A Semi-accurate Look into Fermi

[trinibwoy]

I see. But what about bump mapping? Can't something be coded that would use the tessellation engine to convert all such bump maps into actual polygons?

Nope, bump mapping is just a pixel shader trick that changes the color of a pixel to simulate detail. Tessellation happens much earlier in the pipeline and affects the underlying geometry of the scene.

[Manicdan]

Then you have wrong info. A2 was already above 500mhz, A1 was supposedly just shy of said mhz though.

Edit- Hmmm... with the info that has come to light from a different forum, it seems metal respins don't drastically help clocks.

so A1 was a little under 500
A2 was over (but probably not 600 or they would have said so)
so A3 can probably be assumed at 650ish?

[NickF]

so A1 was a little under 500
A2 was over (but probably not 600 or they would have said so)
so A3 can probably be assumed at 650ish?

Their target was 750. We're looking at them missing that mark by a good margin, so 650ish sounds like good speculation to me

[XS2K]

@skugpezz
So the overscanning is great in your opinion?I think that Eyefinity has quite a few problems but it's not the tech it's the developers and the support from ATi's software part.Anyhow I think 3D Vision is better after experiencing it and also has a larger install base in games all we need now are more 120hz displays.

[Manicdan]

@skugpezz
So the overscanning is great in your opinion?I think that Eyefinity has quite a few problems but it's not the tech it's the developers and the support from ATi's software part.Anyhow I think 3D Vision is better after experiencing it and also has a larger install base in games all we need now are more 120hz displays.

id rather see someone make glasses that know where you are, and let you look around objects as you try to move your head. it dosnt have to be by much at all, enough just so you know its happening, and you should really feel like your in the game then.

[skugpezz]

@skugpezz
So the overscanning is great in your opinion?I think that Eyefinity has quite a few problems but it's not the tech it's the developers and the support from ATi's software part.Anyhow I think 3D Vision is better after experiencing it and also has a larger install base in games all we need now are more 120hz displays.

Yes eyefinity is awesome IMO. It is in its early stages too. I wont make decision on a card because of it though. it is doing what previous video cards cant, unlike physx......

[annihilat0r]

Nope, bump mapping is just a pixel shader trick that changes the color of a pixel to simulate detail. Tessellation happens much earlier in the pipeline and affects the underlying geometry of the scene.

Well actually I know that, but don't bump maps contain the depth information (that will be "fakely" applied to the object), and using the already developed bump maps can't this information be somehow converted to actual polygon info?

I'm not talking about doing this real-time, either. The bump maps could be taken and pre-converted into whatever, and then be used in game. Obviously I don't know how to actually do this or anything, but this line of thinking leads me to believe that this should be able to be done.

With parallax maps it could be done even better, if it's really a possibility

[JohnZS]

Did you read the article? Hotlots should be back at the beginning of the month. No Janurary paper launch, unless you call showing/demoing GF100 at an event a paper launch. The paper launch will come in Feb/March and no real availability, if you thought HD58x0 availability was bad just wait, until Q2.

Edit- Hmmm... with the info that has come to light from a different forum, it seems metal respins don't drastically help clocks.

That was what I read in the article, however I have the opinion that Charlie can exaggerate and perhaps use a bit of an artistic licence when it comes to reporting. I am still hopeful for Widespread availability in March.

I thought the Radeon 58xx launch here in the UK was awful, units have only just started shipping here in very limited quantities this past week, are you going to say that Fermi would be worse? and that we would need to wait until MAY?

Perhaps the respins are not only for hopeful increases in clockspeed but for bug fixes too?

John

[trinibwoy]

Well actually I know that, but don't bump maps contain the depth information (that will be "fakely" applied to the object), and using the already developed bump maps can't this information be somehow converted to actual polygon info?

I'm not talking about doing this real-time, either. The bump maps could be taken and pre-converted into whatever, and then be used in game. Obviously I don't know how to actually do this or anything, but this line of thinking leads me to believe that this should be able to be done.

With parallax maps it could be done even better, if it's really a possibility

Normal maps don't contain depth, they store normals used in lighting calculations. The normal determines how light interacts with the object so faking the normal lets you simulate a different interaction per-pixel - i.e make you think light is bouncing off of the object in a certain way that gives the impression of additional geometric detail that really isn't there. But this all happens in 2D. There's no way to take the data from a normal map and simulate 3-dimensional polygon information. It would be like asking the GPU to take a jpeg and convert it into a 3 dimensional model of the scene that it was based on. Parallax mapping is no different, it's just fancier normal mapping that takes self shadowing into account.

Going from polygons -> more polygons -> pixels is easy. Doesn't work in reverse though

[josiahsuarez]

my computer desk isn't big enough for 3 monitors

[JohnZS]

my computer desk isn't big enough for 3 monitors

I don't know what that has to do with Fermi, unless you are implying it is going to be a huuuuuuuuuuuge card and you will need 3 monitors to see it @ 100% scale on a JPEG image.

John

[SKYMTL]

Regarding Fermi now, from the start I have personally said that given presumed yields, diesize, scalability and production cost, fermi must deliver 40%+ higher than evergreen performance in games to have a good run. From the looks of it, it might just get that (and more). Thats good for all consumers, since both companies will have almost equal perf/$ ratios, and the resulting price competition will benefit all.

Lets hope that the greens match that goal, it would be a very interesting 6months ahead.

I'd also like to see nvidia try to address the actual moneymakers (ie the mainstream and lower performance market segments).

Let's break this down because you make some good points. First of all, not much is publicly known about how well the Fermi architecture will scale when it comes to anything but the extreme high end. That is to say that it may take a lot of doing for NVIDIA to adapt it to lower end, more mainstream cards. NVIDIA's past issues with scalability is one of the many reasons why we are only now seeing lower-end 200 series cards. ATI on the other hand is an expert on scalability as is evidenced by the quick succession of 5000-series releases. Only time will tell what happens when the Fermi architecture gets scaled down but one thing is clear: ATI has shown that NVIDIA can't continue to use G80 derivatives if they hope to compete in the sub-$200 price category.

from info I got it still cant make 500mhz .

False. And no, I won't elaborate.

my computer desk isn't big enough for 3 monitors

3x 17"

[Helloworld_98]

3x 17"

I would agree but they're quite hard to find now

[Clairvoyant129]

if it makes you sleep better at night LOL. an nvidia card cant even do half that res with 10fps . All I see in the links are smooth gameplay and high rez 7680 x 1600 lets see your gtx 280 do half that or even 19X12 smoothly

I'm not going to take sides in some fanboy bickering but you do keep posting negative and false info on NV threads.

Btw, your 5770 ain't all that special, slower than 4870 512MB and GTX 260 Core 192... bandwidth starved with its tiny 128bit bus.

[skugpezz]

I'm not going to take sides in some fanboy bickering but you do keep posting negative and false info on NV threads.

Btw, your 5770 ain't all that special, slower than 4870 512MB and GTX 260 Core 192.

yea my 5770 is my backup card, I had to send the 5850 back . I am happy with it while I await my new 5850 . Your 5870 isnt special either, a 5970 is faster

[003]

There's one quick easy way to solve this NDA bullcrap, take a photo of your NDA agreement (with appropriate parts blacked out) and post it. Don't do it & you're just blowing hot air and probably deserve at least a temporary ban if not a permanent one. If you don't prove it, you'll both be labelled as liars, shills, [insert insulting term here] etc. I think the saying goes a little something like this:


Put up or shut up!

I've already stated I am not the one under NDA. I know someone who is and they have technically broken NDA by talking to me. I am not going to risk getting them into some very hot legal water. And that is a terrible idea. I can't say for sure but I'd wager Nvidia doesn't sign too many NDAs on Fermi with private individuals. It would be very easy to create a short list of possibilities of people that broke NDA with a photo like that.

[Clairvoyant129]

yea my 5770 is my backup card, I had to send the 5850 back . I am happy with it while I await my new 5850 . Your 5870 isnt special either, a 5970 is faster

Whatever makes you sleep better at night.

[skugpezz]

Whatever makes you sleep better at night.

lol good one dude

[LordEC911]

so A1 was a little under 500
A2 was over (but probably not 600 or they would have said so)
so A3 can probably be assumed at 650ish?

I would say 650 would be near the upper end of the possible clockspeed. I have been saying we would see G200-esque clocks for quite awhile now and no one wanted to believe me, though they were of the green tinted variants, just like how I have been saying that GF100 is very similar in size to the 65nm G200.

Their target was 750. We're looking at them missing that mark by a good margin, so 650ish sounds like good speculation to me

Yeah. Say good-bye to the estimated 40% advantage over Cypress.

That was what I read in the article, however I have the opinion that Charlie can exaggerate and perhaps use a bit of an artistic licence when it comes to reporting. I am still hopeful for Widespread availability in March.

I thought the Radeon 58xx launch here in the UK was awful, units have only just started shipping here in very limited quantities this past week, are you going to say that Fermi would be worse? and that we would need to wait until MAY?

Perhaps the respins are not only for hopeful increases in clockspeed but for bug fixes too?

John

He doesn't often exaggerate with the facts, when you can find them in his articles.

The UK has been getting a constant supply of cards, while that supply wasn't exactly huge, there was a supply. You seem to be implying no cards in the UK, which is false.

Yes Fermi availability will be worse, depending when Nvidia gives TSMC the greenlight on full production and if the yield numbers I have heard are still accurate for A3. I doubt TSMC will be running full tilt to ramp over the Lunar New Year and that Nvidia will have all the capacity they need to get all the boards ready in around 2-3 months for a real launch.

Let's break this down because you make some good points. First of all, not much is publicly known about how well the Fermi architecture will scale when it comes to anything but the extreme high end. That is to say that it may take a lot of doing for NVIDIA to adapt it to lower end, more mainstream cards. NVIDIA's past issues with scalability is one of the many reasons why we are only now seeing lower-end 200 series cards. ATI on the other hand is an expert on scalability as is evidenced by the quick succession of 5000-series releases. Only time will tell what happens when the Fermi architecture gets scaled down but one thing is clear: ATI has shown that NVIDIA can't continue to use G80 derivatives if they hope to compete in the sub-$200 price category.

Yep. With still absolutely no whispers of any of the GF100 derivatives taping out and the end of the year coming, the window on a 1H launch for any of them is getting smaller and smaller, unless Nvidia pulls a miracle and launches on A1 silicon.

[Clairvoyant129]

Like most have mentioned here, I don't think Fermi will have a 40% advantage over comparable Evergreen and I doubt NV will be able to hit their intended clock speed.

[Teemax]

I've already stated I am not the one under NDA. I know someone who is and they have technically broken NDA by talking to me. I am not going to risk getting them into some very hot legal water. And that is a terrible idea. I can't say for sure but I'd wager Nvidia doesn't sign too many NDAs on Fermi with private individuals. It would be very easy to create a short list of possibilities of people that broke NDA with a photo like that.

So it's a friend of a friend or something like that, right?

Actually, me too, a friend of my second-best friend's cousin told me that HD 6870 will release exactly 2 month's after Fermi's availability, and it will completely decimate Fermi.

But I can't back it up. You know, NDA and stuff...

[Movieman]

I can see this thread going right into the sewer.
gentlemen: Please present your arguements without the personal attacks.
Also when making claims, back it up or it's just hot air..
Thanks for reading.

[Florinmocanu]

I see. But what about bump mapping? Can't something be coded that would use the tessellation engine to convert all such bump maps into actual polygons?

you don't need that. you can do displacement through tesselation with the bump maps.

I work as a 3d artist, usually displacement deals with big details and bump mapping deals with the really tiny details which don't need extra geometry to be believable.

[Movieman]

you don't need that. you can do displacement through tesselation with the bump maps.

I work as a 3d artist, usually displacement deals with big details and bump mapping deals with the really tiny details which don't need extra geometry to be believable.

OT: Love the avatar!

[flippin_waffles]

Judging by your post history I'm convinced that you are Charlie.

Hmm, I don't know Charlie, but he probably wouldn't exactly be thrilled to see this statement.

Anyway, if fermi doesn't have something similar to Eyefinity for gamers, in my opinion it's not something i'd even consider a worthwhile upgrade regardless of how 'fast' it may or may not be. Frame rates add nothing to an experience, at least for myself.