What are your first impressions of the Parhelia ?

[APEXNETHOR]

Originally posted by Zyn
Im waiting for someone, hopefully a hardocp review to peltier watercool the parhelia and bring the gpu up to about 300+mhz.

hey why not?
that should kill some of the slack off.

Now yer talkin hehee.

[blade-be]

well, the results are in, it sure looks as though the big M dissapointed quite a lot of people out there ...
what do you guys think the main reason is? Matrox bragging too much or us hoping too much ?

[CHHAS]

Definetely the last part, Matrox has IMHO lived up to everything they promised.

[afterburn]

Im not really disappointed. I had hoped for more, but one can always hope. Actually, the more I read about P., the more enthousiastic I get. I'm just waiting for it to hit the shelves and I can get some real people's experience with the card and espcially its drivers. After all, it wouldn't be the first time M's drivers aren't quite as they should when they are released. That said, you always hear more bad stories than good and I for one never had any problems with the G400 drivers even though I was pretty early getting one of those.

[Cheesekeeper]

Originally posted by blade-be
well, the results are in, it sure looks as though the big M dissapointed quite a lot of people out there ...
what do you guys think the main reason is? Matrox bragging too much or us hoping too much ?

Neither. Matrox weren't able to ship cards running at the same speed as the Alpha boards, due to heat problems when a smaller cooler was attached. If the shipping cards were able to run at the expected rates, the benchmarks would have looked a lot better. It is a pity that they didn't take a little extra time to look at a better cooling solution, rather than attach a baby heatsink, and crank the clock down though......

[gt40]

bunch of whiners! (you know who you are)
I for one am extremely impressed with the technology in this puppy.
Ok it isn't a fps card, but frankly anything above 30 fps (constant)is overhead to the human eye - but anything that enhances the visual quality is eye candy even at 0 fps!
My main reason for having this card is for CAD and photo/video editing, all of which will gain benefit from the perhalia.
The onlythings stopping me at this point is the price, and the fact that my rainbowrunner won't work with it ( anybody have a rt2500 cheap )

Great Job, Matrox

[WaltC]

Originally posted by blade-be
*most* of your *current* games
what about doom3, ut2k3, ... ? I'm wondering how this baby is gonna perform on future games. If it can hardly keep a decent framerate with current games with the features that make this card worthwile (16x FAA, 8x aniso), how the hell is it gonna run future games at a decent level ?

Where did you hear the P supports 8x AF? Everything I've read thus far says that the current drivers support 16-tap only, which is 2x. The reports are mixed as to whether this is a "driver bug" or a deliberate limitation employed by Matrox so as to limit perfomance degradation.

I for one would really like to know what's what with regard to anisotropic filtering and the P. I put up a thread in the gamer's forum but thus far no one has responded definitively.

[WaltC]

Originally posted by moreau

http://www.anandtech.com/video/showdoc.html?i=1645&p=15

If that is true, then it doesn't look so good for the GF4 Ti 4600, either.

1024x768, W/Anti-Aliasing

Parhelia = 43.6
4600 = 42.8

matrox is using 16x AA while the Nvidia uses 4x.

Cheers

Which doesn't mean a whole lot really because 4x FSAA is AA'ing a ton more pixels per frame than 16x FAA.

Compare P's 4x FSAA mode to the GF4's 4x FSAA mode to get apple to apples.

[WaltC]

Originally posted by Mumrik



Only according to you guys...

Read these graphs

I'm not convinced...

I saw this, too, but it's difficult to understand your comment given that the minimum fps looks ~38 and maximum is near 80 for Serious Sam SE. I haven't read a review yet which talks about "chugging" game play with P. The fact is that you don't need even 100 fps to get smooth gameplay. (I own a GF4 4600.) 38 fps certainly won't "chug."

[WaltC]

Originally posted by Myrv
Actually what's annoying me the most is the requirement for .NET

What were they thinking? How is this card going to work under Linux?

They've said they'd be releasing Linux drivers for P, so that's not a concern. However, putting up with .NET under Windows seems a negative to me--I really don't want to install .NET.

[TdB]

well, if you look careully on those graphs, you will see that the minimum framerate drops about 7 fps going from 1024*768*32 to 1600*1200*32 while the high-peaks drops about 15fps on parhelia, thus the fps delta and the avg fps gets smaller without hurting the lowpeaks much.
on the gf4ti the fps delta is huge in all resolutions, meaning that the gf4ti framerates drops more linearly with resolution increases, including the lowpeaks, that is the avg fps the highpeaks and the lowpeaks differences stays big, so that the difference in lowpeaks between the gf4ti and parhelia decreases as the resolution goes up.

and about the serious sam test i noticed this:

we let the application's self-tuning mechanisms do their thing

do we know if serious sam detects the parhelia properly, what if the self-tuning doesn´t work, I think the self-tuning is hardcoded in the game, and it doesn´t looks like all 4 TMUs is in use in that test.

[WaltC]

Originally posted by stmok
MATROX NEVER STATED : THIS IS A GF4 Killer or a BRUTE FORCE
FPS dominating card. If you believed that, you are gravely mistaken and will be extremely disappointed. (As many of you are!)

I agree. Many have not listened to Matrox at all on this point. Rather, they've made their own independent decisions relative to what P should be and so it's hardly surprising they're disappointed. This product has many other uses and applications besides 3D gaming, but at the same time is the best 3D gaming card Matrox has ever made.

And if you're expecting to run DOOM III on a Parhelia with high frame rates, DON'T! You'll be disappointed here as well!
If you want all out performance, upcoming cards with ATI's R300 or Nvidia's NV30 and NV35 are the choice.

Truthfully, though, I think the concern is whether something like Doom 3 will be playable on the P. At least for smooth game play with all of the eye-candy turned on. I think that's the real concern there.

I personally think the Parhelia is great for my needs.

I study Aero Engineering and run alot of apps such as CAD, Finite Element Analysis, Matlab, etc. Sharp, high quality images are a benefit here for me.

If your concern is primarily 2D, you can't beat the P, I think.

Most of the time I'm in the Windows desktop, surfing the web, watching a movie or edit some video clips.

Occasionally, I play Jedi Knight II. I'm happy with 50fps and high quality settings at 1024x768x32bit. Highest frames per second as possible are not a requirement for me.

Obviously, your uses are different and gaming is your priority, hence your disappointment in this card.

That's about the size of it for gamers, I think.

Performance-wise...Put it in Jedi terms...

Parhelia = Jedi Apprentice
Radeon 8500 = Jedi Knight/Master
GF4 Ti 4600 = Jedi Knight/Master

The Apprentice has "not yet in full control of his Force powers , his/her light saber techniques are not fully mastered" (THE DRIVERS!), and is in effect as not as effective as the Knights or Masters. (Low benchmark numbers)

Matrox itself expects drivers to improve performance 10%.

So when you dumped them all in a Jedi vs Sith "bar fight", don't expect the Jedi Apprentice to beat a a Sith of a higher level in an all out match. [/B]

I don't think that's all that good of an analogy because there are some areas (Tri-head support and surround) where the P is the indisputable Master and the other guys aren't even in the game at the apprentice level...

[Ryu Connor]

http://www.matrox.com/mga/products/parhelia512/technology/supersample.cfm

It supports 8-Tap anisotropic filtering, which with bilinear filtering as a base will make for thirty-two samples. With trilinear filtering as your base it will make for sixty-four samples.

A driver issue at current prevents use of higher than 2-Tap anisotropic filtering. Which is eight samples with bilinear and sixteen samples with trilinear.

Compare P's 4x FSAA mode to the GF4's 4x FSAA mode to get apple to apples.

This is a common error. The GeForce 3 and 4 series of cards do not do Full Scene anti-aliasing. They do MSAA or multi-sampling antialiasing. MSAA is a edge based technique that can use various tecniques to acheive it's goal.

2X AA on the GeForce 3 and 4 grabs edge data and puts it through a Rotated Grid scheme.
Qunicux AA uses rotated grid, but then samples three points in order to add a bit of blur to the image.
4X AA uses a ordered grid scheme.

Regardless of the scheme used to antialias the GeForce 3 and 4 cards are always doing just the edges of the scene not the entire scene.

The Parhelia's 16x FAA is a edge based scheme also, so comparing it against the MSAA of the GeForce 3 and 4 is completely fair. Meanwhile the Parhelia's 4X FSAA is just that, Full Scene Ordered Grid antialiasing, which is akin to the SmoothVision the ATI serires uses and the antialiasing found in the older GeForce 2 series of cards.

http://www.tech-report.com/reviews/2001q2/geforce3/index.x?pg=16

http://www.tech-report.com/reviews/2001q4/vidcards/index.x?pg=14

http://www.tech-report.com/reviews/2002q2/parhelia/index.x?pg=9

[WaltC]

Multisampling is not "edge AA"

Originally posted by Ryu Connor
[B]http://www.matrox.com/mga/products/p...upersample.cfm

It supports 8-Tap anisotropic filtering, which with bilinear filtering as a base will make for thirty-two samples. With trilinear filtering as your base it will make for sixty-four samples.

A driver issue at current prevents use of higher than 2-Tap anisotropic filtering. Which is eight samples with bilinear and sixteen samples with trilinear.

Here's a quote from the Tech Report preview (which you should know by heart...):

http://www.tech-report.com/reviews/2002q2/parhelia/index.x?pg=10

The best sort of texture filtering we tend to see is anisotropic filtering. Unfortunately, with current drivers, the "most advanced texture filtering units" can't do better than 2X (16-sample) anisotropic filtering. I noticed this limitation and asked Matrox about it, and they confirmed to me that current drivers are limited to 2X aniso for performance reasons. The hardware can do 8X (64-sample) aniso, and Matrox is considering enabling that capability in future drivers. Given that the GF4 Ti can do 8X aniso and the Radeon 8500 can (with some caveats) handle 16X aniso, I think enabling stronger forms of anisotropic filtering would be wise.

In other reports (like yours above) I've read the current limitation is a driver bug which Matrox plans to correct. So which is it? Is it a bug which will be corrected, or is it a feature deliberately turned off in the current drivers for performance reasons, which Matrox "is considering" turning back on? Do you know? This is the question I've been asking.

This is a common error. The GeForce 3 and 4 series of cards do not do Full Scene anti-aliasing. They do MSAA or multi-sampling antialiasing. MSAA is a edge based technique that can use various tecniques to acheive it's goal.

2X AA on the GeForce 3 and 4 grabs edge data and puts it through a Rotated Grid scheme.
Qunicux AA uses rotated grid, but then samples three points in order to add a bit of blur to the image.
4X AA uses a ordered grid scheme.

Regardless of the scheme used to antialias the GeForce 3 and 4 cards are always doing just the edges of the scene not the entire scene.

Sorry, but that simply isn't possible for three reasons I can think of off-hand....

First of all, I get clear, noticeable and objectionable full screen blurring when I set my GF4 4600 to 4x FSAA--there's simply no way to deny it. I don't ever use it as a result. If all that was being done were "edges" I would never get the kind of full screen blurring I get here. In 2x FSAA, I get much less blurring--but I still get blurring that I can notice. And so I don't use 2x FSAA, either. (By contrast, with my V5 5.5K I got very little blurring if any in 2x FSAA, and more obvious blurring in 4x FSAA--although nowhere near as much as I see with my GF4 4600.)

Second, one of the strong points to 3dfx's hardware Rotated Grid scheme was the treatment of near horizontal and vertical lines. The V5 5.5K handled these and AA'ed them very well. On my GF4 they simply are skipped--and they look completely non AA'ed. This is actually a common symptom of ordered grid FSAA--certainly not any type of rotated grid FSAA I'm familiar with. I've noticed this in both 2x and 4x FSAA. If nVidia is doing some sort of rotated grid, it certainly isn't the calibre or type of rotated grid FSAA 3dfx did in hardware with the V5--that was far better. But in any event "rotated grid" just like the term "multisampling" has nothing to do with edge AA, anyway. "Multisampling" btw, simply means you don't have to take a sample of every pixel on the screen in every frame, it means you use copies of like pixels across frames to buttress performance. It has nothing whatever to do with edge AA. I don't know if that's what you mean by "multisampling"--edge AA--if so, it's the oddest and most incorrect definition I recall hearing.

Third, just check the performance differential in terms of percentages between the Parhelia's no FAA mode and their 16x FAA mode compared with nVidia's no FSAA mode and it's 4x FSAA mode. You'll see that the percentage of perfomance drop is far greater in the GF4 than it is with the Parhelia (which we know is doing edges.) If GF4 was merely doing "edge AA" we would expect to see no blurring of the entire image, and we'd expect a similar performance drain because the numbers of pixels would be similar--actually, though, since the GF4 is only doing "4x edge AA" by your accounting, the percentage of drop should be far less in the GF4 since the Parhelia is doing 16x FAA. The performance percentages cannot lie: by your accounts the GF4 should be AA'ing roughly the same number of pixels as the Parhelia, but using only 4 samples, whereas the Parhelia uses 16 samples, therefore the Parhelia should suffer a far greater drop in performance than the GF4 when AA'ing. But in fact we see just the opposite occur. The reason is abundantly clear and obvious: the GF4 is AA'ing a great number of pixels more, hence the performance drop is much greater on the GF4 even though the GF4 is is taking 1/4 the samples per pixel and on a multisampling basis to boot.

I'll throw in a fourth and fifth reason here:

Four, common sense clearly dictates that if nVidia was doing merely edge AA that they would not restrict themselves to 4x, but would at least go to 16x like Matrox has done for the same reason Matrox did. Performance.

Five, I've seen several pictures on web sites which have been magnified a couple of times, and in which nVidia's 4x FSAA and Pahelia's 16x FAA have been compared. In every example that I've seen, in the magnified images from Parhelia, you can plainly see where several pixels have been missed--this is exactly what you'd expect from an "edge" algorithm, because such an algorithm will never be able to get all of them in every frame. However, in the GF4 4x FSAA images, ALL of the pixels are AA'ed, without exception, so that when magnified there are none missed. You can only get this kind of coverage when you are AA'ing all of the pixels--there simply can be no algorithm to get them all on an "edge" basis. Also, even when magnified, the image portions all appeared slightly blurred compared to the Parhelia, in which you see blurring only at the edges in the magnified images, where you'd expect it to be in an "edge-only" algorithm.

I see no evidence whatsoever that the GF4 is doing edge AA of the type the Parhelia is doing. Moreover, the evidence that the GF4 is continuing with nVidia's trend of FSAA (full screen AA) is obvious and compelling, at least to me... We can discuss multisampling a bit more if you're still fuzzy on that--however, you may not have meant to imply that "multisampling" means "edge AA" at all, and if you didn't, my apologies. But in that case I have no earthly idea where you picked up "edge AA" from in the first place. MSAA != edge AA.

[Ryu Connor]

I'm immensely disappointed at this very moment Walt C. You didn't read the links I provided, did you?