[digitalwanderer]

Originally Posted by DemoCoder

Personally if NV or ATI are within 15% performance of each other, I think they share the crown, and we all win.

If they have the same image quality, then yes. 8)

[John Reynolds]

Originally Posted by digitalwanderer

If they have the same image quality, then yes. 8)

I think a better word would be comparable.

[digitalwanderer]

Originally Posted by John Reynolds

I think a better word would be comparable.

Fair enough. It ain't exactly gotta be exact, but it does have to be close there too.

[Florin]

Originally Posted by digitalwanderer

If they have the same image quality, then yes. 8)

And the same level of driver support, yeah 8)

Linux users, dual head gamers and people who experiment with 64-bit OSes badly need some tlc from ATI

[THe_KELRaTH]

What do you think the chances are that ATI will implement an AA format / technique that will remove jaggies on Alphatest artifacts?
While the NV35 offered a solution with it's mixed modes they were too slow to be useful in most cases the NV40 would seem to be fast enough based on the UT2004 performance.

[Mintmaster]

Originally Posted by Chalnoth

I request you quote exactly what you claim I contradicted.

I think I'll give it a shot.

Originally Posted by Chalnoth

Besides, I don't care all that much about performance anymore. For me it's all about the image quality and reliability.

Originally Posted by Chalnoth

Oh, and I do expect the NV40 to outperform the R420 (something that becomes somewhat less likely if ATI went for a sub-PS/VS 3.0 card), given that they have a much longer history of higher-performing parts.

Obviously you do care or you wouldn't say this. And besides, higher performance means higher image quality settings are possible. If you read Brent's most recent review, you'd know what I mean.

Originally Posted by Chalnoth

This is why I continue to complain about ATI's texture aliasing, and why I am rather excited about the pictures. That FSAA looks very nice....I just hope it's what we'll get.

Right. You hate ATI's aliasing, but you are excited about the lost details in the blur of these pictures. The FSAA looks "very nice", even though it's the same as NV30's 4xOGMSAA 9-tap blur.

Originally Posted by Chalnoth

Fortunately, we really haven't had much of a problem with a part that has speed, but doesn't advanced technology, significantly outclassing a part that has much better technology for the future...

Well, already replied to that in here. I guess you didn't specifically contradict yourself, but I'm sure if I look hard enough you'll have said something about the GF4MX holding back pixel shader adoption.

Don't get me wrong, I value your opinion much more than say HB and maybe even more than DW and DT, but T2K has a point.

Anyone else care to make a contribution?

[Chalnoth]

Originally Posted by StealthHawk

If they had gotten NV30 out earlier in 2002 it would have been clocked lower than the NV30 we saw in 2003.

I'm not so sure. Essentially, it may well be that the NV30 that was released was not the original NV30 that nVidia had planned. Remember the rumors that the NV30 was delayed because they couldn't get the low-k dielectrics to work?

Also recall that, apparently, Microsoft did not add _pp to DirectX 9 until very late, and there was as yet no indication that the NV30 would require integer precision to work at speed. The first indication of that came around August (though nobody noticed it then), when nVidia released their "CineFX" white papers.

And then there are the statements from some people on these boards who have seen pictures of the NV30's layout that claim that its design appeared to be almost entirely auto-routed. Auto-routing should be the first stage of design, with further refinement being used to improve the design in successive steps. Only if a design is horribly rushed should it be almost completely auto-routed.

Also recall that the NV35 essentially doubled the FP32 performance of the NV30 while only barely increasing the transistor counts.

Then there's the fact that each previous generation has had improved image quality:
128->TNT: 32-bit color, MIP mapping
TNT->GeForce: 2x anisotropic filtering, FSAA, good trilinear
GeForce->GeForce3: 8x anisotropic filtering, improved FSAA

....but the GeForce FX has essentially the same FSAA quality as the GeForce4, and the anisotropic filtering quality is worse (i.e. brilinear).

I therefore propose that the NV30 that we see was nVidia's attempt to release something, to not end up like 3dfx. I propose that we have not seen what nVidia originally intended to release as the NV30. I further propose that the NV40 will be the technological derivative of this planned NV30, not the NV30 that was sold. These early rumors of it being a 16x1 architecture seem to be one first step in verifying this. Anyway, there's not much time left to wait.

Even if NVIDIA re-energizes themselves, you're forgetting that this isn't the same old-ATI of generations past either.

Yes, ATI does have Dave Orton. Their culture has apparently shifted towards a much more aggressive stance. If this wasn't the case, I'd say that nVidia would certainly come ahead in this next generation. But no, it will be an interesting battle (in technology, performance, and market share).

[MMJ]

What is this NV40 sample like noise-wise? Is it as quiet as 9800pro's?

[Chalnoth]

Originally Posted by Mintmaster

Obviously you do care or you wouldn't say this. And besides, higher performance means higher image quality settings are possible. If you read Brent's most recent review, you'd know what I mean.

What you're seeing is me separate my personal cares about what I want for my next video card, and what kinds of things will make other people want a specific video card.

Right. You hate ATI's aliasing, but you are excited about the lost details in the blur of these pictures. The FSAA looks "very nice", even though it's the same as NV30's 4xOGMSAA 9-tap blur.

I don't think that's true at all. And no, I don't like the Quincunx/4x9 modes. The screenshots posted so far are vastly too complex to tell with any reasonable certainty what the exact FSAA mode used was. But no, I don't think it's any derivative of 4x OGMSAA, given that the low-angled edges show very little aliasing.

[Chalnoth]

Originally Posted by MMJ

What is this NV40 sample like noise-wise? Is it as quiet as 9800pro's?

Apparently it has two molex connectors. That would seem to indicate lots of power consumption, which in turn means lots of heat generated, which in turn means a powerful cooling solution. Furthermore, as an early sample, I'm sure not much care has taken toward quietness, as functionality is much more important.

I'm sure some manufacturers will come out with quieter cooling solutions, however.

[pocketmoon_]

Originally Posted by Chalnoth

Apparently it has two molex connectors. That would seem to indicate lots of power consumption

I think thats just down to initial PCB layout design. It probably a matter of re-routing some power tracks to get back to the single molex.

[volt]

Wait, so the other molex is there just "in case"

[russo121]

Originally Posted by Chalnoth

...I therefore propose that the NV30 that we see was nVidia's attempt to release something, to not end up like 3dfx. I propose that we have not seen what nVidia originally intended to release as the NV30. I further propose that the NV40 will be the technological derivative of this planned NV30, not the NV30 that was sold. These early rumors of it being a 16x1 architecture seem to be one first step in verifying this. Anyway, there's not much time left to wait.

And I propose you go to the doctor This thread is killing your brain...

[pocketmoon_]

Originally Posted by volt

Wait, so the other molex is there just "in case"

I think it's probably needed in the version that's out there. Perhaps It's tricky to route the power to everywhere on the card, especially as track density and number of pcb layers increases. Motherboards don't have a single power connector - well my recent ones don't. That makes it easier for the PCB designers. Perhaps (damn I used that word again) the complexity of the graphics card PCB's is getting to that point where power routing is an issue.


Is there an emoticon for talking out my @rse ?

[Tahir2]

Originally Posted by russo121

And I propose you go to the doctor This thread is killing your brain...

I propose that the NV30 was a figment of NVIDIA's imagination and in fact it did not ever exist. It was just a bad NVIDIA nightmare.

P.S. In all seriousness your proposal is ridiculous, as the NV35 was what then? An attempt to fix the NV30 that was meant to be an NV40 really? Makes me wonder how you perceive reality sometimes Chalnoth. That was way beyond 'rose tinted glasses' and definitely into 'been eating too many Magic Mushrooms!!' territory.

[volt]

Originally Posted by pocketmoon_

I think it's probably needed in the version that's out there. Perhaps It's tricky to route the power to everywhere on the card, especially as track density and number of pcb layers increases. Motherboards don't have a single power connector - well my recent ones don't. That makes it easier for the PCB designers. Perhaps (damn I used that word again) the complexity of the graphics card PCB's is getting to that point where power routing is an issue.


Is there an emoticon for talking out my @rse ?

Seems reasonable, though maybe the A0-A1 samples require second molex for better stability. Maybe retail boards will have just one -- which I hope for.

By the way, does anyone know PCB layer count for NV4x? I'd assume it will be a tad higher than current 12 for NV3x -- or I'm tad wrong heh

[FUDie]

Originally Posted by Chalnoth

Then there's the fact that each previous generation has had improved image quality:
128->TNT: 32-bit color, MIP mapping
TNT->GeForce: 2x anisotropic filtering, FSAA, good trilinear
GeForce->GeForce3: 8x anisotropic filtering, improved FSAA

....but the GeForce FX has essentially the same FSAA quality as the GeForce4, and the anisotropic filtering quality is worse (i.e. brilinear).

This is a pretty skewed look at things. First, the SSAA on the GeForce would have worked just fine on the TNT, had the drivers exposed it, so you can't really call that a strength of the GeForce over the TNT. Also, I believe the TNT did support true trilinear, but it wasn't used because of the performance hit.

But at least you acknowledge some of the flaws on the FX.

I therefore propose that the NV30 that we see was nVidia's attempt to release something, to not end up like 3dfx. I propose that we have not seen what nVidia originally intended to release as the NV30. I further propose that the NV40 will be the technological derivative of this planned NV30, not the NV30 that was sold. These early rumors of it being a 16x1 architecture seem to be one first step in verifying this. Anyway, there's not much time left to wait.

I think that you should hope that NV40 has nothing in common with NV30.

-FUDie

[AlphaWolf]

Originally Posted by Tahir

P.S. In all seriousness your proposal is ridiculous, as the NV35 was what then? An attempt to fix the NV30 that was meant to be an NV40 really? Makes me wonder how you perceive reality sometimes Chalnoth. That was way beyond 'rose tinted glasses' and definitely into 'been eating too many Magic Mushrooms!!' territory.

I think those glasses most definitely have a tint of green, not rose.

[Mintmaster]

Originally Posted by Chalnoth

I don't think that's true at all. And no, I don't like the Quincunx/4x9 modes. The screenshots posted so far are vastly too complex to tell with any reasonable certainty what the exact FSAA mode used was. But no, I don't think it's any derivative of 4x OGMSAA, given that the low-angled edges show very little aliasing.

Did you see this? What about this? There's no way you're going to say Bambers is wrong, as he's one of the most knowledgeable people on these boards when it comes to AA.

Hate to bust your bubble (well, not really), but looks like pretty damn convincing evidence to me. It looked like 3x3 or 4x4 OG to me before I managed to get through that thread, but when I saw Hyp-X's comparison shot there was no doubt.

Compare the NV40 pic (assuming it's NV40) with ATI's rendering. When you look at the details on the wall there is a very large difference. ATI's textures are much clearer, and would be so even if you nudge the LOD a bit to reduce "ATI's horrible aliasing" ( ) that you hate so much.

It looks like NVidia didn't adopt any new antialiasing strategy, such as jittered sample locations. They're just going to remarket Quincux as better AA. The edges are not bad (ATI's 4x is still better), but Quincux will always blur textures because the same texture sample is smeared over 4 pixels (albeit a weighted smear). You might as well just play in lower resolution - I'll take ATI's 6xAA at 1024x768 over that at 1280x960 anyday.

EDIT: Of course, there's the most likely possibility that these are just fake shots, and we have absolutely no idea what NV40 will bring to the table.

[Pete]

Originally Posted by Evildeus

Well, DB must be under NDA since Cebit .

Thankfully, it seems we don't have long to wait before we're all in on the secret.

There's something that i don't understand otherwise. The 175/210m transistors rumor :/ Partners of Nv say that the NV40 will be 3 times faster than the 5950, and that NV41 will be 2 times faster than 5950. They add that the Nv40 should be available in may/june and the NV41 in july-august.

So if the NV40 175m and 210m were tape out at the same time, and that the 175 is now the Nv41, why is it released 3 months later? Stock issues? I find quite ironic that the 210m should get out before the 175m chip.

nV typically releases the high-end first and the mid-range a few months later, so this wouldn't be out of character for them. ATi did the same with 9700P and 9700 (though I honestly don't remember what happened with 9800/P). nV releasing two chips, one high-end and one mid-range, may jive with Huang's saying they would increase their "vertical range" to span five, rather than three, chips/cards. I don't know how that'll help the market, but I suppose it might help their profits if they extend their card range to accomodate chips that fail qualification at a certain speed.

[Doomtrooper]

Lod bias is for graph wars...you would be surprised at what LOD bias adjustments do for frame rate.

[volt]

Originally Posted by Doomtrooper

Lod bias is for graph wars...you would be surprised at what LOD bias adjustments do for frame rate.

certainly lesser hit than your 8x aniso

[Fred da Roza]

Originally Posted by Mintmaster

Anyone else care to make a contribution?

http://www.beyond3d.com/forum/viewto...ghlight=#13564

Originally Posted by Chalnoth

Yes, there are other reasons to consider an 8500 over a GeForce3, but none of those reasons ever held for me.

24 bit FP isn't enough.

[gokickrocks]

i think the 2nd molex is for powering the fan itself, instead of having it drain power from the card

[Chalnoth]

Originally Posted by Tahir

P.S. In all seriousness your proposal is ridiculous, as the NV35 was what then? An attempt to fix the NV30 that was meant to be an NV40 really?

No, not at all. I don't think the "original" NV30 was to be anything like the NV40. I merely stated that the "original" NV30 will be more of a basis for the NV40 than the one that released was.

Anyway, I think the NV35 was hit with a similar crunch, but it wasn't quite as bad as the NV30's, and so they were able to get it to approach the original design much better (here I reference the fact that the NV35 doubled FP performance, without significantly impacting integer performance, at a pretty tiny cost in transistors).

Why do I now think that the NV30 that we saw was a mistake? Well, there are a number of aspects of the architcture that just seem silly. For example, have you noticed how it's only the floating point registers that are hit with a performance hit? The integer registers are just fine. Why were the integer registers not simply increased in size to accomodate floating point values? The more I learn about the NV30, and the more I think about it, the more it really seems like a product that was thrown together at the last minute, with decisions made merely in the hope of getting a product, any product, out the door.