Reading all these Parhelia reviews...

Chalnoth,

My point was that even if all algorithms would work 100% as expected, that the comparisons would still turn out dubious.

If you prefer then Accuview against FAA. Where and why is it a fair comparison when one algorithm maxes out at 4x and the latter uses 16x samples on edges?

FAA having quite a few glitches/incompatibilities is common knowledge though, but I'm touching a totally different principle here.

How about someone drop the resolution on Parhelia as much to fall close to edge quality of that that 4xOGMS can produce on a GF4 in X resolution and then compare performance? Not fair either. That's why I think that it's hard to make a valid and fair comparison when implementations differ too much.

Accordingly you cannot compare just Multisampling with Supersampling, w/o enabling for the former aniso. Who in his right mind plays with MSAA and no aniso anyway? I don't for one.

 

No, it does help.

 

I believe that was for multisampling...

 

There is also "Ain't good enough for what they're trying to charge for it." Obviously the Parhelia is better than my GeForce 2 in just about every aspect, but it's not $400 bettter, and even moreso when there are alternatives for the same price that are more attractive to me (given the fact that I care more about framerates than triple-head and Gigacolor).

Performance is not everything, but I think the difference in performance between the top 3 cards is greater than the difference in image quality between them. You also can't just look at the overall picture, because everyone has areas they are more concerned about than others. Take the following type of rating as an example. The numbers are my rough guestimate of what I have perceived each card to deserve, without ever having used any of them in my computer. The odds of me caring about whether or not you agree with these numbers are about as high as the odds of me buying a Parhelia.

Image Quality (1-10):
----------
Parhelia: 10
Radeon 8500: 9
GeForce 4 4600: 8

Speed:
----------
Parhelia: 6
Radeon 8500: 8
GeForce 4 4600: 10

Features:
----------
Parhelia: 10
Radeon 8500: 8
GeForce 4 4600: 8

Price:
----------
Parhelia: 2
Radeon 8500: 8
GeForce 4 4600: 4


In this example, the Radeon 8500 would win overall, and the GeForce 4 edging out the Parhelia since it can be found for a hundred dollars less at this time. To a person who will never use 3 monitors, doesn't care about Gigacolor, and will likely replace their card again before any DirectX 9 titles come out, Parhelia drops even further out of contention. If you don't care about the cost or speed, the Parhelia is the top choice.

The question is not if people care only about speed. The question is are there people in the world who don't care about the cost or speed, and do they exist in large enough numbers to make Parhelia a successful product. My guess would be no.

 

Thanks for the correction.

 

Gamepc have published their review of the Parhelia. Again the results aren't fantastic but when they max out the quality in Dungeon Siege and 3Dmark it leads the Ti4600. Not by much but by enough. The other benches are more or less the same as previous reviews.

All the benchmarks should be run at max quality for each card ignoring the differences and concentrating on the end result. Thats what I'd prefer to see rather than a standard run that performed well on GPUs of that generation.

3dcgi, that was my mistake I got JCs .plan all mixed up

 

Looks like a good review, and its good to see the Parhelia performing so well under Max Quality conditions, as it validates Matrox' claims!
Yet, is this really a good way to compare two cards? Parhelia does a higher quality AA (16x to 4x) while the GF4 uses a higher quality Aniso (8tap to 2tap) in the benchmarks. I can't really make up my mind about just how comparable these scores are, but maybe they don't need to be as there is no usefull way to compare the cards on an "even ground" anymore ...

 

I guess we shouldn't read much into the numbers at all, they're just a rule of thumb and a way of making pretty charts. It's the look and feel that counts. My GF4 knocks my R8500 into an cocked hat when it comes to statistics but it achieves it in a very different way. The Radeon *feels* smoother when I'm playing games etc even though the framerates are lower. I personally think the Parhelia is achieving what it has set out to do but people are still looking at it from the nVidia inspired 'frame rate is life' perspective, it can do a whole lot more.

 

If FAA can be made to work 100% correctly, then it would be superior to anything other FSAA method we have today, period.

Absolutely, but you have to be careful which degree of aniso you enable. 4x SS should have roughly the same texture impact as 2-degree anisotropic.

 

I don't have a problem with max quality comparisons, what I fear is if the "end result" is only concentrated on performance and conclusions drawn only from those. Recent example being numerous reviews on Xabre. Funny how only digit life and xbit labs noticed a difference in texture quality.

If then Xabre gets same or more performance than competitor X, with a complete blurfest on default, then the reviewer calls it a day, Xabre is faster and everyone is happy.

 

No objection at all Chalnoth.

edit: sorry Reverend, couldn't help it

 

doh! and i received bloody amounts of beating on the head for starting positive sentences with 'no' and negative with 'yes' while i was studying english as a kid!..

 

i did have to stop myself mentioning an irate Wavey again

 

I keep reading people complaining that Faa 16x isn't fair game when benchmarking against the 8500 or the ti4600 , It is so

the benchmarks should be ran at the highest possible setting for the cards ( too bad if the 8500 or the ti4600 don't go up to 16x faa ) but you have to keep the IQ in mind ( and yes the parhelia AF level is locked in the drivers )

 

I simply disagree with this statement. FAA is again only edge AA. It is not better To Supersample methods when considered as a whole imo.

One other thing i noticed at Nvnews today. Type, made quite a news post regarding parhelia being better when max details are applied. I have a serious problem with that. First the numbers are frankly unplayable except for dungeon siege. Secondly It is totally unfair to compare Matrox 2x aniso to Nvidia's 8x ansio. It looks worse and if the ansio on the PAehelia was running 8x would be slower by quite a margin.

Look, I totally understand why some people are backtracking, or trying to be overly posotive after the initial response to Parhelia. However, making unfair comparisons or calling unplayable numbers a victory worthy of a purchase is simply wrong. Parhelia's Supersample method is VERY much slower than Nvidias offering. Most Nvidia users are playing with 2x FSAA and 4X-8x aniso. This delivers reasonable IQ and more than reasonable FPS. No Real gamer will EVER run anything but a RPG at sub 60 FPS and be happy. The dips for a 30-50 FPS average will be to inhibiting to play.

If this is the way its going to be handled, then I want all reviewers including type to admit that ATI aniso is the best in the industry for Performance/IQ and bencharks shoing that ATi's 8500 beats the 4600 in every benchmark known to man when max aniso is applied to both.

Until that happens, I want a little fairer reviewing done, even if it does not reflect well on Matrox.

 

Matrox the scapegoat?

 

Reverend,

I think there is more to a review than whether or not the card plays games well. Graphics card reviews, for the most part, have been about which card is better than who at what and who is the best at which time (Just like other PC Hardware reviews). Unless you're reading the review from PC Gamer or some other magazine/website more gaming oriented, you're probably going to be reading a technical review regarding it's technology and how it compares to other cards. I don't think anybody on this messageboard buys videocards based on it's own merit, but rather on it's price and performance relative to the competition. Unless, of course, you're on of those fa-... eh, nevermind.

Anyway, "playable 60fps" vs. "rediculously fast frame rates" is the same as the new features in the videocards that aren't being used. Like T&L and Pixel/Vertex Shaders. Many people went, "Who cares about those? It's features are never going to be used anyway!" But then on the other hand, you have a whole lot of people scouring websites, reading the latest articles, all in the interest of finding out how it works, what it does, what it's used for etc; not for the sake of gaming, but for the sake of the technology itself. The matter of whether or not it will be used in games is another perspective all together.

Now, with Parhelia, it's not performing up to hype and a lot of people are disappointed that it isn't lightening-fast. Then back here on the Beyond3D forums, we have people going, "Why?! It's fast enough to play the damn games!" Wow, that's nice, but that isn't the point anymore. It's whether or not it's better than the competition, and I thought Beyond3D was a technology messageboard, not a gamer's messageboard. If you're buying the video card to simply 'play the games' for the sake of playing games, you probably don't even need the latest and greatest video card - or even need to know what the clock speed of the video card is running at, how Anistropic Filtering works, whether or not Matrox's 16x AA works on alpha-textures, etc.

Videocard consumers may have been, at first, only gamers, but now there is another aspect of consumers - those who just want the latest technology to blow out the rest of its competition. The gamer may want 60fps, but the technology freak/power user wants rediculous frame rates that he can't even recognize, functional new features that he will never use, fill-rate that will never be required, or like a few years ago - two videocards working together that will never be stressed.

There are really only two perspectives, the gamer's perspective and the technology perspective. I've always known Beyond3D for it's technical-perspective articles. From Kristof's article on explaining TBR and 3Dfx's 22bit post filter to Daves MultiSampling FSAA article to your rendering interview with John Carmack. With all this "60fps is all I want" talk, are you telling me you're going "gamer" on us?! (Okay, I admit, I went "gamer" as I hardly ever care about the latest tech, anymore. But I don't write for a tech-site! )

-dksuiko

 

lets not forget how the radeon8500 was whenit first came out.
Wasn't as bad as matrox but rough launch.

 

I was just browsing on the Matrox forums (not MURC) and i found a very interesting little thing which Haig said, questions are in bold and his comments are in italics:

I am worried about this one http://www.anandtech.com/video/showdoc.html?i=1645&p=7 At 1024x768 62Fps there 53 on the next page.

On our XP1800+ system under 2k, we're getting 77fps with FAA forced and Aniso on at 1024x768@32bpp.

ok why do you manage to get 77fps and all these sites are lower?

No idea. Ever since I got Parhelia in my dep't, I had all systems reformatted, and disabled all the useless services for my dep't. Perhaps some editors haven't done this, or it could be the fact that we aren't using the same map to benchmark. Epic has a special benchmark app for ut2003 which an nda was required. If these sites have the same benchmark as I do, then something is wrong somewhere.

Right the benchmark Haig is rerfering to seems to be the UT2003 performance test and isn't this test suppose to be alot more demanding then the test Anandtech has done as they have used the actual demo NOT the Performance test. If you goto http://www.anandtech.com/video/showdoc.html?i=1645&p=16 and look at the last graph we can see at 1024x768 with ansio and FAA the parhelia gets 30.4fps now i would of thought the performance test would be more demanding then the test anandtech has done.

Have i missed something here as Anandtechs results are quite far off what Haig gets, the only differences i can see is the benchmark used, anandtech uses athlonxp 2100+ and winxp and haig maybe using newer drivers but i wouldn't of thought that would make this kind of difference. if im getting something wrong here can someone explain to me what it is thanks