Ich Kann nicht Deutsche verstehen

[Matt Burris]

Parhelia has exceeded my expectations anyway, probably because I didn't let the hype get to me. I am really impressed by 16x FAA, and the scores it generates. That reef demo is really sweet, I'd like to have it. Overall, I think the FAA/Aniso performance with this card is very very nice, and I personally I like that, because I use both heavily in all my games. I just wish the pricetag wasn't so high, otherwise I would buy this card without hesitation.

Guess I'll have to resort to whoring myself to Matrox and see if I can get a review unit.

 

[cellarboy]

Not to get off topic or anything, but are Kyle Bennet's stream of bitter little comments regarding the Parhelia getting as old and annoying for anyone else??

 

[Nappe1]

Re: faa

Folks, when's the last time you saw a videocard release that actually exceeded your expectations?

SiS Xabres / 33x series really exceeded (more than that actually, because I didn't expect anything...) my expectations.

 

[PiNkY]

I really think Matrox will have a ton of headway in their drivers...I mean, a TON. It just seems impossible to fathom a product featuring FOUR times the number of Vertex Shaders over an 8500 to lose out that badly...And even with the relatively slow clock, it should still perform much better without all the IQ features enabled.

I thought r200 had 2 vertex shader units and because of that stomped gf3 in Vertex Shader Performance

 

[Randell]

The TRUFORM engine is an additional vertex unit I think, and maybe its used in that test. But it obviously isn't utilised in normal games when Truform isnt used else the 8500 would stomp a Gf3 in games. Of course I may be wrong

 

[Jazz]

Anandtechs review is up as wel http://www.anandtech.com/video/showdoc.html?i=1645

 

[Tahir2]

Having seen the technology analysis at The Tech Report and the review at Hot Hardware I have changed my mind.

There is a l;ot more to Parhelia than first meets the eye in these reviews. I have to admit I am quite impressed and would love to see FAA in front of my very own eyes.

There is more to gaming than FPS

 

[Sharkfood]

I dont understand the hoopla over the price. It's the same price bracket as the other contenders. Remember, a GeForce4 Ti4600 is retail $399, as is the 8500AIW 128MB $399 retail. This puts it in the same category.
GF4 Ti4600:
http://www.compusa.com/products/product_info.asp?product_code=290231
8500AIW
http://www.compusa.com/products/product_info.asp?product_code=293118

It seems both the NVIDIA and ATI equivalents have VIVO/Capture whereas the Parhelia does not. Perhaps "surround gaming" is their attempt to capture a feature for feature against VIVO/Capture?

As far as performance, I can't make heads nor tails of the various websites to get any indication on how this sucker runs or performs. There are some sites showing max anisotropic shots and others saying it's only limited to 2xAF. Some show promising Multitexturing performance with both AA/AF maxed and some show poor benchmarks. There is massive variance and few conclusive screenshots to balance performance. There is also no notes concerning z-buffer errors or overall image quality, driver stability (a couple arbitrary notes on UT, but nothing tangible or useful) and the like.

It's pretty obvious just how useless cookie cutter website reviews have become since absolutely none of the data leads me to a conclusion concerning the overall product quality, performance or stability. I also find it kind of humorous how everyone wants to bust out the Aquanox's and Commanche's of the world in order to paint pretty benchmark graphs. I saw throw in DroneZ, Giants and VulpineGL to finish the "package" in the same vein, lol.

Lastly, does anyone know what the hell Anand is talking about for "16-tap" in his reference to AF? Does this mean 4x on GF4 and 8x on the ATI? I really wish sites would be more specific about AF, especially given the 8500's bilinear usage. It leaves the reader with little clue as towards what is actually being tested. The "AF+AA" tests are the same way and should be more specific as towards what exact settings are being compared.

Oh well, looks like actually getting a Parhelia for hands-on testing will still be the only way to figure out how these cards handle. Doesn't seem much different from the past few months.

 

[Randell]

Or you could wait for Johns review here Shark like I and many others will

 

[BRiT]

Re: faa

Folks, when's the last time you saw a videocard release that actually exceeded your expectations?

The last timie was with 3dfx Voodoo 2 solution with support for SLI. The time before that was 3dfx's Voodoo 1.

The V1 was so far beyond my wildest expectations. The V2 had exceeded my expectation by doubling my previous speed with higher resolutions too and by staying fast enough for an entire year. At that point, I merely added another one with SLI, which kept me blazingly fast for another year.

Any and all other cards have fallen far below my expectations or have merely met them. The list of the fallen expectations includes the dreaded THG-proclaimed "3dfx killer" from PowerVR that never materialized [well years later and never killed anyone], the 3dfx V5 (great card, but I wanted it 7 months earlier) or the V5-6000 (limited qtys), and the NVIDIA GF3/GF4 and ATI-8500 (why couldn't they get FSAA done right?).

I hope this time around NVIDIA/ATI will have nailed the FSAA/FAA aspect, and perhaps throw in something that will go beyond my expectations.

If the BitBoys ever shed their bad luck and get to release a card, it will exceed my expectations. I hope they can do so.

Now, if anyone were to release an upgradable card, say ala-SLI or expansion modules, then that too would far exceed my expectations.

--|BRiT|

 

[hughJ]

will be very interesting to see how this card overclocks..

how much bandwidth overhead do you think there is?

 

[Freon]

Re: faa

I think some of you aren't really _looking_ at those AA scores. note that even at 1024x768x32x16FAA, that card is _still_ pulling 101 frames per second in Quake3. 101 frames per second is a lot. Add the gorgeous edge AA that Matrox has, and you're looking at stellar image qualities. Even the 4600 is only pulling 10 fps more at only 4x AA at this resolution. And, at that AA setting, the Parhelia remains fully playable (above 30fps) all the way up to 16x12x32!

And Quake 3 is an old and simple game by todays standards, even though it was ground breaking at the time. Even the more modern Q3 engine games like SOF2, MOHAA, and RTCW are *much* more demanding. I wouldn't be surprised if the Parhelia couldn't break 60fps at *any* setting in SOF2 without pretty much turning everything in the game settings down. Luckily SOF2 has lots of very good scalibity built in (I think it has like 6 different main texture quality settings, plus 4 increments for each of models, effects, sky, etc).

And I don't consider 30fps good. I consider 60fps average in a benchmark for a $400 the bare minimum, regardless of quality. I doubt I'm alone here. 30fps (average) is a slideshow. I haven't gotten 30fps since Quake days in software mode. Maybe 30fps in a brand new game is acceptable for a value card, but NOT for a $400 premium card.

I dont understand the hoopla over the price. It's the same price bracket as the other contenders.

Remember, a GeForce4 Ti4600 is retail $399, as is the 8500AIW 128MB $399 retail.

By this logic, the Geforce 3 (not Ti) is still $399, too. Fact is, the 8500LE (250/250)can be had for as low as $99 and matches or beats the Parhelia in most benchmarks. It only seems to lose big at 1024x768x32 with AA in a 3+ year old game.

Sorry to sound so bitter, but I guess I'm just severely disappointed in the Parhelia after seeing today's reviews. Call me when the drivers give it a 30% boost across the board. Hopefully this happens before the NV30 and/or RV300 hit the streets...

 

[Dave Glue]

I dont understand the hoopla over the price. It's the same price bracket as the other contenders. Remember, a GeForce4 Ti4600 is retail $399, as is the 8500AIW 128MB $399 retail. This puts it in the same category.

The problem is, you can get a 4200 for half (less than in some cases) the price of the 4600, and still easily beat the Parhelia's current scores. I don't think anyone would be praising the GF4, even despite its performance, if the 4600 was the only model Nvidia offered. And Nvidia certainly would not survive if it sat on the GF2's laurels for 2+ years then only offered a $400 solution as an upgrade path.

And why the comparison with the 8500 AIW? It's a do-everything board which hardly compares to Parhelia's 3-monitor-support, which is at best an extremely niche item that doesn't begin to have the more broad appeal of the AIW's features. You're paying a large premium for the TV functionality, with the Parhelia you're paying a large premium for 3 monitor support, and good AA. At this point IMO, that doesn't being to come close to justifying the $400 price.

 

[nggalai]

I wasn't disappointed about the Parhelia--after all, we didn't learn anything new Matrox didn't say before regarding performance--until I read this little bit on Tech-Report:

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

(my emphasis)

I'm a sucker for texture quality and simply hate any form of texture aliasing. If Matrox really opted to disable anisotropic filtering > 2x due to performance reasons . . . nope. This card is effectively dead in my book.

ta,
.rb

 

[CarstenS]

nggalai,

You've asked about Matrox being known for fine-tuning their drivers for better performance in future releases.

Well, having owned a G400-card i can tell you my impression. With that card, they already had learned half of their driver lessons with the G200 before, as it featured a quite similar Engine and so they only had to refine their drivers with respect to compatibility.
So my answer to your question is basically no!

But, and this is a big but, this is a completely different story, as the whole setup of Parhelia is almost not a bit alike that of their former products. So, my guess would be as follows. They first have to get through with basic functionality and their recognition algorithm for object-edges to be FAA'ed before they actually take their time and make Parhelia perform as it should according to its specs.....so while you won't see a huge performance jump, as games get more and more complex, you won't see any major drop in relative perfomance either.....

Regarding AF, I found the following quite interesting...taken from the Parhelia-Preview on 3D-Center...

Eine TMU, die bilinear filtern kann, liefert exakt 4 Texture-Samples. Um auf 64 Samples pro Pixel zu kommen, müssen alle 16 TMUs denselben Pixel bedienen. Soll pro Takt eine doppelte Texturierung stattfinden, bleiben zum Filtern natürlich nur noch je 32 Samples. Das heisst konkret: Hochwertige anisotrope Filterung kostet kräftig Leistung. Die Framerate wird mit anisotropen Filter voraussichtlich deutlich einbrechen. Allerdings wahrscheinlich von einem sehr hohem Niveau aus, so daß auch anisotrope Filterung mit dem Parhelia gut nutzbar sein wird.

Summary: Each TMU of Parhelia's Pipelines can only do bilinear Filtering, thus delivering 4 Texture-Samples. If you use 8tap-AF and therefore need 64 TexSamples, all 16 TMUs have to cooperate on one pixel (mathematically). In Short: With AF turned on, Framerate will suffer greatly[/i]

 

[LeStoffer]

As far as performance, I can't make heads nor tails of the various websites to get any indication on how this sucker runs or performs. There are some sites showing max anisotropic shots and others saying it's only limited to 2xAF. Some show promising Multitexturing performance with both AA/AF maxed and some show poor benchmarks. There is massive variance and few conclusive screenshots to balance performance. There is also no notes concerning z-buffer errors or overall image quality, driver stability (a couple arbitrary notes on UT, but nothing tangible or useful) and the like.

I couldn't agree more. I did expect Parhelia to be quite a bit faster than GF4/ATI 8500 with FAA and AF turned on, but so far it is a bit of a lowdown.
But I guess drivers need to mature, and I would say that quite a few of the reviews should have had time to mature as well.

 

[bystander]

Firing Squad has a review up of the Matrox Parhelia which isn't so negative.

They evaluate the 2D image quality of the card, Gigacolor and the performance of Neverwinter Nights using Surround Gaming.

 

[Jerry Cornelius]

Hree's another
http://tech-report.com/reviews/2002q2/parhelia/index.x?pg=1

"...and I expect to see some scuffles in the aisles of Best Buy this week as rabid graphics artists fight over that last card. Goatees will be yanked, hemp clothing torn. "

 

[Pete]

TR has the best reviews--humor and graphs, what more can you ask for in a hardware site?

And those are average frame rates. You should be looking at the minimum fps count. 30 fps average in Q3 is definitely not playable. Minimum of 30 fps is a totally different story.

You bring up a good point, an aspect that I was looking for specifically in reviews, but won't get until we see one where the reviewers have the time and the intelligence to consider: minimum fps. Why are we still not seeing framerate over time graphs in reviews, particularly those involving new architectures? I've yet to read Anand's, TR's, or FS's reviews, but I don't expect them to produce the goods. I suppose I'm waiting on B3D--don't let me down, gentlemen.


Edit: ... and the french review (the freview, if you will ) has such a graph. Incredible how similar the peaks and troughs are for every single card.

Edit Edit: And TR has them, too! I'll be quiet now.

Last Edit:

Our Parhelia came stocked with 128MB of DDR 325MHz memory for an effective clock rate of 650MHz – on par with the competition.

WTF? First InqWell botches it, now FS bungles it even more. Is it sleep deprivation? A lack of commitment to their work? Sunspots? You'd think you'd be required to know WTF you're reviewing to get a review board. I'm pretty sure Kyle would have at least gotten his facts straight....

 

[nggalai]

Hi Quasar,

thanks for your information regarding past Matrox drivers. About anisotropic filtering on Parhelia:

Summary: Each TMU of Parhelia's Pipelines can only do bilinear Filtering, thus delivering 4 Texture-Samples. If you use 8tap-AF and therefore need 64 TexSamples, all 16 TMUs have to cooperate on one pixel (mathematically). In Short: With AF turned on, Framerate will suffer greatly

I am aware of that. But I find it interesting how Tech-Report got a completely different answer regarding ommission of AF>2x on the Parhelia than Leo (3DC) did. Tech-Report's Matrox contact made it sound as if it was disabled for performance reasons and that they might enable higher degrees of AF in future drivers, while Leo's Matrox contact made it clear it's a driver bug that will be fixed.

*shrug*

I still would love to see some 4°AF benches on Parhelia, though.

ta,
-Sascha.rb