PC Magazine's IDIOTS screw Matrox in a preview

[Stringy]

Originally posted by xortam
So maybe we're talking about a pippin apple to a Washington apple comparison. Its the reviewers' responsibility to make the reader aware of the differences in the feature comparison. FAA, FSAA super, and FSAA multi are all AA features and they should be compared to each other.

And leaving out a feature Directly aimed at the task at hand is bad?
If the Whole idea is to remove the Alaised edges from the Game or App. what does it matter which "method" is used??
the end result is the reviewers Concern...

Now, if a feature doesn't work in a certain circumstance this should be noted and that's where the other Anti-Alaising method that the P offers comes in...
I have yet to hear of a game that doesn't work at all with FAA... we know it doesn't work on images produced using a Stencil Buffer [ie shaddows] ... but not a whole game...

If the newer Method the P puts forth doesn't show a better resultant image that of course is very important....


anyways, I don't get print anymore, too slow...
Craig
Edit correct Terminology - Frame buffer, Stencil buffer, whats the Diff Thanks UberLord

[UberLord]

Originally posted by Stringy
we know it doesn't work on images produced using a Frame Buffer [ie shaddows]

It's the Stencil buffer, not the frame buffer.

Hard to use FAA if the card is drawing a Frame from the buffer

[xortam]

Folks ... go back and read my first post. I said that the review should have included a comparison with FAA. I'm not going to argue with you on that one.

[Ryu Connor]

Fine.

You make average Joe User understand the differences

I'll bet they have a hard enough time understanding the basic conecpt of AA in the first place!

That has been tried in this very forum before and it didn't go well. Your average gamer does not understand the difference between the AA methods used by each of the current graphics titans.

Needless to say comparing the 4X FSAA OGSS of the Parhelia to the 4X MSAA OGSS of the GeForce 4 is not a apples to apples comparison.

[tigurius]

Benchmarking a backward compatibility feature is not a good idea. It is somthing like you would benchmark new Athlon XP or P4 processors with DOS stuff. Sure, they can run it, but they are not optimizied for it.

[R.Carter]

I haven't seen this review myself.

That said, I think it is fair to compare Matrox Parhelia 4xFSAA to GeForce4 Ti 4600 4xFSAA. As they say, its just a apples to apples comparision.

I feel that they should also have compared the quality of the output between the two cards as well. After all, anti-aliasing is all about improving the visual quality, and one implementation might yield a better result.

The Matrox 16xFAA is one of the major selling features of the new Matrox Parhelia. It is also a totally new method of anti-aliasing. I would expect the readers would be interested in how this method compares to the 4xFSAA method.

Also, there is this feeling that the 16xFAA feature has some compatibility issues. I know in the July 2002 magazine Maximum PC appeared to be unable to get it to run on one of their tests for some strange reason. So these rumored compatibility issue would also be of interest to readers.

I would also expect that the readers of Maximum PC might be interested in how well 16xFAA performs and what the pros and cons of it are.

Unless the reviewer has provided some justification in the review as to why they neglected to test this feature (ie. it doesn't work) I think it should have at least been tested as part of the review.

[[Ch]amsalot]

Dear xortam and others who believe the review was.... uh... "fair,"

I'm not inclined to repeatedly bash my head into a brick wall. This is either a you see it my way or you don't see it my way kind of issue. There's no need to argue.

Simply put, the review didn't mention Parhelia's 16xFAA. Zippo-nada-nothing was mentioned of it. The author leaves the reader with the clear impression that Parhelia simply sucks as an antialiasing solution. Not only is this factually inaccurate, it's horrid journalism.

Moreover, I stand by what I said about comparing Parhelia's 4xFSAA with NVIDIA's. It is neither a fair nor meaningful comparison. No Parhelia owner in their right mind is going to use 4xFSAA when 16xFAA is both faster and visually superior. The only exception will be specific games that have problems with running 16xFAA. (I note that PC Magazine used 3DMark2001 as a benchmark, and hence could have and should have used 16xFAA).

As for the rumors of 16xFAA not functioning in all games, it is true that there have been a handful of problems which have been noted by responsible reviewers. But with specific reference to MaximumPC's inability to run QuakeIII with 16xFAA, everyone should be advised that Will Smith barely had any time with the card for the July preview ("one afternoon" in fact), and he was using very early drivers that didn't support any form of AA in QuakeIII. Tellingly, others have 16xFAA working just fine in QuakeIII:

Tom: In Quake 3, by contrast, Parhelia's 16x Fragment mode leaves no room for complaint. The following shows a quality comparison:

http://www.tomshardware.com/graphic/02q2/020625/parhelia-19.html

Haig: We also have another switch to force FAA 16x on even if the game doesn't support it and it runs fine on Q3.

http://forum.matrox.com/mgaforum/Forum8/HTML/001254.html

MPC will be having a nice review of Parhelia next month (September 2002) issue. You can believe that Will will beat the card to a bloody pulp with benchmarks and compatibility tests and report any image quality discrepancies he finds.

In conclusion, while I certainly can not fault a reviewer for publishing scores for both 4xFSAA and 16xFAA, the bottom line is PC Magazine utterly failed to report anything with respect to16xFAA. Should you disagree with me that such an omission is outright incompetent reporting, see my first sentence of my first paragraph above.

-[Ch]amsalot

[Ryu Connor]

16th July 2002 21:40, Ryu Connor said:

That has been tried in this very forum before and it didn't go well. Your average gamer does not understand the difference between the AA methods used by each of the current graphics titans.

Needless to say comparing the 4X FSAA OGSS of the Parhelia to the 4X MSAA OGSS of the GeForce 3/4 is not a apples to apples comparison.

On 17th July 2002 07:31, R.Carter said:

That said, I think it is fair to compare Matrox Parhelia 4xFSAA to GeForce4 Ti 4600 4xFSAA. As they say, its just a apples to apples comparision.

Thanks for proving my point.

I feel that they should also have compared the quality of the output between the two cards as well. After all, anti-aliasing is all about improving the visual quality, and one implementation might yield a better result.

The 4X FSAA OGSS mode of the Parhelia will provide superior image quality to the 4X MSAA OGSS of the GeForce 3/4. As the Parhelia will be upsampling (and thus filtering) textures before their division and output to the monitor. The 4X MSAA OGSS of the GeForce 4 has no impact upon textures. Texture data is instead reused in order to save resources and increase speed.

Also, there is this feeling that the 16xFAA feature has some compatibility issues.

http://www.matrox.com/mga/products/tech_info/pdfs/parhelia/faa_16x.pdf

Matrox mentions it in their own white paper, but glosses over where the difficulty lies. If I understand correctly the problem is with 3D applications that use the stencil. There's an option in Q3A (has to be enabled via the command line) to turn on the stencil buffer for more realistic shadows. In reality it over does the shadowing a bit, but it it's not completely unpleasing to look at. Traditionally turning on the stencil killed performance to the point that it wasn't worth using.

Maybe at some point I'll be bothered to turn on the stencil and try 16x fragment AA and see what happens.

Being that it's an edge based AA it will not clean up the jaggies on alpha textures nor does it filter textures to get rid of texture shimmer.

http://www.3dcenter.de/artikel/parhelia_aatest/

The algorithm also apparently has difficulty with certain angles. This of course may seem somewhat negative and that's not the case. You asked a pointed question and this is the straight answer. All the AA options have various downsides. I think the Parhelia 16x AA has a pretty significant leg up versus the others options.

[Ch]amsalot:

Nicely done.

[[Ch]amsalot]

Ryu,

Thanks for your props. BTW, I never had the chance to thank you for your response:

Supersampling does a texture read for every subpixel read. So in 4X it makes four subpixel reads and four texture reads. Multisampling make one and only one texture read per subpixel group. So in 4X it makes four subpixel reads and one texture read. That single texture read is then applied to all four subpixels.

Because there is only one texture read it doesn't receive the benefit of being sampled multiple times as in supersampling, which would provide a boost to visual acuity. The end result of multisampling is that subpixels become more accurate and the textures remain untouched. We get flatter edges and better performance.

I understand some of what you said, but I'm not sure how supersampling a texture would improve visual acuity over multisampling's technique which apparently just dumps the same texture back onto the final product (i.e., textures are basically unaffected). What else am I not getting here ???

-[Ch]ams

[Ryu Connor]

I understand some of what you said, but I'm not sure how supersampling a texture would improve visual acuity over multisampling's technique which apparently just dumps the same texture back onto the final product (i.e., textures are basically unaffected). What else am I not getting here ???

Your welcome.

http://firingsquad.gamers.com/hardware/multisamp/default.asp

Take a look at his OGSS example.

1. The engine and API define the scene.
2. The scene is then rendered upsampled. In 4x this means that the 800x600 just became 1600x1200. Everything is getting upsampled the polygons (subpixels) and the textures (pixels).
3. Now the card moves to antialias the image. It starts making subpixel and pixel reads. It takes the four points and divides them together in order to bring the pictures back to 800x600. So by raising the resolution and making the polygons finer, then taking four values from that finer result and dividing them together it creates a smoother less jagged line.

It's pretty much the same story for the textures. You're upsampling them to a higher resolution. Then you are pulling four points. It then divides those points in order to bring the texture back down to the smaller resolution image. The side effect of that is that texture is refined. The division of the four points has provided better data that improves color accuracy and removes some of the noise (think bilinear or trilinear filtering).

You can actually try this for yourself. Go up to a wall or some other long object in one of the Serious Sam games with a GeForce 2 card. Be running in 800x600x32 with 4X FSAA, Trilinear, and 2-Tap anisotropic filtering. Take a screen shot of the long object so that you can see how it goes from highly detailed in the foreground to blurred in the distance.

Then do the same thing with a GeForce 3 or 4 without AA but with Trilinear and 2x, 4x, and 8x aniso filtering. Then compare the images.

You should find that textures are on the GeForce 2 remain cleaner and detailed farther out than a 2x shot and rival the 4x shot.

As for your explanation of the GeForce 3/4 MSAA it is essentially correct.

[Wombat]

Ha ha ha. Silly foft. You think Matrox actually understands/cares how to benefit from alternative OSes?

[Reckless]

Just thought I'd chip in with some abuse...

This thread (especially the starting post) is almost as worthless as the review (I expect). Matrox will not win many NEW customers with the Parhelia. It's $400 - u know GF4/4600 price bracket - and big numbers (benchmarks) sell to the masses. The people that know and care what FSAA and FAA is is about will already know the score before reading the review and they really don't need yet another review with the same load of screenshots. They'd have most likely made up their mind to go with large numbers (nVidia) or the promise of reasonable numbers and better quality.

You seem to think Matrox may possibly lose customers over 1 review? Maybe but very unlikely. Anyone who's got $400 will likely research his/her purchase with more than one review (well, one would hope so and if they don't it's their bad!).

Besides, how about take this stance.... IF they (the mag) really wanted to show the P in bad light then why not pick a game where FAA doesn't work or seriouly slows the game down (going by what's been said on this board). No, they didn't take the attach approach, they merely compared 2 comparable AA mechanisms instead of the P's more resource effective method. It was fair in a sense but obviously u didn't agree!

To end, I'd say you'd better chill out a bit! After all, the review hasn't rewritten the Parhelia story.

[[Ch]amsalot]

Hey Reckless, thanks for your comments

My main point was that it was just a lousy preview, unfair to Matrox. I doubt anyone who is seriously considering Parhelia would pay much attention to PC Magazine anyway

-[Ch]amsalot

[Ryu Connor]

Besides, how about take this stance.... IF they wanted to show the P in bad light then why not pick a game where FAA doesn't work or seriouly slows the game down (going by what's been said on this board). No, they didn't take the attach approach they merely compared 2 comparable AA mechanisms instead of the P's more resource effective method.

Well, in a way they did just that. I'll touch on the non-AA performance aspect a little later, but comparing Parhelia's 4x against the GeForce 3/4's 4x AA is a guaranteed way to make the Parhelia lose. That goes the same for the RADEON as well.

To end, I'd say you'd better chill out a bit! After all, the review hasn't rewritten the Parhelia story.

In general (but not always) the card is slower than a 4600 and if all you do is game, that has some pros and cons you'll have to weigh heavily.

The point of using AA is to improve image quality. Since AA has the trade-off of a performance hit there is certainly a need to see how AA performs on a particular card. The performance is a objective fact. You can even objectively make screen captures to show the difference between the methods. The method a person perfers is usually a subjective choice or is based completely upon the performance of the card.

When it comes to Parhelia's 16x AA we have a situation where the card performs better than it's peers. By removing the 16x AA results and only showing that Parhelia's 4x FSAA performance you have effectively eliminated any results that will show the reader positive features that the card can provide. True, even if the 16x AA results were included the reader might decide the card is not for him, but by omitting them completely you've basically failed to do your job as a reviwer and have effectively made up your mind for the reader. At minimum it's poor journalism due to sloppiness or incompetence. At worst it's called stacking the deck and that kind of bias has no place in reviews.

[Indiana]

Originally posted by ralf
I think xortam is absolutely right, of course it´s a valid comparison. The purpose of including 4xFSAA is of course that people might want to use it.
THG:s review, albeit less than good in other respects, did show pics of 4xFSAA and 16xFAA where you could clearly see that 4xFSAA produced better image quality - not just the edges but overall clarity of the textures. So for the IQ concerned user it could be preferable, hence the validity in benchmarking the performance.

DOH! You sure don't know what you're talking about. Fragment AA does not blur textures, in fact it does not alter the textures in any way. If you're comparing AA quality you should look at the edges and the 16x FAA is superior to NVidias 4x FSAA and VASTLY superior to any of these Quincunx modes (except for slight problems with few games)

The worse texture quality of the Parhelia simply comes from it's non-sufficient anisotropic filtering, but that is a totally seperate issue.