Slashdot Mirror
Matrox's New Three-Head Video Card
This Anonymous Coward was one of many readers to point to sites with information on Matrox's upcoming Parhelia-512 graphics card: "It appears that some foreign hardware sites have violated NDA and posted some very juicy details on Matrox's next generation hardware. iXBT's review can be found here(1), and a MURC posting with some other pics from China can be found here (2).
It looks like the real deal. Will Matrox wake up from their long slumber in the 3D gaming market, or will this card be another stopgap like the G550 was?" Update: 05/12 14:07 GMT by T : Alexander Medvedev of ixbt.com points to the English version now online as well, and notes : "Please note, we can't violate NDA becouse we _do _not _sign _anything
with Matrox Graphics. And never receive any info from Matrox."
Surround video Pr0n...
No, it locked a Geforce 4 Ti4600's in a box and fed them only the bare minimum to keep them alive.
for sport.
whilst rendering Jedi Knight II at 3840x1024 / 32bit at 100FPS.
It seems this thing will kick major (major) major ass...
The thing works with 10 bit resolution
has 5 outputs, and 2 display controllers (dunno how they will have 3 monitors attached)
There is a glyph antialiasing unit (ahem...)
DVD/HDTV decoder (10 bits) and also a 10 bit video digital interface.
more info at
http://ixbt.com/video2/parhelia512/chip_diagr.jpg
The only things that I can make out are :-
:)
1) It has some VERY fancy graphics tricks up its sleave.
2) 16xFSAA
3) 3 Head support for "surround gaming" which appears to be supported by the driver rather than requiring the developers to support it.
4) Support for 256mb of DDR ram.
The rumours I've head about this card are that its also FAST enough to give nvida some competition. Hopefully tom will have some benchmarks soon
Look closely at the lower right corner of this screenshot. This is the one without the 16xFAA too. Pretty disappointing really.
Here is the screenshot from 3DMark 2001.
For those too lazy to look it shows a paltry 3 FPS.
I can't envision hooking up 3 CRTs
I can, it's about to be winter down here.
Dave
I write a blog now, you should be afraid.
The register has more details (and isn't dead):l
Go http://www.theregister.co.uk/content/54/25238.htm
DWR is Ajax for Java
80 million transistors
Technological standard 0.15 gm
Graphic nucleus/kernel and memory work with the clock frequency up to 350 MHz
Valuable is 256 bits (!) OF DDR the busbar/tire of the memory
The capacity of local memory on the order of 20 GB/sek
Capacity of local storage 64/128/256 MB.
AGP of 2kh/yakh/8kh including regimes/conditions SBA and FastShrites
4 piksel'nykh conveyors
4 textural blocks on each conveyor (!)
To fillrate:up to 1.4 gigas-peaktorrent and up to 5.6 gigas-flowtorrent
Apical sheydery of version 2.0 (Vertekh Syuader 2.0), four parallel fulfilling blocks
Piksel'nye sheydery of the version of 1.3 (Pikhel Of syuader 1.3), 4 textural + 5 combination stages on each piksel'nom conveyor, with the possibility of the association/unification of conveyors in pairs (we obtain 2 conveyors on 10 combination stages)
YEMBM and DOTE the imposition of the relief
Fixed/recorded T & L DKH8 (including the extended possibilities of matrix blendinga and skininga).Is actually special apical sheyder
Construction, storage in the local memory and conclusion/derivation to the monitor of image with the accuracy of 10 bits to the component of color (!). the technology of the 10- bits Of gigaCholor
Two built in the chip, 400 MHz, 10 bits to the channel RAMDACH, which use technology Of ultraSyuarp
Valuable of 10 bits.> 10 bits are tables for the arbitrary Gamma-korrekqii the concluded image
DVD and YUDTV of video decoder with the accuracy (at the output/yield) of 10 bits
Is supported the conclusion of image in the permissions/resolutions up to 20ya8khyshche'khe2bpp8shch Hz
Built-in the chip interface of TV -Out with 10 bit accuracy signal shaping
Two digital TDMS of interface for the digital outputs/yields or external RAMDACH.Is supported permission/resolution up to y920khy200khe2bpp
Two*** TRANSLATION ENDS HERE ***
There was also a mention of glyph antialiasing... And 64 / 128 bit per pixel colour...
L33t haxx0r: Notice the on-topic first post above.
Matrox Driver Problems: We are experiencing major driver difficulties with Matrox products under Windows XP. All of these are with the most recent Intel motherboards and Matrox G400, G450, and G550 adapters. We are using the latest Matrox drivers from the Matrox website. We have also tried the Microsoft certified drivers, which are much worse. We have tested with clean installations of Windows XP, as well as upgrades from Windows SE.
Our Win XP clean install test machine takes 18 seconds to display 97 items when doing a DIR directory listing. This appears to be caused by bad interactions between the Matrox drivers (with a new Matrox G550 adapter) and Windows XP. We are testing with a 2 GHz Pentium 4 and a new Intel motherboard.
We often see artifacts in DOS windows. Little colored vertical bars are left on the screen after some operations.
When we reported these things by telephone, the technical support representative, Bob Alionis, was very reluctant to deal with any matter that could not be solved quickly. He told us to try a video adapter from another manufacturer. This was difficult for us, since we have been building computers only with Matrox cards. Also, if an adapter from another manufacturer worked well, why would we go back to Matrox?
We tried an ATI Radeon card, and it worked better. We would be reluctant to switch to selling ATI cards because of our perception that ATI often has driver problems.
We haven't tried nVidia yet. Do nVidia chipset cards display business applications crisply? None of our customers run games, so sharpness at 1600 x 1200 resolution on 19" monitors is the most important criteria.
There is apparently no e-mail address for Matrox technical support. Matrox did not respond to e-mail sent to sales. Matrox did not respond to e-mail about technical problems sent to the RMA department.
Things have changed at Matrox. They are apparently trying to keep the number of tech support calls down by making it complicated to report a problem. The paragraph below is an exact quote from a message sent by a Matrox RMA department representative. The phone number mentioned is in Canada. Apparently Matrox does not have a U.S. number.
Jump through hoops RMA procedure:
"You can obtain an RMA for your board through Tech Support. Just make sure that you have registered your Matrox board on our web site http://www.matrox.com/mga/registration/home.cfm and have selected the option 'Obtain your tech support client id number...' at the Registration Menu. Once you obtain the client id number, just call 514-685-0270, then select option 1, followed by option 4, and then finally option 1 to reach the Tech Support queue to speak to a technician. For additional information on the RMA procedure, e-mail rma@matrox.com or call 514-822-6000 and ask for the RMA Department."
We wonder if Matrox is unable to fix its driver problems, and they are trying to avoid taking calls about them.
That's the way Matrox's tech support procedures have been since the days of the g200. After you go through their faq, you'll find a tech support e-mail address, where you can ask for support. After it's been determined that the problem is not a result of user error, the techie(who responds from his/her own address within Matrox) can then give you a RMA number.
They're not trying to make anybody jump through hoops, they're ensuring that the user has done everything they can to get the card working before tying up their resources by processing an unnecessary RMA.
Karma: Dyn-o-mite!(mostly affected by Jimmy Walker reading your comments)
Check out Maximum PC's May issue: this freak has 4 LCD screens (1 AGP & 3 PCI cards) running from one PC -- looks like somethng out of The Matrix..... He had to do some custom fabrication on the screens, but pretty cool..
...we are from the government - we are here to help...
Neomagic, however, eventually learned the folly of having an anti-Linux policy, and were forced to leave the Laptop chipset market altogether; I am sure that the various laptop makers did not appreciate all of the returns from people who wanted to use Linux. In fact, NeoMagic's support web page srill prominently discusses Linux drivers.
- Sam
The secret to enjoying Slashdot is to realize that it should not be taken too seriously.
I hope they're getting ready to make an explosion at Siggraph this year. :)
But the big question is - where the hell did all this come from? Did they hire a few people away from nvidia? Did they run across some brand new wunderkind? Or is this what happens when you shelve new product development for a few years and focus on delivering a new product three years, not six months from now?
Is the architecture modular and well-designed enough that Matrox can continue to compete when the other guys catch up?
And is Matrox (I hope) back?
They have a few screenshots of different games which they've tricked into supporting it at the first link above. And I have to admit - it makes me a little drooly. :)
It's also a brilliant move for Matrox: If they keep throwing out 3 head cards at a premium price - after buying one Matrox 3-headed card, who's not going to keep purchasing Matrox cards? If you got this set-up, would you disable one or two of your gaming monitors just to get the new nvidia whizbang that might be 10-20% faster?
I mean, if you look at the performance charts at end, you'll notice that they are cheap Excel graphs. Personally, if I were as big a company as Matrox I'd use something better than Excel to make graphs. Or at least, they wouldn't be that crappy.
Spoken like a person who has no idea how computers are used in the real world. Yes, there are programs that use 3D, including modelling and games. They continue to form a very small part of the market.
In evidence I present the Matrox G200 MMS; a four-head video card based on the marginally 3D-capable G200 chip. Matrox sells these by the bucketload into businesses like finance, who give some value to a card that can present four screens of 2D information. I also present the 10 Top Selling Games of 2001. There's exactly zero games in there that can't be played on a G400, and one that would like a more powerful card (Black and White).
Yes, Matrox realizes that they cannot compete in the high-end 3D gaming market with the G550. What you need to realize is that most of the computer users in the world don't need any3D, let alone more than what a G400 can deliver.
All of my video card money goes to Matrox. End of story. I won't waste my time with these half-assed, hostile companies like ATI and Nvidia. (Yeah, so I'm a little behind the curve on this whole "3D game" thing...;) I'm extremely happy that Matrox will produce a good 3D card! I have owned the Millenium II, Marvel G200, and G450 dualhead and been extremely happy with them all. Now I'm just waiting for the Marvel G1000!
--Bob
1^2=1; (-1)^2=1; 1^2=(-1)^2; 1=-1; 1=0.
I've been using matrox dualhead videocards for quite some time now on my studio machines. It's really handy for extending the cubase desktop on various screens, because there's too much information for one screen.
It's very important that the computer runs rocksolid for high end audio-applications, even the chip-set on the motherboard can be the cause of a lot of problems, that won't show up during non-audio applications. Luckely the matrox drivers a really solid, compared to other brands.
I was just ready to order a non-agp dualhead card in order to attach more screens to this machine, but now this card shows up and I will deffinately buy it as soon as it comes out.
I'm not laughing. I own a Vax. There were some volcanologist graduate students, that were poking around the neighborhood last summer, they thought that they had discovered some unknown thermal vent on an obscure satellite heat map. They were disappointed when I explained it was just the trusty MicroVax. I felt kinda bad, to tell the truth.
First, that massive 20 GB/s of bandwidth is going to be needed, every bit of it. There is no bandwidth-saving logic on the chip at all, unlike ATI & nVidia's latest. Since occlusion detection can make a significant difference, and Z compression & fast Z clear also help a great deal (ATI claims their 8.8 GB/s performs like a 12+ GB/s system, a 36% boost), the Parhelia could be considered to have only 55% more bandwidth than a GF4 Ti4600 instead of 110%. If the next-gen offerings from ATI & nVidia have similar memory specs, the Parhelia could be at a significant disadvantage almost as soon as it comes out.
Second, the Digit-Life article mentions that early scores (from very raw drivers) show a mere 20-30% increase in scores over a Ti4600. Now admittedly this should increase, but Matrox are not known for their 3D driver optimisations, and nVidia are. A unified driver architecture will give you a head start right out of the gate, as you can take some advantage of previous optimisations immediately, whereas Matrox will have more work in front of them to get their drivers performing near the potential of the hardware. Look at ATI; it took them 6 months of focussed effort (and the odd quality hack along the way) to get their drivers up to scratch. Matrox have not traditionally given their 3D side or their software side as much attention, in my experience.
To me, while the triple-head feature could be useful to some (though I dislike external DACs - it's difficult to sync them closely to internal DACs, causing monitor beats), the 10 bit colour is to be applauded, and the vertex handling sounds very nice, anyone looking for performance would be better advised to wait for R300 and NV30.
On a slightly different note, was anyone else disappointed by the quality of the 16x AA screenshots? I expected more. The edge-only AA feature sounds like a very good idea (though it will not help alpha textures, just like multisampled implementations), but I'm a bit jaded after the miracles promised by ATI's SmoothVision didn't exactly set the world on fire. Guess we'll have to wait for performance figures.
Also, I wonder what their yields will be like. 80 million transistors on a 0.15 micron process sounds like something that's difficult to do cheaply.
Why would anyone engrave "Elbereth"?