the big parhelia review conclusions thread

[Bilbo]

Just my 2p.
But who gives a monkeys whether it gives 97fps in Quake III.
Quake III is old hat now and any of the games using its engine have tweaked it with extra features that are hard on the video card. RTCW and Jedi Knight II to name two. The one benchmark I saw for RTCW (sorry seen so many today that I can't remember where) had the Parhelia at less than half the framerate of the GF4 Ti 4600 and would be noticebly slower and less playable.
You can say whatever you want but it is noticeable when framerates are under 60fps. So to get the best out of your gaming you need 60fps+. As a next generation card it should be able to hold its own on at least current generation games, not last generation. When DoomIII comes out then from current stats it looks unlikely that it will be playable. The frame rates showing up on these reviews are very lame for such an expensive card.
And before you flame me, I've had every Matrox card since the Millenium including both versions of the Mystique and been a staunch supporter all this time, but after the G400 they have seriously lost it.

[Afronaut]

Originally posted by nggalai
Regarding your question: the Parhelia seems to be very little bandwidth-limited (surprised, anybody? ), hence higher resolutions won't tax it as much as the GF4 Ti4600. Bummer is that as of now, those higher resolutions with FAA are too slow to be interesting for most gamers.

ta,
.rb

Currently FAA will not work with any use of stencil in a game, which is one thing Epic had to disable in the UT2003 demo in order for FAA to properly work. Although the demand isn't necessarily great for UT2003 to support stencil right now, eventual support is necessary and if FAA doesn't properly work with it enabled then Parhelia will be forced into 4X supersampling mode which is no better than what the Radeon offers.

Why is it so hard for some to face up to the facts?

Do I have to go back and pull the defending threads for Head Casting?

[thop]

Originally posted by Bilbo
When DoomIII comes out then from current stats it looks unlikely that it will be playable.

i'm looking forward to hear John Carmack on that if Matrox lets him. maybe D3 will use more of the advanced features of the parhelia but even if i dont see how it will be playable when JC says the 4600 will probably get around 30FPS.

[mallowman]

I am having a very hard time reconciling what the reviews are showing with the chip tech in the whitepapers. On paper this chip looked so much better than it has turned out and while it is clocked low compared to other cards the performance features seemed that they ought to make up for a lot more of it than they did.

I'm going to reference some of the 3DMark test and while I agree that the overall 3DMark number doesn't really give much information, the results of the tests that give something besides FPS as their measurment do seem to contain useful results.

<ul><li>Massive vertex shader array
<p>The Parhelia has twice as many vertex shader arrays as any other card, so shouldn't it be able to push twice as many transformed vertexes per clock as any other card too? Even with the reduced clock speed it should outperform other cards for triangle throughput, but it clearly loses the triangle tests in 3DMark. I'm assuming that vertex shader units are what is responsible for hardware TNL, but maybe I am wrong about that. Is it a problem with the pipelines being too deep kinda like the P4 has? Are they optimized for DX9 vertex ops and therefore slower for DX8/OpenGL because of some sort of translation in the drivers? Is it possible that somehow a massive driver bug turned off TNL or made it work real slow or something? Maybe someone else has some idea about this.</p></li>
<li>512bit GPU
<p>This seems to have been sort of a bust. I was expecting something like tricks of add/mult/div 16 32-bit numbers in one 512 bit register in one clock giving a big boost. Or maybe some operations where whole homogenous (x,y,z,w) vectors are loaded into a register and a transformation is applied to the whole vectors at once (4 of these vectors would fit at 32bit precision into a 512 bit register), dramatically increasing transformed coordinate throughput. At least it was my understanding that being a 512 bit GPU it would have 512 bit registers and could do tricks like this.</p></li>
<li>Anisotrophic filter fiasco
<p>If the filtering level truly had to be capped because of performance limitations then it raises some serious questions in my mind. Was the chip overengineered? Were so many really good features built in that it became unwieldy and too complex and thus couldn't get clocked high enough to get the performance planned?</p></li>
<li>Equally massive pixel shading array
<p>Is this another instance of too deep pipelines to be able to do the common/simple case really quick or in reduced clock ticks and thus with the lower clock speed can't get real competitve performance in either the simple or complex case?</p></li></ul>

I was really excited about the Parhelia and still am. It has been said that we will have to wait for games that will take advantage of P's advanced features, but I don't see how the performance will go up any, just not take as big a hit as other cards. Granted this will make the relative performance of the Parhelia look better, but will it be enough?

In my eyes it looks like the Parhelia is a superbly engineered product that may have ran into some physical boundaries of the chip making process. It seems like the architecture is constrained by temperature and die size and thus has to be clocked lower than optimally. I hope someone has some thoughts on the above but I suspect that it will take someone with intimate technical knowledge of the Parhelia's architecture to get real answers to the questions, which pretty much means a Matrox engineer, who probably wouldn't be allowed to answer these questions (at least not if they want to keep their jobs ).

(Sorry that this post is so long, but I've been wondering about these things for a while and only got around to writing them down now)

[Mumrik]

Well, quite frankly, Amen mallowman...

[aldamon]

Sorry guys, but the whole Parhelia launch smells like the putrid Voodoo5 launch. All of the benchmarks are terrible and everyone who is supporting this card refuses to look beyond suspect FSAA (UT2003), subjective IQ, performance in THREE-YEAR old games and vaporware features (remember motion blur and T buffer?).

I have to admit I was extremely intrigued by this card, but now I wouldn't even buy it for $200, let alone $400. Damn you Matrox!

[Ryu Connor]

What disturbs me about all this is the cards performance when the CPU is no longer part of the performance picture. Take a look at the Chip.de review for a great example.

Look at the bare results where it's just resolution comparisons. Parhelia loses across the board. Normally you'd just chalk the situation up to not having enough fillrate or bandwidth for the processor to max out.

Then you see the resolution + AA scores. At this point the other cards are starting to fall like flies because you are devouring everybit of fillrate and bandwidth they have. Yet Parhelia moves to perform faster than them. Sometimes twice as fast and sometimes just a hair quicker. So in the set of just a few seconds you've watched a card that appears to be fillrate and bandwidth limited show that it is anything but fillrate and bandwidth limited.

Maybe a great deal of it is due to the fact that the Parhelia's four texture per pipline is being limited to two texture by current software. The texture limitation may be partly to blame, but what it really looks like is that the current driver just isn't passing all the information. It gives the impression that Parhelia is just sitting there and twirling it's thumbs most of the time.

Of course, I could be completely wrong and the chip just isn't up to the task of crunching fast enough. The discrepency between the CPU dependent and fillrate dependent results though, just leave me thinking that the Parhelia driver may be very raw and unoptimized.

[Sinistral]

Originally posted by aldamon
everyone who is supporting this card refuses to look beyond suspect FSAA (UT2003), subjective IQ, performance in THREE-YEAR old games and vaporware features (remember motion blur and T buffer?).

- And everyone who is slamming the card fails to realize what the card has to offer other then Super High Frame Rates in First Person Shooters and gaming all together. -

[notagain]

good news ? or

http://forum.hardware.fr/forum2.php3?post=501547&cat=1&interface=0&config=&page=1&cache=cache&p=1&sondage=&owntopic=&subcat=#t1557041

well new drivers seems to have pumped up the volume a bit.
if we get +10% each day then it wont be so bad wont it be?

[aldamon]

Originally posted by Sinistral


- And everyone who is slamming the card fails to realize what the card has to offer other then Super High Frame Rates in First Person Shooters and gaming all together. -

Yep you got me. $400 for a card that performs as fast as a GF2 in some situations and all you need to take advantage of its amazing "extra" 2D / 3D features is a Photoshop plugin, MORE money to buy two more monitors and a massive push from the developer community.

Boy you got me.....Matrox is brilliant

[Dr Mordrid]

You're ignoring the fact that the low frame rates are mainly due to the absense of an outdated culling methodology that is used in these older products.

Since Matrox has issued a DirectX8 SDK for using Displacement Mapping and Distance Adaptive Tesselation new games can start using it right away.

Also: DirrectX9 will be using DM and DAT so Parheia's best performance lies in the future. Unless NVIDIA and ATI use this Matrox-licensed technology they will be the ones stuck in the past

Dr. Mordrid

[Sinistral]

Originally posted by aldamon
Yep you got me. $400 for a card that performs as fast as a GF2 in some situations and all you need to take advantage of its amazing "extra" 2D features is a Photoshop plugin, enough MORE money to buy two more monitors, and a massive push from the developer community to support its 3D paper features.

Boy you got me.....Matrox is brilliant

- Ok, how bout being able to play a Stragety game on 3 Monitors. Offering more view of your terrain (there are games on PC's that arent FPShooters...)? Offering pretty much 3 times the desktop workspace. And im pretty sure the Proffesional market shadows the Gaming market quite easily.

And just my 2Cents about the whole situation. If you arent going to use the Cards features, and are looking for a Hardcare Game only card. Then stick with ATI or Nvidia. -

[mallowman]

Originally posted by Sinistral


- And everyone who is slamming the card fails to realize what the card has to offer other then Super High Frame Rates in First Person Shooters and gaming all together. -

Those other things the Parhelias has to offer is precisely the reason I was originially excited about it and even more why I still am.

One thing that I am especially interested in is the depth adaptive tesselation, I think this may well be the biggest, if not then certainly the neatest, innovation in graphics tech I've seen recently. I think that once lots of applications start using higher order surfaces to define almost all of their objects that the parhelia can be a real winner and quite possibly see better performance than we are seeing now. (So I guess my previous statement about performance not going up is not 100% accurate to what I think) Being able to apply LOD techniques to geometry without having to check in the renderer how far away things are, so it can be used with more than just terrain, is I think extremely cool.

Another thing I am curious is is that if the Parhelia was quick enough to do UT2003 in surround gamming at E3, then why are the tests the reviewers are doing coming out with numbers that indicate (extrapolating from single head performace, which I know is not 100% accurate) that surround gaming with UT2003 and even some current games wouldn't be playable.

I really feel that we are not finding out the whole picture here and I want to know what it is we aren't finding out and why. My questions above are part of trying to find that out.

[mikeul]

Quote :
- And everyone who is slamming the card fails to realize what the card has to offer other then Super High Frame Rates in First Person Shooters and gaming all together.

Get yourself a Voodoo5 then.
Its FSAA is superior in image quality.

[Sinistral]

- Yeah but does the Voodoo5 offer great 2D quality? Triple head and True Dual Head? 10-Bit Gigacolour? Etc... -