[Star_Hunter]

Originally Posted by The Baron

R420 is capable of 32x0.

It is not...

NoAA 2xAA 4xAA
Chip C Z C Z C Z
---------------------------
R300 8 8 8 8 4 4
R420 16 16 16 16 8 8
NV35 4 8 4 8 4 4
NV40 16 32 16 16 8 8

[Tim Murray]

According to the launch material, it is capable of 32x0. Maybe that was a lie. I dunno.

[Richard]

Originally Posted by Cryect

Well considering it doesn't use the physics engine or AI in timedemos its absolutely worthless except for video card comparison. The point of the article was to show how the game plays on a variety of systems not how a variety of systems can play timedemos.

That's why I said "I'd prefer" and not "They should have..."

[anaqer]

Originally Posted by The Baron

R420 is capable of 32x0.

That's kind of the reason why I asked that particular question back on page one...
(BTW, is that exclusive to R420? Didn't somebody say it works the same way on R3x0 too?)

[Tim Murray]

Originally Posted by anaqer

That's kind of the reason why I asked that particular question back on page one...
(BTW, is that exclusive to R420? Didn't somebody say it works the same way on R3x0 too?)

Mmm, reading the launch materials again, it says, "Up to 32 Z/stencil ops per clock." Interpret that as you will.

[jimmyjames123]

Originally Posted by The Baron

According to the launch material, it is capable of 32x0. Maybe that was a lie. I dunno.

I think what Star_Hunter is showing is that the "double-pumped" z-stencil nature of the R420 hardware is only applicable when AA is turned on, yet the number "32" would not necessarily be applicable because of the effect of AA on z/stencil ops. So, the behavior is similar to the NV40 with AA on, but not the same with AA off (and Star_Hunter's numbers show exactly that). In a game like Doom3 where almost all cards are struggling to enable any AA at 1600x1200 in the first place, this comes across as an advantage for NV.

[John Reynolds]

Hmm, finally read through the article. I'm going to have to try the game at 2048x1536 on my Intel system (P4 3.4EE, 2GB of RAM, and a PCIe 6800 GT @ 400/550).

[bloodbob]

Maybe the 420 is capable of 32x0 as such if its doing a double sided stencil test :P

[jimmyjames123]

Originally Posted by Star_Hunter

It is not...

NoAA 2xAA 4xAA
Chip C Z C Z C Z
---------------------------
R300 8 8 8 8 4 4
R420 16 16 16 16 8 8
NV35 4 8 4 8 4 4
NV40 16 32 16 16 8 8

Star_Hunter, could you explain how you came up with the numbers for the R300 and NV35? I guess I'm not sure why the R300 follows exactly the same pattern as the R420, and why the NV35 follows a different pattern than the NV40. Thanks.

[Reverend]

Hey guys, not to divert the topic, but Dave should be VERY busy this weekend or next week. Our relationship with NV isn't as bad as I thought.

[jimmyjames123]

On an aside, I think people should lay off of Dave a bit on that issue. The way I see it, he has very good connections and access to some information and is somewhat "in the know" with ATI, presumably significantly more so than with NVIDIA, and that translates on the forum at times. This issue is not so evident at many other major hardware review websites because their owners/writers often tend to stay out of the forums almost completely. Ultimately there are pluses and minuses to spending significant time on the forums. Stay on the forums for a while and try to speak one's mind, and one could be subject to a lot of verbal abuse. Stay away from the forums, and one could lose touch with many of the enthusiast community. I still like the former though. My advice to any writer/reviewer who is posting on forums is to try to show as much impartiality as they can (even if they have to bite their lip at times), and to try to generate and/or further discussion on a wide variety of topics in the industry.

[trinibwoy]

Originally Posted by Reverend

Hey guys, not to divert the topic, but Dave should be VERY busy this weekend or next week. Our relationship with NV isn't as bad as I thought.

SLI!!!

[mono]

Originally Posted by John Reynolds

Hmm, finally read through the article. I'm going to have to try the game at 2048x1536 on my Intel system (P4 3.4EE, 2GB of RAM, and a PCIe 6800 GT @ 400/550).

But that I have a mere 1GB of RAM, I would essentially have the [H] ultimate system, albeit my Opteron 150's registered RAM requirement takes a slight hit. My GT is also clocked at Ultra speeds. However the Dell 2001FP LCD maxes at 1600x1200 so I'll have to just suck it up and turn on some FSAA.

Anyone else considering making Tuesday or Wednesday a "sick" day from work?

[Reverend]

Originally Posted by trinibwoy

SLI!!!

Nah... DOOM3.

[Reverend]

Kyle, if you're reading this, why did you use FRAPS+gameplay? OK, you explained that this removes demo tinkering. But why not record a demo while playing the game (=demo represents actual gameplay) and then timedemo'ing that demo? Saves time, wouldn't it? And nobody will have the demo you recorded. And results are actually comparable across all those systems you used, without any doubt about truly comparable data accuracy.

[John Reynolds]

Originally Posted by mono

But that I have a mere 1GB of RAM, I would essentially have the [H] ultimate system, albeit my Opteron 150's registered RAM requirement takes a slight hit. My GT is also clocked at Ultra speeds. However the Dell 2001FP LCD maxes at 1600x1200 so I'll have to just suck it up and turn on some FSAA.

Anyone else considering making Tuesday or Wednesday a "sick" day from work?

Ohh, my new boss put me in for a bonus that I just got this week that would cover a 2001FP's cost. Hmm, how do I tell the wife she won't be getting that new furniture?

Sick day? Cough, nah, hasn't. . .cough, sniffle. . .crossed my mind.

[Scott_Arm]

Originally Posted by Reverend

Kyle, if you're reading this, why did you use FRAPS+gameplay? OK, you explained that this removes demo tinkering. But why not record a demo while playing the game (=demo represents actual gameplay) and then timedemo'ing that demo? Saves time, wouldn't it? And nobody will have the demo you recorded. And results are actually comparable across all those systems you used, without any doubt about truly comparable data accuracy.

I may be wrong, but I remember something about the timedemos not reflecting gameplay because the AI and physics calculations are removed. It's purely for testing graphics processing power, or something. So, he was testing real gameplay on a variety of systems so people would understand what they could expect with their CPU and graphics card combination. I think it was an amazing job.

[T2k]

Originally Posted by Star_Hunter

It is not...

NoAA 2xAA 4xAA
Chip C Z C Z C Z
---------------------------
R300 8 8 8 8 4 4
R420 16 16 16 16 8 8
NV35 4 8 4 8 4 4
NV40 16 32 16 16 8 8

:?

Well, I dunno WTH are these numbers from but IT IS.

ATI's X800-chips can render two Z/stencil ops per pipeline per clock - which makes XT PE capable for 32 Z/stencil ops per clock.

Read first: last two sentences on page 11 of http://www.ati.com/products/radeonx8...WhitePaper.pdf

If multisample anti-aliasing is enabled, however, the number of stencil and Z operations is doubled to two per clock cycle.

[T2k]

Originally Posted by anaqer

That's kind of the reason why I asked that particular question back on page one...
(BTW, is that exclusive to R420? Didn't somebody say it works the same way on R3x0 too?)

? I never heard about any 16 pipes R3xx chip...

[Star_Hunter]

Originally Posted by T2k

:?

Well, I dunno WTH are these numbers from but IT IS.

ATI's X800-chips can render two Z/stencil ops per pipeline per clock - which makes XT PE capable for 32 Z/stencil ops per clock.

Read first: last two sentences on page 11 of http://www.ati.com/products/radeonx8...WhitePaper.pdf

It was from this thread

http://www.beyond3d.com/forum/viewto...sc&start=0

[jimmyjames123]

Originally Posted by T2K

ATI's X800-chips can render two Z/stencil ops per pipeline per clock - which makes XT PE capable for 32 Z/stencil ops per clock.

They can render two Z/stencil ops per clock using AA, but doesn't the use of AA in itself effectively reduce the maximum Z/stencil ops?

[ChrisRay]

My main concern with the hardocp benches is how the 6800 NU is going to react with AA.

I know it only has 128 Megs of vram (But it still can do 8 Zixels @ 4x AA.) But it does loot its huge zfill advantage. I am mostly aiming for 4x/16x @ 1024x768 with 20 FPS + minimum with an Athlon XP 3200+ system with 1 gig of memory.

They can render two Z/stencil ops per clock using AA, but doesn't the use of AA in itself effectively reduce the maximum Z/stencil ops?

Pretty much. The z fill rate never goes above its no fill maximum. So it doesnt have an advantage when AA if thats what they are talking about.

This pretty much show how Zpixels look on both architectures.

NoAA 2xAA 4xAA
Chip C Z C Z C Z
---------------------------
R300 8 8 8 8 4 4
R420 16 16 16 16 8 8
NV35 4 8 4 8 4 4
NV40 16 32 16 16 8 8



As oposed above cept the

Geforce 6800 Non Ultra is


No AA : 12/24

2xAA 12/12

4xAA 8/8

[jimmyjames123]

Originally Posted by ChrisRay

As oposed above cept the

Geforce 6800 Non Ultra is


No AA : 12/24

2xAA 12/12

4xAA 8/8

And now we are starting to get a better understanding of one reason why the 6800 Non-Ultra (and 6800GT/U of course) seems to do so well with Doom3. The interesting thing about Doom3 is that virtually all the graphics cards struggle to maintain playable framerates with any AA enabled at 1600x1200, so NV's Z/stencil performance with no AA really helps to give them the advantage in this scenario. Also, [H]OCP claims that increasing resolution has a more significant positive impact on image quality than increasing AA, so this really bodes well for a 6 series type of architecture. Whoever said that the NV 6800 cards were designed with Doom3 in mind surely were not kidding!

Looking at the PCIe data, the 6800U leads the X800XT PE by 32% with 8xAF enabled at 1600x1200, and leads by 27% with 4xAA/8xAF enabled. Looking at the AGP data, the 6800U leads the X800XT PE by 24% with 8xAF enabled at 1600x1200, and leads by 21% with 4xAA/8xAF enabled. The 6800GT leads the X800Pro by 52% with 8xAF enabled at 1600x1200, and leads by 69% with 4xAA/8xAF enabled. Finally, the 5900 Ultra trails the 9800XT by 5% with 8xAF enabled at 1024x768, and leads by 19% with 4xAA/8xAF enabled. Interesting results, to say the least.

[FUDie]

Originally Posted by ChrisRay

My main concern with the hardocp benches is how the 6800 NU is going to react with AA.

I know it only has 128 Megs of vram (But it still can do 8 Zixels @ 4x AA.) But it does loot its huge zfill advantage. I am mostly aiming for 4x/16x @ 1024x768 with 20 FPS + minimum with an Athlon XP 3200+ system with 1 gig of memory.

Pretty much. The z fill rate never goes above its no fill maximum. So it doesnt have an advantage when AA if thats what they are talking about.

This pretty much show how Zpixels look on both architectures.

NoAA 2xAA 4xAA
Chip C Z C Z C Z
---------------------------
R300 8 8 8 8 4 4
R420 16 16 16 16 8 8
NV35 4 8 4 8 4 4
NV40 16 32 16 16 8 8



As oposed above cept the

Geforce 6800 Non Ultra is


No AA : 12/24

2xAA 12/12

4xAA 8/8

You don't really think that, clock per clock, the 12 pipe 6800 non-Ultra is the same speed as the 16 pipe 6800 Ultra in 4x AA, do you?

-FUDie

[ChrisRay]

Clock per clock? With same memory configuration? Yes I do Fudie.

Apparently you missed this. Since we were discussing zfill. I definately considered it relevent,

http://www.beyond3d.com/forum/viewto...sc&start=0