What exactly is Matrox up to .... get the <facts>

[SmellYfeeT]

Very INTERESTING discussion going on, seems like the thread is becoming a hardware discussion forum
SuperFly brough up many good points on why going with 256bit is technically challenging , and that is not the case only for graphics card. There are many video processing chips as well that have the same issue, the BIG part of the problem is not having to many pins, but in fact lies under the packaging of the silicon, where you have all these thin gold wires connecting the die to the package. and when you cranck up the frequency on so many wires so close to each other you get severe Trasmission Line problems. One of the solution to this problem is the FlipChip technology that connects the die pins somehow directly to the pacakge and indirectly to the PCB. Which ofcourse is more expensive. So it is not Impossible to do this, but rather expensive. I don't know this as a fact, but the Parhelia design has a 256bit memory bus, however the way it works is different and MORE efficient than Nvidia's Cross bar technique.

[SmellYfeeT]

by the way, regarding the discussion on the embeded DRAM, I remeber 2-3 years ago back in University, I heard from a classmate of mine that either ATI or Matorx(can't remeber exactly) was working on a design with a memory manufacturer at that time, to include parts of the pixel operations on the RAM <this is the otherway around> I guess that idea got canned somehow, cause I don't know of any card that are built with that technology.
Another issue with embeded rams are also LOW YIELD rates of the silicon, which does not make it an attractive option to graphics chips. Also there are technical reports about this that shows large amount of Cache will not improve the performace of hardware components that perform mathematically intesive operation if they have low memroy speed. So my guess is expect a very high performance memory system with normal on board memory. should we start the guessing game on memory frequency, my guess > 350Mhz no really, it will be very high!

PS. I am very disappointed by now that Matrox has not announced anything! Therefore, we definitely won't see the card by March 2002! I will bying an 8500 all-in-wonder before Christmass now that it is out (go ahead Boooo me...)

[mute]

They're new card will be out right after DirectX 9 goes public..

and that's 100% SURE .. if they don't can it ..

[Unam]

Might be missing something obvious, but what would prevent Rambus from widing the data path and cranking up the clock speed of RDRAM? I've read a few online articles that state they are going to be doing just that for some of Intels future chipsets. Surely that would be cheaper than FCRAM...

[superfly]

That's exactly what RAMBUS will be doing within the next year,PC 1066(533 mhz) will be available when intel introduces p4's running with a 533 mhz fsb by mid year using a 850 chipset certified to run at that speed(133 mhz x4).


The main question is if pc 1066 will still use a 16 bit bus or if Rambus will transition right away to 32 bit and if they do,will intel still use 2 memory channels,which would total 8.4 gigs/sec of bandwith or stay with only one at 4.2 gigs/sec.


Just a question though,doesn't the gamecube also use rambus as well,because i seem to have read somewhere that it does.

[KngtRider]

the next nvidia mobile gpu has embedded ram, nvidia's fabs (TSMC+UMC) get/will get high yields out of those chips, and NEED to as many OEMS use that gpu, DELL, toshiba etc etc and shipping many units of those notebooks around the world

EDIT: gamecube doesnt use rambus, n64 does

www.icrontic.com < gamecube

http://www.icrontic.com/gc5.htm <processors+ram

the last time i seem MoSys Ram used was on the ET6000 and needed to be special parts to suit that gpu

note the CPU and GPU have ©2000 on them though the unit is a 2001 build, 1yr lag for registering of intellecutal properpty>shipping, so can use that approx on matrox since its a new and big new part

[CHHAS]

I just hope that we'll see a good 3D performer from Matrox soon, this Gefart 2 Pro is killing me, maybe I should put the G400 back in and play in lower res, might even look better too

[KeiFront]

Originally posted by Unam
Might be missing something obvious, but what would prevent Rambus from widing the data path and cranking up the clock speed of RDRAM? I've read a few online articles that state they are going to be doing just that for some of Intels future chipsets. Surely that would be cheaper than FCRAM...

FCRAM should cost around $10 more than DDR-SDRAM. That's still cheaper than RDRAM.

Does anybody know the ETA of DirectX 9.

[CHHAS]

Heard some rumors about March.

[tr4veler]

I though it was the PS2 that used RDRAM ?

[Ali]

I still have stuck in the back of my mind that Carmack interview about 18 months ago saying Matrox has solved the bandwidth issue. That means that they must have had a working, or near working simulation at least back then.

Also, Matrox has signed up with ATI on the shader specs for OpenGL, rather than use the nVidea designed specs. That means a programable T&L unit.

Amiga said Matroxs next product would be kick ass, and I assume they didnt meant he G550.

Haig has been dropping hints about the memory used being interesting.

There was an interview a while ago with a Matrox guy saying they needed to solve the Anti-Alising problem. Lots of bandwidth should do that. Huge supersampling would be best, as textures wouldnt get washed out like Geforce3.

Displacement Mapping is almost a given, and that needs huge amounts of bandwidth.

And lastly, Matrox always makes a balanced product. So the core will have enough pipes to feed all that bandwidth that we know they must have. Should be a VERY fast product.

Ali

[[GDI]Raptor]

FCRAM looks like the best solution for Matrox. I realy hope they have fixed all problems. High Quality Anti-Alising had been very nice, whith a bandwich of 19,2Gb/s, it looks realy good. Displacement Mapping is also a impresive feature, and I think Matrox will use it in their chip, because they have presented it at several trade shows.

DirectX 9 is due in march/april, that is what I can read on the MSDN homepages.

I hope Haig can give us a hint on when we can see an announcment from Matrox, Sometime in 2002 is not an good answer...

[superfly]

I think that ultimately,it isn't wise for any given chip maker to implement different versions of pixel shaders,vertex shaders or what have you so that that the company who developed it can boast about having a feature that other cards don't since all that does is create confusion with users and delays support from game developers,it's been proven time and time again.


So the real issue should be how well each feature is implemented in order to see who has the best image quality/speed compromise,not if the feature is actually there or not.

I know that won't go down well with each company's markting dept,but ultimately it's unavoidable if we want to see those features get used(and abused...),by game developers as soon as possible.


As far as antialiasing issue is concerned,from the admitedly few shots i've seen so far,the Radion 8500 does a pretty kickass job thank's to the variable algorithm they use since it keeps textures from looking washed out like the GF3.All it would need now is the anisotropic capabilites of the gf3(at least 64 tap preferably 128 tap) to make it look perfect.


Can't wait to see what ATI has up their sleeve with the R300 chip,which should be on the final stages of developement,but regardless it clearly shows that ATI is on the right path.

[KngtRider]

ATI has/will always have D3D and GL problems, since the r6c its thair nature inherinet with the company, would be suprising if they fix EVERYTHING inc trilinear

100% chance displacement mapping is the new chip, no doubt. They have engineers working on that as shown by meltdown

http://www.nvidia.com/docs/lo/983/SUPP/detonatorxp.pdf

this document released with 23.10 dets shows the feature of dets XP inc GL caps supported by their ICD

doc shows what matrox need to implement on the new part

i learned 2 new things from nvidia pdf today

1) Panel Link and TMDS are trademarks of Silicon Image, Inc.

2) no support whatssoever for referece driver, and the board makers wouldnt be able to help in code optimisations cause they just custom the drivers not rewrite core

now the 1000 dollar question is will I have a new matrox card in MY system in 2002

good think though is that no one in AUS hw review/oc community likes matrox so i got a headstart to keep track of the chip and try to get aus exclusive heh heh

now the remaining issue, case badges

stuff the memory

we want matrox case badges again now

mII had 1

BTW n64 did use rambus

[superfly]

Well let's give them the credit that the hardware behind the 8500 is pretty impressive,especially at the price these cards are going for and i'm pretty sure that ATi can't possibly be making a lot of money on them,especially when it's quite easy to find the cards going for well under 250$,something users haven't seen in a couple of years,at least when it comes to high end video cards.


And it can't be denied that they've already made quite a few improvements in the driver dept for a card that's only been available very recently,so i guessing that with a couple more driver releases,most of the issues that still remain will be addressed and with any luck we'll see the card operating at it's full potential.


And get this,according to ixbt,nvidia is supposed to anounce the NV25 by late january and ATI will apparently spoil it somewhat by launching even higher clocked versions of the radion 8500(300 mhz core,300 mhz memory).

But again,according to the site,they'll have two models available,the radion 8700 which is just the higher clocked version mentioned above and get this,the radion 8800 that also runs at the same clock as the 8700 but packs 128 megs of 300 mhz DDR sdram.

Damn,128 megs...until recently most users didn't have that much running their entire system,let alone just for the video card....