9200 cards instead of the 9600), this is a necessary factor for consideration along with the shader performance comparisons and other performance characteristics.
Addressing Kyle's stance as I understand it:
Knowing this for "months" in advance and "behind the scenes" just makes you a witting accomplice in the misrepresentation.
Attacking an application's use as a benchmark for not being "applicable to gaming" (in your opinion, without establishing objective factors about that opinion being valid) and supporting that opinion with how it is being targetted and is therefore inapplicable to games in general,
while knowing specifically how another application you yourself are using for benchmarking is being intentionally targetted in the same way, and still proposing the use of the latter without even mentioning it,
just makes you hypocritical.
Thinking the image quality difference is insignificant yourself, based purely on limited selection of screenshots and with no objective criteria established to indicate their usefulness while supporting them as a universal representation (remember, hardware review, not UT 2k3 review), doesn't give you the right to speak for more than your eye sight and the particular still screen shots used to support it, or indicate anything to counter observations of how the particular screen shots are lacking.
These are things that objective criteria, and objective discussion of your provided examples and what might be wrong with them, are used to evaluate. Someone criticizing you for proposing your subjective evaluation of limited cases and your basis for selecting them as universally representative of an issue, does not require anything more than disagreement.
Deleting, ignoring, or muting someone as an alternative to discussing objective criteria they present, by proclaiming them as having an agenda, is simply avoiding dealing with that disagreement, it does not make you any more "right".
The "size" of your web site doesn't change this, or make you any more or less "human" and potentially mistaken as anyone else. Your conduct does. Maintaining that as a factor in your discussion seems a rather meaningless proxy for the discussion. That's popularity, and it comes and goes without direct association to merit. Confusing this with anything else just encourages comparisons to politics and politicians.
