Nvidia Fermi GF100 working desktop card spotted on Facebook

[gumballguy]

lol love your sarcasm .

But i think we are all missing the point here, THAT IS ONE AWESOME LOOKING DRAGON on that screen

I know thats a joke

BUT

Have you looked at the stupid With/without tesellation screenshots of that dragon?

"Without" they have INTENTIONALLY left out enough poly's to not produce the spikes. If a game developer wanted those spikes there, without tesellation, they would make them appear by adding sufficient polys to do so.

IMHO its not a fair comparison whatsoever. Its virtually like its a different model.

Not saying tesellation sucks, it has its uses, but that dragon is an outright lie.

[Vivi]

I disagree, just look at NVIDIA's marketing in the past. When they had a winner, it was always well anounced and there was hardly any ambiguity as they were launched on time. Its the samething with intel vs amd, the company with the better product is always sure to launch first.

Just think about how fast the 4870 and 5870 came out compared to the r600, albeit supply was an issue at launch, but there were no major delays. If anything, the 5870 came out earlier than I expected, I thought they would have milked the 4870x2 a bit longer.

I do agree, the debate is senseless over whether that is a real card because more than likely NV will never reveal the truth unless it is advantageous to them (which I don't see happening). However there is also no real discussion needed on whether fermi will be game changing, simply put, if it didn't have problems, it would have been out already (there's no way you can deny that logic, because if its not true, then NV's product must either suck or their management is retarded).

i see your logic, but the 295 was released way after 4870x2 and defeats (kind of) it. (in the overclocking world). You see, my money is on that they might be worse than amd air, but as soon as you ln2 the 40nm process it might reach something silly like 1400 gpu core. Thats what im hoping for

[Solus Corvus]

I know thats a joke

BUT

Have you looked at the stupid With/without tesellation screenshots of that dragon?

"Without" they have INTENTIONALLY left out enough poly's to not produce the spikes. If a game developer wanted those spikes there, without tesellation, they would make them appear by adding sufficient polys to do so.

IMHO its not a fair comparison whatsoever. Its virtually like its a different model.

Not saying tesellation sucks, it has its uses, but that dragon is an outright lie.

That misses the point of tessellation, IMO. If they really wanted to they could have made the basic dragon model have the same number of polygons as the tessellated version, or even more. That's something we could do a long time ago. The point is to show off how tessellation allows scaling from the lowend to the highend with a single set of content.

[jaredpace]

looks like its on the same unidentified PCB as the "mock-up" prototype. (Just not sawed in half heheheh)

[Manicdan]

That misses the point of tessellation, IMO. If they really wanted to they could have made the basic dragon model have the same number of polygons as the tessellated version, or even more. That's something we could do a long time ago. The point is to show off how tessellation allows scaling from the lowend to the highend with a single set of content.

for some things, maybe the spikes, without the right polygons, the textures wont line up right. its one thing to show off 2 models one with and without uber amounts of polygons, its more important to do it right when you actually use it in a real game/benchmark with textures and lighting

for the small things, bump mapping did ok, for larger items, you kinda have to put in enough minimum polys for the textures to look right. tessellation isnt going to be a quick magic wand that makes them all go together perfectly. otherwise the result will be without tessellation, games will be as piss poor at B:AA is without physx (being that they could have easily done something, not just remove everything)

[Solus Corvus]

Isn't that basically what I said? Scaling between low detail and high detail with one set of content.

[AliG]

i see your logic, but the 295 was released way after 4870x2 and defeats (kind of) it. (in the overclocking world). You see, my money is on that they might be worse than amd air, but as soon as you ln2 the 40nm process it might reach something silly like 1400 gpu core. Thats what im hoping for

Performance may be slightly higher, but not enough to justify the price increase imo. Its value comes at the extreme end as you pointed out, which supports that it had far less sales than the 4870x2 (simply because it came out so late). Sales are all that matters in the end, so perhaps NV has something at a similar performance level and is spending a lot of time with drivers? I think the 5870 surprised all of us by doubling the 4870's shader count, including NV and that's why we're seeing all the delays.

[annihilat0r]

5970 is a launch and of course Nvidia is going to do whatever it can to steal some of their competitor's publicity. What's with you people?

[Solus Corvus]

5970 is a launch and of course Nvidia is going to do whatever it can to steal some of their competitor's publicity. What's with you people?

The 5970 is so awesome we don't even need to talk about it.

[Blacky]

The 5970 is so awesome we don't even need to talk about it.

I don't know why they even launched such card at this time, when they can barely supply their 5870 & 5850 line, looks like is going to be a ghost launch to me

[tajoh111]

This is has got to be the worst counter promotion I have ever seen, whether the card is real or not doesn't matter. Its that they don't have anything gamewise to demo.

Just a damn picture of the card.

[orangekiwii]

I'm strongly considering getting a 5970...

I want Fermi... but without ANY idea of how it will perform relative to 5970... whats the point of waiting?

Best case scenario - its a little faster than 5970

most likely - a little slower...

worst... barely faster than 5870


I want to see something that will hold me off from buying a 5970... and right now nvidia isn't doing anything... add that to the fact I need a new card soon as mines in the process of dying D:

[Blacky]

^ probably less $$ on fermi with "similar" performance, could be wrong on the $$ part though

[Sam_oslo]

A real counter needs to show 3D-scores and FPS for at least a few games. Looking at the cover of the card (even with GPU inside ) is not good enough .

It is so sad that it becomes funny to browse the test-sites, and see the review of those "super-duper GT 240 with custom cooling and stuff", on the same page as 5970.

[George_o/c]

I thought the same for an instance I hope they don't face several shortage issues like ATi, if it does the prices on e-tailers is going to be overrated...

TSMC low 40nm yields not only have on impact on ATI, but also on NVIDIA as well. Have you forgotten already ? http://www.digitimes.com/news/a20091030PD209.html

Yeah, it doesn't look half as good when rendered on a 5870!

I know thats a joke

BUT

Have you looked at the stupid With/without tesellation screenshots of that dragon?

"Without" they have INTENTIONALLY left out enough poly's to not produce the spikes. If a game developer wanted those spikes there, without tesellation, they would make them appear by adding sufficient polys to do so.

IMHO its not a fair comparison whatsoever. Its virtually like its a different model.

Not saying tesellation sucks, it has its uses, but that dragon is an outright lie.

Have you seen Tesselation with/without pictures from confidential AMD PDF ? The difference is obvious there as far as I can tell, but I'm not supposed to show it... You can't just put more polygons there dude.

I don't know why they even launched such card at this time, when they can barely supply their 5870 & 5850 line, looks like is going to be a ghost launch to me

Is that supposed to be a justification for NVIDIA not having a card outta there at the time being ?

^ probably less $$ on fermi with "similar" performance, could be wrong on the $$ part though

Performance will be better. But do not expect these cards to arrive at less than $600... NVIDIA has shown many times at the past, that they are generous when it comes to high-end VGAs.

[Blacky]

TSMC low 40nm yields not only have on impact on ATI, but also on NVIDIA as well. Have you forgotten already ? http://www.digitimes.com/news/a20091030PD209.html

^ I'm aware of that but things can get better when they finally reach the mass production stage.... is not their fault, its TSMC...

Is that supposed to be a justification for NVIDIA not having a card outta there at the time being ?

Nop just doesn't make sense to launch a card that uses an additional chip when they have shortage on their "single" other products...

Performance will be better. But do not expect these cards to arrive at less than $600... NVIDIA has shown many times at the past, that they are generous when it comes to high-end VGAs.

I hope not (on the price part) but knowing nvidia that could happen (like GT200 did at start)

[gumballguy]

for some things, maybe the spikes, without the right polygons, the textures wont line up right. its one thing to show off 2 models one with and without uber amounts of polygons, its more important to do it right when you actually use it in a real game/benchmark with textures and lighting

for the small things, bump mapping did ok, for larger items, you kinda have to put in enough minimum polys for the textures to look right. tessellation isnt going to be a quick magic wand that makes them all go together perfectly. otherwise the result will be without tessellation, games will be as piss poor at B:AA is without physx (being that they could have easily done something, not just remove everything)

+1

Isn't that basically what I said? Scaling between low detail and high detail with one set of content.

I didnt take manicdan's comment to be what you said. Thats probably just interpretation. Anyway, semantics aren't important. In a nutshell:

I was delineating between adding polys to scale, and polys that should be there to start with.

The spikes should be there to start with. And werent. They are required to make the dragon look like it is supposed to and would have been included by any rational game developer. Imho this breaks the comparison between with/without tesselation for the dragon and makes me suspicious about what they've done.

The roof should be done with tesselation. And was. This is perfectly fine for comparison.

Have you seen Tesselation with/without pictures from confidential AMD PDF ? The difference is obvious there as far as I can tell, but I'm not supposed to show it... You can't just put more polygons there dude.

No, I don't have access to this. But again, and as in manic dan's post, tessellation cannot logically be the answer to low poly counts. Where you know you need them, you need to put them to start with. For anywhere where it isn't necessary to make it look right, use tesselation, provided it makes it look better (ie: turn it on for ancillary objects like roofs if no undesirable warping/texture effects occur).

There was some interesting weirdness happening with blockwork in a building too (not the dragon scene) where thickness was added to specific blocks on the edges of a building & near the windows. This screwed up the textures royally by stretching them. As an aside, blockwork wasn't extruded to that degree in real life, so it looked outright wrong. But again I suspect that it went past extrapolation into being a different polygon model (but reused textures). Why? If there's no information on varying block thickness to start with, you can't just imagine it and produce the very specific effect that they did. I can acknowledge there may actually have been some depth info to do this with, but crikey, they would have had to scale slight differences in object depths * 3000% to make it look that way - which is not a serviceable example of how tesselation should apply. A generic algorithm can't do 3000% scaling everywhere... it'll make an unintelligible mess.

The pictures I were looking at when I noticed these were on [H], but you can find them anywhere if you wanna look..

I have little respect for Unigine and what I believe is a mockery of a bench. This is because I believe they are using two completely different models for with/without tesselation. You don't have to agree with me there. In the very least they are being very selective with the way they construct their models so that it exaggerates the difference in an unrealisitc, unreasonable fashion.


I'm sure people will hate my entire post for the extremity of suggesting they were two different models. Be objective:
1) Look at the pics I'm talking about.
2) Even if you dont believe most of my post, you can't deny that textures need to consider depth.

Geez this post started out nice and to the point. Now I've really overdone it

[blob]

looks like its on the same unidentified PCB as the "mock-up" prototype. (Just not sawed in half heheheh)

Compare the PCB on the vizworld picture with the mockup PCB
The fermi based card shown on SC09 isn't a tesla but a Geforce.
(The tesla card is laying in front of the system)

[Sgt.McRuff]

Nop just doesn't make sense to launch a card that uses an additional chip when they have shortage on their "single" other products...

The 5970 as supposedly binned for lower voltage, so I believe then been planning to bin like this before 5870 came out.

[FischOderAal]

[IMG]http://tweakers.net/ext/f/YSErZieixGmDpRVti4W2mxvX/full.jpg[./IMG]
Compare the PCB on the vizworld picture with the mockup PCB
The fermi based card shown on SC09 isn't a tesla but a Geforce.
(The tesla card is laying in front of the system)

Ha! Nice find!

[ajaidev]

[]http://tweakers.net/ext/f/YSErZieixGmDpRVti4W2mxvX/full.jpg[/]
Compare the PCB on the vizworld picture with the mockup PCB
The fermi based card shown on SC09 isn't a tesla but a Geforce.
(The tesla card is laying in front of the system)

gr8 find Geforce hoo

Jan is here, i hope so are you GTX 380

[mrcape]

Fermi and the 5970 both seem tasty, I want both of course. The 5970 price point at launch is $100 too high. I bet it will encourage a lot of people to wait and see what fermi will cost and what it can do.

[jaredpace]

Compare the PCB on the vizworld picture with the mockup PCB
The fermi based card shown on SC09 isn't a tesla but a Geforce.
(The tesla card is laying in front of the system)

thanks for the picture, blob. yah the card at GTC is teh same as the one at SC09, just its working this time

[Florinmocanu]

+1



I didnt take manicdan's comment to be what you said. Thats probably just interpretation. Anyway, semantics aren't important. In a nutshell:

I was delineating between adding polys to scale, and polys that should be there to start with.

The spikes should be there to start with. And werent. They are required to make the dragon look like it is supposed to and would have been included by any rational game developer. Imho this breaks the comparison between with/without tesselation for the dragon and makes me suspicious about what they've done.

The roof should be done with tesselation. And was. This is perfectly fine for comparison.



No, I don't have access to this. But again, and as in manic dan's post, tessellation cannot logically be the answer to low poly counts. Where you know you need them, you need to put them to start with. For anywhere where it isn't necessary to make it look right, use tesselation, provided it makes it look better (ie: turn it on for ancillary objects like roofs if no undesirable warping/texture effects occur).

There was some interesting weirdness happening with blockwork in a building too (not the dragon scene) where thickness was added to specific blocks on the edges of a building & near the windows. This screwed up the textures royally by stretching them. As an aside, blockwork wasn't extruded to that degree in real life, so it looked outright wrong. But again I suspect that it went past extrapolation into being a different polygon model (but reused textures). Why? If there's no information on varying block thickness to start with, you can't just imagine it and produce the very specific effect that they did. I can acknowledge there may actually have been some depth info to do this with, but crikey, they would have had to scale slight differences in object depths * 3000% to make it look that way - which is not a serviceable example of how tesselation should apply. A generic algorithm can't do 3000% scaling everywhere... it'll make an unintelligible mess.

The pictures I were looking at when I noticed these were on [H], but you can find them anywhere if you wanna look..

I have little respect for Unigine and what I believe is a mockery of a bench. This is because I believe they are using two completely different models for with/without tesselation. You don't have to agree with me there. In the very least they are being very selective with the way they construct their models so that it exaggerates the difference in an unrealisitc, unreasonable fashion.


I'm sure people will hate my entire post for the extremity of suggesting they were two different models. Be objective:
1) Look at the pics I'm talking about.
2) Even if you dont believe most of my post, you can't deny that textures need to consider depth.

Geez this post started out nice and to the point. Now I've really overdone it

They didn't use another model lol, they used displacement through tesselation.

How is displacement done? You get a flat plane, you use a black and white texture (which you don't see in-game, it's just used by the engine to do the displacement) to show the high (white) and low (black) parts of the wall and through that the wall gets to look realistic, even if in reality it's just a flat plane.

That's what the purpose of tesselation is, to create new geometry through the gpu, so that you don't raise the base poly count that much.

The tesselation in DX 11 is no the Trueform engine in the Radeon Cards, this is something a bit more complex, it does smoothing and displacement at the same time.

[fiveprime]

I'm really enjoying this thread, I don't think satellite intel analysts could have extracted so much "information".