Next Version Of Parhelia

[3dcgi]

Originally posted by Nuno
Crippled 64 Mb P: Big Mistake

Issues to correct on a new Parhelia rev:

- Redesign memory controler to 4 64 bit independent buses with bandwidth saving features.

I remember reading they actually have 4 independent memory controllers. I'm thinking I might have read it on ExtremeTech, but I'm not sure.

- Get some 2.0 pixel shaders with all the transistor space that´s left after removing all those useless texture units

I doubt this is an even trade. Floating point support will add a lot more hardware.

[Nuno]

If you want 9700 performance you have to have 8 pipes, 4 ain't gonna cut it.

R9700 has 8 pipelines with a single texture unit each. It´ll only double the speed of a 4 pipe design when single-texturing. When multitexturing its speed it´s roughly the same as a 4 pipe, double texturing unit.

The big advantage of R9700 isn´t about pipelines, but on the agressive clock settings for such a big chip and the 256 bit memory bus. On theorethical pixel pumping numbers it equals a overclocked Ti4600.

I remember reading they actually have 4 independent memory controllers.

hmmm. Maybe it´s so. But is that a "cross-bar" design like Nvidia´s?

I doubt this is an even trade. Floating point support will add a lot more hardware.

Of course, I´m just speculating I don´t pretend to be a chip design expert or sugest that redesign the Parhelia core is a 5-minute task. But I get the impression that much of Parhelia silicon isn´t being that much usefull right now. And why not .13u?

Anyway ATI managed to do a .15u chip with equal or superior complexity as Parhelia and clock it 100Mhz higher and with some confortable overclocking headroom. Ok, just forget that external power conector It´s a pretty smart option, though.

[DGhost]

keep in mind Matrox was doing """crossbar""" memory controllers in the G200 and G400 series iirc... also something that iirc, NVidia didn't start using until much later... the term """Crossbar""" was just some NVidia marketing fluff...

the 4 texture units is a good idea, as the next generation of games will often require more than 4 textures to be applied for each pixel. the problem that it faces is the fact that it doesn't allow more than 4 textures to be applied through a loopback interface. if they did that a lot of developers would be happy....

keep in mind that its multitextured performance is better than that of a 9700...

several people did a study on the memory bandwidth saving features (z-buffer compression)... it resulted in a minimal performance inprovements over when they were disabled... the only thing that really made a difference is fast Z Clears... which the Parhelia does support...

plus, the time it would make that much of a difference is in memory bandwidth limitations... ie, a GF4MX would benefit a lot more than GF4Ti... but with the memory bandwidth that the Parhelia has makes it almost pointless... and all they would have to do is crank the memory speed up...

plus i seem to recall Matrox talking about having fairly efficent texture buffers and caches...

the trilinear + ansioscopic filtering stuff could use improvements...

2.0 pixel shaders would be nice, but thats not a cost saving measure as was already said, the whole pixel pipeline would take a massive rehaul to do it...

the only other thing that could really use improvements without doing a complete core redesign pretty much would be improving the clock speed... if it was another 100mhz faster it could own the 9700 fairly quickly... especially with FAA enabled... its an impressive architecture and all the problems that it has right now are due to its low clock speed. with a higher clock speed it could really cook.

that being said, i do not know what the parhelia was supposed to be... if it was supposed to have 8 pixel pipelines instead of 4... or what... but... the rest of the core could be competetive with the GF4 and the 9700....

[3dcgi]

Originally posted by Nuno

hmmm. Maybe it´s so. But is that a "cross-bar" design like Nvidia´s?

I don't believe it is a "cross-bar" like Nvidia's, but I don't remember the article giving any details.

[dmantione]

Matrox future

The whole point is that designing competetive 3d cards costs an awfull lot of money and Matrox has driven itself in a small corner of the market that doesn't provide enough revenues to maintain competetive even in that market.

With the Parhelia Matrox did try to get out of the circle, but it looks like they cannot. Instead of awaiting bankruptcy I hope they take action against it, because they can do quite a lot.

For example:
- Use third party chips. Matrox did this in the past with PowerVR chips and they can do it again. Matrox of course equips their boards with their high quality output electronics and uses it to differentiate it's products from the made in Taiwan cards. With the revenues they can fund their chip designers to develop the next generation chips.
- Alliance. Both Nvidia and ATi have more chip design teams than Matrox does. While Matrox has to choose between improving an existing design or working on the new one, ATi and Nvidia do both at the same time. By making an alliance with another company that has the same problem (3dlabs, trident, sis?) they would be able to do the same.
- Takeover. If there is really no other way out, Matrox can try to be taken over by a larger, richer manufacturer. Unlike 3dfx, Matrox could continue to sell cards under their own brand, using usual Matrox features for marketing. Their chip design teams would be integrated in their new owner and be able to develop better chips.

Now, let's hope Matrox is smart enough to not await bankrupcy.

Daniel Mantione

[The PIT]

Yeh get Matrox and ATI together. Matrox too make sure they get the display quality, dual head right and drivers right.

[DGhost]

ATI to mismanage it, make their tech support crappy, etc...

[dmantione]

Well, I must say ATi's driver brewers do much better drivers than Matrox currently. I'm very dissapointed by the lack of Win9x drivers for Parhelia. The 2000 drivers? Fun with .net.

[bsdgeek]

The two companies' new logo would sport a purple color and the joint effort would be called ATrox

[DGhost]

heh, ATI's drivers have more compliancy issues than Matrox's do... along with more stability problems...

Matrox knows how to make drivers... the Parhelias are still better than ATI's... a lot less problems than any ATI driver set i have played with...

[103er-Fan]

Originally posted by DGhost
heh, ATI's drivers have more compliancy issues than Matrox's do... along with more stability problems...

Matrox knows how to make drivers... the Parhelias are still better than ATI's... a lot less problems than any ATI driver set i have played with...

Agreed! Matrox really knows how to make good drivers, the last time I used such stable drivers was back to late decembre 2000, with the good old Voodoo5 5500

[Chrono_Wanderer]

Originally posted by dmantione
Well, I must say ATi's driver brewers do much better drivers than Matrox currently. I'm very dissapointed by the lack of Win9x drivers for Parhelia. The 2000 drivers? Fun with .net.

heh first time to hear that ATI makes good drivers

Trust me. My R8500 would be in my faster P4 machine than my slower Athlon if their drivers are small & stable (compared to G-series drivers). Also, i hate having to DL the control panel seperatly... IMO ATI should make two options for end-users:

1) Put the drivers together so that boardband users like most of us do not have to dl twice.

2) Seperate the drivers like what they are doing now, for the 56K users.

P.S. Plez u big M marketing ppl we do not need a $250 Parhelia that has triple-head support while other more important features are cut off. After all, anyone who can buy 3 displays should be able to afford the current P.

[K6-III]

The main thing about Matrox drivers is that problems actually get fixed...rather than sometimes/rarely fixing something, but usually breaking something else in the process...

[ElDonAntonio]

It would be cool if Intel or IBM financed Matrox Graphics. Intel could expand its graphics dominance, it already has a large portion of the integrated graphics, and with some money shot in Matrox's arm, I'm sure they would output some GREAT cards really fast. Besides, it would allow them to compete more with the nvidia/AMD alliance.
As for IBM, well...that's the big mamoth in the computer world with endless pockets

But I don't know how that would work because Matrox is a private company.

[Wombat]

Money isn't Matrox's problem. They've had all the money they needed to do things right. Handing them more capital would just let the same idiots mismanage mroe funds.