Nvidia/Ati - GPU roadmap

[overclocked]

I dont really think you need more than 4x AA / 8xAF.

 

[Sage]

Only 30% On Parhelia? C'mon... noooo way. You've gotta be kidding... turn on everything (AA, AF) and start playing UT2003 on three monitors...

Well keep in mind 16x FAA has a very surprisingly small performance hit.

And if you turn on AA or AF, the same % hit will be applied to both resolutions, meaning the performance difference will remain the same, ~30%.

30% for tripling the work done?
i dont buy it.
Either that or the game was CPU bound before you did SG

actually, most games really do take merely a 30pct performance hit. A 30pct hit can be rather large in many games if your performance is already... ehhh... well... parhelia-like

 

[Althornin]

30% for tripling the work done?
i dont buy it.
Either that or the game was CPU bound before you did SG

actually, most games really do take merely a 30pct performance hit. A 30pct hit can be rather large in many games if your performance is already... ehhh... well... parhelia-like [/quote]

do you have any reviews that show this?
Because you stating it again does little to nothing to convince me.
enabling SG doesnt magically make the chip almost three times as efficient, so that when doing 3 times the work it barely loses any performance....

 

[Arun]

I dont really think you need more than 4x AA / 8xAF.

And I really don't think we need more than 640K.


Uttar

 

[Prime]

Aren't there a number of memory modules being produced in 110nm?

Micron's GDDR3 is .11u per this release on the front page of Rage3D:

GDDR3, which is fabricated on Micron's 0.11-micron process, is the fastest memory device available in the market today, providing the highest performance in both per-pin bandwidth and aggregate bandwidth, Micron said.

 

[Sage]

I dont really think you need more than 4x AA / 8xAF.

and apparently you havn't seen V5 6000 8xFSAA.....

 

[overclocked]

Hehe it would be fun to own a rare card like that.

 

[Dave H]

I dont really think you need more than 4x AA / 8xAF.

and apparently you havn't seen V5 6000 8xFSAA.....

I certainly haven't seen it, but I have a hard time believing it would be much better than R3xx's 6xMSAA/16xAF. It offers a lower level of texture antialiasing where it's most needed, it's not gamma corrected, and the default sample pattern (from what I've read) is a (sub-optimal) rotated grid (although apparently it can be reprogrammed to a sparse sample pattern).

None of which is to say that we don't need (well, perhaps "need" is a strong word ) more than 6xMSAA/16xAF in the long run.

 

[Sage]

actually, the sampling pattern on the V5 is superior to even the R3x0. It is only because the R3x0 uses high level of anisotropy and gamma corrected fsaa that it is able to compete with the IQ offered by the V5 at 4x. Remember, the V5 is actually supersampling which produces VERY good texture quality, and at 8x, in my opinion, better than 16x AF.

 

[Althornin]

actually, the sampling pattern on the V5 is superior to even the R3x0. It is only because the R3x0 uses high level of anisotropy and gamma corrected fsaa that it is able to compete with the IQ offered by the V5 at 4x. Remember, the V5 is actually supersampling which produces VERY good texture quality, and at 8x, in my opinion, better than 16x AF.

Actually, the sampling pattern is NOT better, according to recent threads here (*recent = past few months).

And Colourless can confirm that indeed, the DEFAULT 8x sample pattern blew. Sure, he figured otu a way to set the sample pattern, so you could choose a sparse grid - but why should he have to?
As for your 8x SS = (texture aliasing wise) 16xAniso, i'd disagree.

 

[Sage]

but the 4x sampling pattern looks better (to my eyes) than ATi 6x

 

[Althornin]

but the 4x sampling pattern looks better (to my eyes) than ATi 6x

the 4x pattern is simply a rotated grid.
Its WORSE for edge AA than ATI's, because it lacks gamma correction.

 

[Ailuros]

but the 4x sampling pattern looks better (to my eyes) than ATi 6x

the 4x pattern is simply a rotated grid.
Its WORSE for edge AA than ATI's, because it lacks gamma correction.

Due to gamma correction only?

In terms of edge antialiasing it's rather a 4*4 grid vs 6*6 grid + gamma correction.

As far as Supersampling producing better texture quality than 16x sample Anisotropic (apart from some corner cases), someone is either very shortsighted even in front of a monitor or never has used either/or in real time.

 

[stevem]

How sure are you of this? Id like to think that some of the vendors (abit, asus,...) would love to add in their press release how they can do multi-video cards with their PCI-E motherboards. I hope someone tries it. Or is the problem that current version of PCI-E only has the ability to have one 16x tunnel. That brings up the question as to whether a 8x tunnel could be used for the second video card.

The 16x slot maxes out first gen available bandwidth. Vendors that decide on a mix of 8x & 4x slots are just dividing up the bandwidth of the single 16x slot. High bandwidth 16x/8x/4x slots are connected via the MCH. Lower bandwidth 1x slots are via the ICH. This probably means CSA will be deprecated as 250MB/s bidirectional from 1x devices will be more than adequate for GB LAN. I don't know how this may relate to AMD64 multiple PTP busses with PCI-E 16x via HT.

Edit: The other benefit of the 16x slot is the 60W power budget & cooling schematics. I wonder how surprised people will be with NV30 magnitude cooling systems on next gen cards...