New NV40 rumours(Stoled/Quoted from Uttar)

[digitalwanderer]

Originally posted by The Baron
free AA to me is AA with <5% performance hit. that'd be nice.
But it wouldn't be free, it'd be a "less than 5% performance hit". :(

I dig the concept you're aiming at, but I hate the term "free AA" since it probably never will be. A very minor performance hit, yes; free, no.

[The Baron]

why would AA ever be free? it has to render something effectively multiple times, so you could reduce the hit by quite a bit with enough color compression and the like, but you're never going to make it free.

that'd be like alchemy, homes. :)

[volt]

Originally posted by hkultala
FSAA just requires more work to be done than no FSAA.

If your FSAA is completely free, it means that you render suboptimally when not doing antialiasing.

Getting as close as possible to doing just the necessarily amount more than without AA is what should be preferred,
"not wasting anything when doing AA"

Originally posted by The Baron
free AA to me is AA with <5% performance hit. that'd be nice.

[Nv40]

by the info here.. and by looking at the sucess of the R300.. seems -very likely- that ATI will look to repeat again their strategy with the R300.

R420 -> brute force performance with minimal increase in Dx9 Features.. most transistors used for speed.

Nv40 -> tons of new features with (including new AA and AF modes)- with Ok performance in PS2.0/Vs2.0 .if Nvidia still support Fp16 and lower seems very likely that they dont have enough confidence that their Fp32 PixelS performance will be enough to compete with ATI. ;)

my prediction is that ..

1)NVidia with the Nv40 will look to atract gamers and compete solely by IQ/features/with ok speed in PS2.0. since lately the focus in IQ comparisons is greater,than the focus in performance and they have been critized alot the their IQ. they will not look to compete in price (in the highend) since its very small the enthusiast market and they lead the market share. but Nvidia will rely mostly to make money in the low end $100-$200-market. to repeat again their selling sucess with the fx5200.

2)ATI in the other will try to re-gain their market share at all cost, to sell lots of cards even in the highend. and will go ALL the way for speed ,at a great price- in the R420 ,since they have been congratulated enough by their IQ by gamers,so i really doubt they will touch AA or AF .. an minimal increase in Dx9 features ,just to save transistors in the card .

i predict that Nvidia will look with the Nv40 to be the Ultimate Doom3 card and games in that engine and that ATI will want to make the R420 the absolute and only way to go for Hl2 . nobody spend 6 million in a contract for nothing. :) the sucess in sales of ATI and Nvidia will be mostly around those two games alone. :)

it will be VERy Interesting the fight in 2004 for games performance , i dont think that there will be and all around card for all games ,but more ATI and NVidia winning and losing benchmarks in many games... :D

[zakelwe]

My maths might be incorrect but 48GB/sec bandwidth is GDDR2 with frequency 750Mhz, but doesn't this memory comes in at 800, 700 and 600Mhz for the top end ????

Is not 800Mhz ( 51.2GB/sec ) for the Ultra, 700Mhz ( 44.8 GB/sec ) for the non ultra and 600 for the XT//MX :) ( 38.4Gb /sec ) more likely ?

Regards

Andy

[StealthHawk]

Originally posted by hkultala
FSAA just requires more work to be done than no FSAA.

If your FSAA is completely free, it means that you render suboptimally when not doing antialiasing.

Getting as close as possible to doing just the necessarily amount more than without AA is what should be preferred,
"not wasting anything when doing AA"

Wouldn't you want hardware/drivers to be designed to be optimal for high IQ using features like FSAA though? Hardware already reached the point where it's pretty damn fast without FSAA/AF.

Personally I wouldn't care if regular rendering was not being done at 100% efficiency if that meant that FSAA performance would be higher.

The GeForceFX has already done a good job at getting low performance hit 2x FSAA into the market place. Granted 2x FSAA isn't all that great, but it's nice to be able to enable it for a much smaller performance hit than on other (previous) cards. With a high end gfFX you can basically set 2x FSAA on and forget about it.

[StealthHawk]

Originally posted by zakelwe
My maths might be incorrect but 48GB/sec bandwidth is GDDR2 with frequency 750Mhz, but doesn't this memory comes in at 800, 700 and 600Mhz for the top end ????

Your math is correct. The formula for calculating bandwidth is (bus width) / 8 * memory speed = MB/sec.

The fastest DDRII memory in use today is on the gfFX5900Ultra, coming in at 475MHz. Assuming NV40 does indeed launch with 48GB/sec of memory bandwidth, it will be using 750MHz DDRII as you suggested. I don't know what the fastest memory actually available right now is. Nor what the fastest memory will be once NV40 actually comes out.

[Hanners]

Originally posted by The Baron
why would AA ever be free? it has to render something effectively multiple times, so you could reduce the hit by quite a bit with enough color compression and the like, but you're never going to make it free.

that'd be like alchemy, homes. :)

Indeed... Nothing is free in 3D.

[nobie]

The fastest DDRII memory in use today is on the gfFX5900Ultra, coming in at 475MHz. Assuming NV40 does indeed launch with 48GB/sec of memory bandwidth, it will be using 750MHz DDRII as you suggested. I don't know what the fastest memory actually available right now is. Nor what the fastest memory will be once NV40 actually comes out. [/B]

http://www.xbitlabs.com/news/memory/display/20030828165211.html
"It looks like GDDR2 memory starts gaining its popularity as Samsung Electronics today announced production of its 256Mb Graphics Double Data Rate2 (GDDR2) SDRAM. The 256Mb capacity will enable building graphics cards with up to 512MB of memory using PCBs of typical sizes. Besides, the 1600MHz DRAM chips will allow extremely fast graphics cards.

Samsung claims that the new devices had already been supplied to leading graphics card manufacturers, suggesting that the actual products with such amazing memory are not too far away. I believe that we will hardly see any graphics cards with 512MB of mind-blowing fast memory in late 2003 or early 2004, but a year from now there definitely will be high-end graphics products with 512MB of memory. Such graphics cards may be based on the successors of the NV40 and R420 designs, for example."

[Blacklash]

As long as the NV40 runs the games I play with good frame rates and solid IQ, I will be buying it.

[oqvist]

Originally posted by Nv40
by the info here.. and by looking at the sucess of the R300.. seems -very likely- that ATI will look to repeat again their strategy with the R300.

R420 -> brute force performance with minimal increase in Dx9 Features.. most transistors used for speed.

Nv40 -> tons of new features with (including new AA and AF modes)- with Ok performance in PS2.0/Vs2.0 .if Nvidia still support Fp16 and lower seems very likely that they dont have enough confidence that their Fp32 PixelS performance will be enough to compete with ATI. ;)

my prediction is that ..

1)NVidia with the Nv40 will look to atract gamers and compete solely by IQ/features/with ok speed in PS2.0. since lately the focus in IQ comparisons is greater,than the focus in performance and they have been critized alot the their IQ. they will not look to compete in price (in the highend) since its very small the enthusiast market and they lead the market share. but Nvidia will rely mostly to make money in the low end $100-$200-market. to repeat again their selling sucess with the fx5200.

2)ATI in the other will try to re-gain their market share at all cost, to sell lots of cards even in the highend. and will go ALL the way for speed ,at a great price- in the R420 ,since they have been congratulated enough by their IQ by gamers,so i really doubt they will touch AA or AF .. an minimal increase in Dx9 features ,just to save transistors in the card .

i predict that Nvidia will look with the Nv40 to be the Ultimate Doom3 card and games in that engine and that ATI will want to make the R420 the absolute and only way to go for Hl2 . nobody spend 6 million in a contract for nothing. :) the sucess in sales of ATI and Nvidia will be mostly around those two games alone. :)

it will be VERy Interesting the fight in 2004 for games performance , i dont think that there will be and all around card for all games ,but more ATI and NVidia winning and losing benchmarks in many games... :D

And yet ATI has superior FSAA, superior filtering and superior PS 2.0 performance?? Isn´t it really the other way??? I mean the NV35 scores best with no antialiasing and no anisotropic and at lower res generally.

But as it is now ATI both got the best performance and IQ.

2. ATI obtained their awesome AA and AF buy hard work and some clever choices by their programmers. And ATI has just about always opted more for IQ then nVidia. Even in old rage PRO days ATI had superior image quality over the TNT but couldn´t in anyway compete in performance and drivers.

And everybody knows how the 8500 forced nVidia to improve their image quality considerably with their Geforce 4 line. Geforce 3:s was solely optimized for speed really and it showed.

nVidia got a lot to proof with it´s NV40. For me it´s the definitive win or loose video card for nVidia. If the NV40 is such a failure as previous FX cards I am wondering if nVidia is prepared to continue trying to compete in the high end enthusiast market or if they will pull their efforts mainly on the budget market.

But I am as curious as anyone how it will be with the NV40 vs the R420. I am just hoping to god nVidia will skip their "optimization" scheme with that one. Hopefully they will do if it performs like they hope.

It should be able to compare better now. It seems like the R420 will be an 8 pipeline card and so it seems the NV40 will be at least. It´s just a question on how effective the nVidia employees can make their new baby. They seem have lacked aim on how they best should develop their FX series somewhat. The FX series seemed quite unbalanced.

[Cotita]

Originally posted by StealthHawk
Your math is correct. The formula for calculating bandwidth is (bus width) / 8 * memory speed = MB/sec.

The fastest DDRII memory in use today is on the gfFX5900Ultra, coming in at 475MHz. Assuming NV40 does indeed launch with 48GB/sec of memory bandwidth, it will be using 750MHz DDRII as you suggested. I don't know what the fastest memory actually available right now is. Nor what the fastest memory will be once NV40 actually comes out.

Actually the fastest DDR2 memory in use today is on the now discontinued Geforce FX5800 ultra which runs at 500mhz, the Geforce fx5900ultra runs at 450mhz and is DDR1 the 5950 ultra runs at 475mhz and is also DDR1

[Cotita]

I think that instead of Full screen antialiasing, nvidia and ATI should implement edge antialiasing a la matrox parhelia, at least as an option.

[The Baron]

Parhelia had edge AA, but it was quite buggy, no?

[StealthHawk]

Originally posted by Cotita
Actually the fastest DDR2 memory in use today is on the now discontinued Geforce FX5800 ultra which runs at 500mhz, the Geforce fx5900ultra runs at 450mhz and is DDR1 the 5950 ultra runs at 475mhz and is also DDR1

You're right about that- my mistake :o

[StealthHawk]

Originally posted by oqvist
2. ATI obtained their awesome AA and AF buy hard work and some clever choices by their programmers. And ATI has just about always opted more for IQ then nVidia. Even in old rage PRO days ATI had superior image quality over the TNT but couldn´t in anyway compete in performance and drivers.

Most people would say that NVIDIA has always had the edge in AF quality due to the fact that NVIDIA's AF doesn't compromise on "off-angles" as much as ATI's does.

And everybody knows how the 8500 forced nVidia to improve their image quality considerably with their Geforce 4 line. Geforce 3:s was solely optimized for speed really and it showed.

Um, what? The gf4 is the same thing as a gf3 except it's faster and hence allows you to dial up the IQ more. The IQ is the same.

[-=DVS=-]

Whatever they will bring is definetly much better then todays cards lets just all hope it doesn't have leaf blower ;) or massive heatsink.

Also looking forward to Power Vr cards :drooling:

[oqvist]

Originally posted by StealthHawk
Most people would say that NVIDIA has always had the edge in AF quality due to the fact that NVIDIA's AF doesn't compromise on "off-angles" as much as ATI's does.



Um, what? The gf4 is the same thing as a gf3 except it's faster and hence allows you to dial up the IQ more. The IQ is the same.

Nope! 2D is vastly improved on the gf 4 and they wasn´t as blurry in resolutions above 1024x768.

The IQ is night and day. Believe me I traded a asus Geforce 3 ti-500 to a asus Geforce 4 ti-4600. And I could run very much the same settings on the ti-500 and ti-4600 at that time. This is no secret. 95 % would agree that the 8500:s image quality was far superior to the Geforce 3:s. Between the Geforce 4 and 8500 the gap was diminished immensely.

And ATI do support higher level of AF. And to complain about ATI not offering full AF when nVidia don´t do full Trilinear... Well... The image quality doesn´t suffer from ATI not rendering off angles and it´s still better than nVidias anisotropic in real games even if they are really equal.

[Cotita]

Originally posted by The Baron
Parhelia had edge AA, but it was quite buggy, no?

I really don't know if they made it work right, but I see no reason why wouldn't nvidia or ATI couldn't make it better.

[Cotita]

Originally posted by oqvist
Nope! 2D is vastly improved on the gf 4 and they wasn´t as blurry in resolutions above 1024x768.

The IQ is night and day. Believe me I traded a asus Geforce 3 ti-500 to a asus Geforce 4 ti-4600. And I could run very much the same settings on the ti-500 and ti-4600 at that time. This is no secret. 95 % would agree that the 8500:s image quality was far superior to the Geforce 3:s. Between the Geforce 4 and 8500 the gap was diminished immensely.

And ATI do support higher level of AF. And to complain about ATI not offering full AF when nVidia don´t do full Trilinear... Well... The image quality doesn´t suffer from ATI not rendering off angles and it´s still better than nVidias anisotropic in real games even if they are really equal.

I have a palit daytona geforce 3 ti200 and a visiontek geforce4 ti 4600, I put my brothers and my pc side by side with same 17" acer monitor and there is no difference in IQ in either 2d or 3d.

[StealthHawk]

Originally posted by oqvist
Nope! 2D is vastly improved on the gf 4 and they wasn´t as blurry in resolutions above 1024x768.

The IQ is night and day. Believe me I traded a asus Geforce 3 ti-500 to a asus Geforce 4 ti-4600. And I could run very much the same settings on the ti-500 and ti-4600 at that time. This is no secret. 95 % would agree that the 8500:s image quality was far superior to the Geforce 3:s. Between the Geforce 4 and 8500 the gap was diminished immensely.

It's a difference of RF filters. NVIDIA didn't do anything to the core logic of the chips to account for the difference. It's not NVIDIA's fault that manufacturers generally used cheap RF filters for the gf3. You can fault them for not making board manufacturers use better filters, but really it's all the fault of the board manufacturers.

In other words, the gf3 was not slower because most board manufacturers used poor filters. The gf4 did not get any slower or faster because it used higher quality filters either.

And ATI do support higher level of AF. And to complain about ATI not offering full AF when nVidia don´t do full Trilinear... Well... The image quality doesn´t suffer from ATI not rendering off angles and it´s still better than nVidias anisotropic in real games even if they are really equal.

The difference between 16x and 8x is pretty small. You have look far into the horizon to see the difference.

NVIDIA does trilinear in both gf3 and gf4. So I don't see any relevance to your point when comparing AF between the r8500 and the gf3 or gf4. The AF on the r8500 is even weaker than on the r9700, it only supports bilinear AF, and has more off-angle problems than the r9700.

When comparing NV3x AF against R3xx AF, I still bet lots of people would prefer NVIDIA's brilinear AF to ATI's trilinear with off-angle reduced filtering AF. The brilinear that NVIDIA is using does not decrease IQ much. It's still bad that it's being forced on users when the hardware is capable of more, but...the thing is that it may be you opinion that ATI's AF looks better. But you can find lots of people who prefer NVIDIA's implementation and think it looks better. Saying that one implementation is better than another when both have different IQ tradeoffs is ludicrous, as it is completely subjective to determine which IQ tradeoff is worse.

[Viral]

IMO nvidias AF is better, but has a larger performance hit. IMO ATI's AA is better, but has a larger performance hit.

When both are enabled, they both have strengths and weaknesses that in the end makes the overall IQ very similar, and both cards recieve a similar performance hit. I am glad Nvidia came this far, and i hope they push their IQ even further with the NV4x.

[Hanners]

Originally posted by Viral
IMO nvidias AF is better, but has a larger performance hit. IMO ATI's AA is better, but has a larger performance hit.

I can't really think of any scenarios where ATis AA takes a much larger performance hit that nVidias...

[Viral]

Have you seen benchmarks where they test AA and AF individually? last i remember, ATI's AA generally took a bigger performance hit when enabled. This was comparing R350 to NV35, so perhaps memory bandwidth was the sole reason for this.

If what you are saying is true, and NV3x takes a larger performance hit from 8xAF than R3xx, yet R3xx doesn't take a larger performance hit from 4xAA than nvidia, then wouldn't benchmarks show nvidia being equal/ahead/closer to ATI with AA/AF off, but further behind when both are turned on? it should, however this has not generally been the case in any reviews i've read.

[oqvist]

Originally posted by StealthHawk
It's a difference of RF filters. NVIDIA didn't do anything to the core logic of the chips to account for the difference. It's not NVIDIA's fault that manufacturers generally used cheap RF filters for the gf3. You can fault them for not making board manufacturers use better filters, but really it's all the fault of the board manufacturers.

Yes it is! It doesn´t matter whos fault it´s. I am just stating the obvious. The geforce 3:s suffered from it nevertheless. And they also got lower RAMDAC which didn´t help either.

In other words, the gf3 was not slower because most board manufacturers used poor filters. The gf4 did not get any slower or faster because it used higher quality filters either.

Slower? You mean it was more blurry because of the bad filters? I never said the ti-500 was as fast as a ti-4600. I just said I could compare the image quality under the same settings...



The difference between 16x and 8x is pretty small. You have look far into the horizon to see the difference.

NVIDIA does trilinear in both gf3 and gf4. So I don't see any relevance to your point when comparing AF between the r8500 and the gf3 or gf4. The AF on the r8500 is even weaker than on the r9700, it only supports bilinear AF, and has more off-angle problems than the r9700.

Yes Geforce 3 and geforce 4 but not on the FX series in all games...

When comparing NV3x AF against R3xx AF, I still bet lots of people would prefer NVIDIA's brilinear AF to ATI's trilinear with off-angle reduced filtering AF. The brilinear that NVIDIA is using does not decrease IQ much. It's still bad that it's being forced on users when the hardware is capable of more, but...the thing is that it may be you opinion that ATI's AF looks better. But you can find lots of people who prefer NVIDIA's implementation and think it looks better. Saying that one implementation is better than another when both have different IQ tradeoffs is ludicrous, as it is completely subjective to determine which IQ tradeoff is worse.

I think ATI:s tradeoff in image quality is significantly lower than nVidias tradeoff. But the fact that ATI takes a much less performance hit also helps ATI:s anisotropic filtering allowing you to push it even further.

Of course it´s subjective. Image quality is a very subjective matter.