Parhelia 2 suggestions

[piaxVirus]

Since the crystal ball has died down. I thought I would start a thread about suggestions for possible improvements Matrox could make for the Parheila 2 chip. This is not about some mythical card that will be release in 3 months. This is about what you would like to see in the next version of the Parhelia core, that will not be out within the next year or year and a half.

.09 micron fab process
8 Textures per pass
Multiple loop backs (upto 24 possible textures per pass)
6 Rendering Piplines
2.0 PixelShader
16X Anisotropic Filtering
12-16bits per channel internal precision
Programmable floating point pixels and textures
Gigacolor with full 10bit stencil buffer
Independent Triplehead
Independent Dualhead plus TV out
DDR-II or QDR 256bit memmory interface
8x AGP 3.0
Improved TV/HDTV output quality
MPEG 1/2 decoder
iDCT on GPU
RSN v2 (I'm not joking. I want my RSN! )

Occlusion culling should not be required a year from now. So I did not include that.

[SpiralDragon]

well i was going to say "Independent Dualhead plus TV out " that would be grat for my line of work... sound and vid editing.... one monitor is not enugh and with two on i loose the tv out... (no more monitoring) how does that play in for sound.... lip sync..... while i am working on dule monitors i like to out put the vid to the TV espesialy when there is need for dubing.... 19' monitor is not always enugh to sopt the details..... i have a 46' TV for that.... and maybe some better power protection... for some strange reason both my G450 and G550 have a busted second head due to grounding problems beyond my prevention.... darnd electricians in this country.....arghhh..

[SpiralDragon]

P.S. "Independent Dualhead plus TV out " especialy since non of the matrox video products suport AVID express DV x ..... only that crapy toy software premier.... bloody gastly bit of work.... how ever did i manage to work on it ... ahhh now i remeber.... anything from AVID was tooooo expensive untill they came up with their student license fee....

[[GDI]Raptor]

I would sugest 8 Renderig Piplines, ATI R300 is going to have that and also 100% DirectX 9.1 compatible and improved 16xFAA

[BuddMan]

Hmm...12-16 bits per color channel. I dunno about that.

I think they should work on the alpha channel first, maybe try for a 10/10/10/10. However, I think it depends on what the next generation cards, the R300 and NV30, will do. If they can pull something higher than 10-bits per color channel, then I would hope that Matrox would definately try to follow.

[Hyp-X]

Originally posted by BuddMan
Hmm...12-16 bits per color channel. I dunno about that.

I think they should work on the alpha channel first, maybe try for a 10/10/10/10. However, I think it depends on what the next generation cards, the R300 and NV30, will do. If they can pull something higher than 10-bits per color channel, then I would hope that Matrox would definately try to follow.

Appearently R300 will support 32bit per color channel internal, and possibly 16bit per color channel external. (128bit / 64bit).

And that shouldn't be a surprise, at least the first one is a PS2.0 requirement.

As for the ramdac 10bit/channel is sufficient.

[Drizzt]

It's already 10/10/10/10.

[BuddMan]

My understanding was that it was 10/10/10/2 Drizzt.

[piaxVirus]

I am not sure the R300 is going to have 8 pipelines. I think it's 4 pipelines with 8 textures per pass. One of the ATI reps was quoted as saying most of the specs and pics floating around the web are wrong. I am starting to think the R300 won't live up to the hype.

The Parhelia has 10/10/10/10, but only with custom applications. I am not sure of the specifics. There is a limitation of 32bits, either in directX or in windows. So in order to get 10bit per color. They had to sacrifice 8bits of the alpha channel. That is what Gigacolor is. It's not full 40bit color, but 10bit per color in a 32bit limitation.

Both the R300 and the NV30 should support at least 40bit color. Since it's part of DX9. Will they find a way around this 32bit barrier is what I want to know.

I think they may support 12-16bit per channel internal processing. But 32bit (128bit total) per channel sounds too complex to fit on a .15 micron die. I expect the NV30 to have 16bit (64bit) per color.

[thop]

RSNaD - Render Subject Nude and Drunk

[Buback]

I would like to see the bugs worked out with FAA. maybe it's just a driver issue, but they seem pretty adamant that it won't work with stencil buffers.

occlusion culling wouldn't be a bad idea, as it would only extend the lifecycle of the card.

is the current card OpenGL2.0 compliant? i would like to see that.

bub

[piaxVirus]

Nude chicks drunk = Nude chicks puking. It does not look good.

It's like that Seinfeld episode. Where he saw his gf sneeze while nude and was grossed out. Except this would be with full color saturated texture.

[BuddMan]

Well, 64-bit or even 48-bit color (external that is) would be exiciting, but do we really need it right now? I don't think the R300 nor the NV30 would go above what DX9 calls for, which is a 10/10/10/10 output (right?). Internally, it could indeed be 128-bit color, and I could see why they would do that. But wouldn't 128-bit internally be overkill? Bah...all this speculation is making my head hurt.

[Ssin]

piax : never listen to whatever ATI has to say, they're only a bunch of liers and I've never seen them live up to their promises. The Radeon line of product was supposed to be revolutionary.. and before that, the rage chip was what they saw as the best technological avancement... They also did promise a huge improvement in drivers.... never saw so unstable drivers in my entire life. I had a built by ATI radeon 7500 in my box once. It only lasted 5 days in it. Now the drivers wont remove even tho the cards not there anymore...

[UtwigMU]

Let's analyse Parhelia.

If I make a GXXX analogy I'd say that Parhelia is something like G300 (between G200 and G400).
- excellent 2D like entire GXXX line
- lot's of features like G400
- 3D speed not on par with competition much like G200

For a company to be successful it is crucial to gain share in all segments or substantial share in high margin segments
Parhelia is a good card but if it outbenched the competition in most benchmarks (many people just go to tom's) it would sell even better.

In Hi-end 2D (DTP and video) parhelia is the ultimate card (gigacolor, glyph antialiasing, good IQ)
Unfortunately macs dominate this segment OTOH I think PC penetration is bigger
For 3D (CAD, modelling) it will suffice for midrange if opengl ICD will be OK - awaiting Ace's review
For 3D gaming alone price performance is not the best and it's performance is not the highest so I don't see many gamers getting one for gaming alone
For business/home 2D (DH) G4/550 has better price performance
The soho freelancer/enthusiast/gamer will also be satisfied with multiple features which none of the cards have

What Matrox should do is to develop a kind of powered by Matrox programme for service bureaus, marketing agencies and commercial printers with special offers, subscriptions and successful user stories
The business segment should be informed of the productivity boost of large desktop real estate (AFAIK P is the only card that can do dual DVI 1600×1200)
(brokers, architects, IT, ...) but I believe MMS line suits them better.
For gamers there should be cool factor like bundles, case stickers, blue pcb perhaps but without performance crown it will be harder to take a big share at this price

I think that Matrox jumped back on the 3D performance wagon in the right moment with DX9 and Longhorn later on the horizon

By the time parhelia 2 comes arround drivers and core will be optimized an we can expect better if not best performance comparing to competition

Matrox should develop 2 products
Midrange (G350 analogy)
Parhelia 2 (2LE/2MX)
- dieshrink to 9µm - considering UMC will skip 13µm process
- core cleanup
- full DX9 features - 2.0 pixelshader, 16 textures/clock
- appropriate clockbump 300/350DDR
- AGP 3.0 if enough boards out
This product should position itself between midrange gaming and entry workstation cards (200-300$ range) thus endangering nVs Quadro line

Hirange (G400 analogy)
- dieshrink to 9µm
- crossbar memory or revision of current memory controller
- lossless memory compression
- 8 rendering pipelines/4 tex filters per pass
- 2.0 pixelshaders
- stencil buffering
- clock 350/400 (I'm guessing here)
- 16bits/CC 128colour
- hardware video alpha
- HW MPEG decoding
- improved 16FAA or 4FSAA w/ minimum performance penalty
- independent 3H/3DVI
- 3D/CAD certified OpenGL icd, good workstation performance
This product should position itself between performance gaming and midrange PRO (400-500$) again diminishing nV30/40 quadro line

Thus Matrox would be fully 3D competitive and considering info about roadmap on anand I hope they will go into this direction.

About RSN - fairy in G400 tech demo was pointing into right direction, where did they got lost. Perhaps nudecasting should be considered as well - your ultimate cyber6 experince.