nVidia NV20 info...my gawd!! ;)

[Shazam]

Yes, but does it make coffee?

 

[Warrenton]

First off in regards to DaveB3D...

You apparently have inside sources, so why are you telling us? Isn't that violation of NDA? Maybe youre the one pulling stuff out your rectum?

 

[Varborta]

Just wondering do those HIGH premium card will drive MX price down?
And, does the NEW card gonna make the visual better?

 

[Sephiroth_IX]

<< You apparently have inside sources, so why are you telling us? Isn't that violation of NDA? Maybe youre the one pulling stuff out your rectum? >>

Thank you, Warrenton, for ruining our only halfway inside source. shut up.

 

[BFG10K]

I said this was going to happen and I was right. nVidia is combining their monster fillrate with a HSR scheme. This card is going to absolutely demolish everything.

The Kyro and Radeon boards have HSR schemes but with mediocre fillrates thay can't exploit the advantage they have given themselves. The NV20 will confirm for the 5th time in a row that nVidia are the speed king of the PC 3D graphics industry.

 

[DaveB3D]

I'm telling you because I don't think people should be told crap and then come to find out something totally different.

I'm not violating anything. If I were, I wouldn't be telling you.

 

[DaveB3D]

Well future boards will see a definate quality improvement over anything we currently have. Pixel and vertex shaders are going to allow for some simply mind blowing effects. Add to that true multi-sampling AA, anisotropic filtering and very power T&amp;L units and things are going to be a considerably step up. I'm really looking forward to seeing what Carmack does. I can tell he has some good things in the works and I think the new hardware is going to make it look amazing. It will be nothing like we have right now...

 

[Auric]

Folks who are whinging about the prices should understand something: you are not obligated to buy it. Generally, you will not be considered &quot;inadequate&quot; and have &quot;sand kicked in your face&quot; if you don't have the most expensive gaming viddy card available. Do you lament that [insert pricey automobile] exists just because you cannot afford it? If you do there is something wrong with you. Check your head. I think it is fantastic that there is a wide range of choices. It's even better if it drives competition and lowers the price of parts with higher performance, features and quality.

Right now I cannot justify spending more than $150 on a viddy card because none of them are particularly great, partly because the full power of the high-end ones does not improve the experience enough for the extra $$$. The software has to catch up and the industry is held back by that mutual dependency of one always waiting on the other to advance.

Remember peeps, the goal is realism. We are not anywhere near that. We are still in blocky cartoonland. The viddy chip will be the most important part of the drive towards realism for the forseeable future. If you cannot justify $500 to play some cheesy game like Quake 3 faster that is understandable, but how many of you would drop $1500 for a viddy card if it offered a pseudo-real visual experience? Prolly a lot I'm a thinkin'. Prolly a lot of people would drop $5000 to really be immersed in a game.

Of course, that will require advancement of displays, controls, voice recognition, bandwidth, and so on too. Bottom line: if you like games don't complain about better stuff just cause it's out of your range. It's a free market and self regulating price-wise. Okay, so that wasn't the bottom line.

If they build it and people do not come, well then they will lower the price where possible. I'm sure they have lots of brainy actuaries with charts and graphs and slide rules calculating what the market will support.

Blah blah blah

 

[Tulkas]

NV20 specs look very interesting.

The price is what i was expecting. As for the people who will be purchasing this card i think that for those like myself who didnt upgarde after TNT2 or Voodoo3 it isnt all that bad.

I was saving my money for as card like this... if the card lives up to its expectations.

Look at all cards between the TNT2 and the NV20. I could have upgraded twice in that time and seen a good performance boost.

But instead i resisted the dark force of UPGRADING

I think that this Xmas is gonna make me very happy! Whatever card i choose

 

[ElFenix]

5 times in a row! nvidia has hardly gotten 5 products out the door! there was the riva, the zx, the twintexel, the tnt2, the gf, and the gf2. thats six. and everything before the tnt2 was slower than the competition. the tnt2 wasn't even as fast as a v3 when it came out, had to wait for the tnt2u variant, and then that got beat by the g400max. so then the gf comes out, is still slower than the max at 32bit. the ddr variant is really what blew everything out of the water. fine, they were undisputed king then. a couple months pass, the gf2 comes out its really damn fast, but a short while later the radeon comes out and is faster in 32bit (not to mention that the voodoo has better fsaa quality)! so we get some driver hacks and later a gf2u variant so that nvidia can assert some dominance. so out of all 6 of nvidia's chips only one variant was undsiputed king of speed until the next nvidia release.

 

[Wingznut]

&quot;stumbled into information they obtained about the NV20 in a Chaintech Beach party where they spoke with Roy Taylor from NVIDIA&quot;

Sounds reliable to me.

 

[Soccerman]

&quot;NV20 is not a tile based architecture. NVIDIA is not using a tiler and they have no plans on using one any time soon. Rather it is a traditional architecture with occlusion detection. It has been said that in a case of 10x overdraw you'll save like 4x of it, or basically 40%. This is somewhat optimistic though.

I must wonder what you think Tile Based rendering is then? please inform me of what I've been told multiple times (in fact, I think I have a paper (PDF) on it somewhere on this computer..).

sure, there could be a few differences here and there, but basically from what nVidia has stated, they take off what is NOT visible, so that it does not get rendered, and therefor does NOT waste A) Bandwidth, and B) Fillrate.

if you are simply saying that it isn't the same, because they call it a different name, then I'm sorry to say, but you have yet to learn much abuot how this industry works!

 

[Mday]

how much RAM... ???

yes, it's still an issue. ;-)

/me just fell down from reading that...

 

[Soccerman]

reading what?

&quot;Add to that true multi-sampling AA,&quot;

uhh, 3dfx AND nVidia, AND ATi already have multi sample FSAA. in other words, there is more then ONE sample per pixel. nVidia's method of doing that is by rendering everything in a high res, and then scaling down to the res that yuo run at.

3dfx's method, is somewhat different. it actually uses each chip to render the exact same frame multiple times. it then puts all these frames and, a crude, yet explanetory term, Jitters these images less then a pixel around, so that you get in effect a very good FSAA (some say it's slightly better then nVidia's FSAA)

 

[DaveB3D]

Well tile based is to general of a term. All 3D accelerators tile to some extent (be it just memory tiling, they tile). A more accurate term would be deferred rendering. To sum it up, you basically upload your triangle data, bin it, and then procced to render the scene one tile at a time. From there, you check for pixel visibility (dropping everything you don't see), render your texturing passes, AA ect and then once you are done you output that tile to vid memory and do the next tile. That is a very rough explaination. Deferred rendering is a very specific technology. I can explain specifics if anybody cares to know.

With NVIDIA's occlusion detection, they are not getting 100% overdraw removal as with a tiler. They are likely to get 40% at best and it is per-triangle (perhaps per-primitive, but I doubt it). If things were sorted, you could get improved results if you could render back to front.

I have a lot to learn about how the industry works eh? lol. Perhaps you should just learn to tell different technologies apart.

 

[Soccerman]

heh well at least you responded.. I'm too tired right now to answer, so I'll answer in the morning..

 

[DaveB3D]

ah geez.. No they don't.

With multi-sampling you need a single texel fetch, with super-sampling it depends on your sample rate. So n sample = n texel fetches.

And yes, I know what 3dfx and NVIDIA do. I did write the 3dfx FSAA whitepaper and the article, NVIDIA's FSAA Investigated (http://www.beyond3d.com/articles/NVIDIA_FSAA/)

 

[Sunner]

Soccerman, I dont mean to imply you dont know what you're saying, but Dave is right, I'd trust him on this one either way, the guys over at B3D knows quite a bit about this stuff you know.

 

[Charles]

Wow! I would be happy if it will be able to do 1024x768x32 2xFSAA.

It's just a bit expensive.

What will that call it? GeForce3?

 

[DaveB3D]

Is that all you want? At least use 4x AA.

 

[Whizzy]

But what about 64bit rendering ?? Carmack once said that he's gonna need it !! especially for smoke/fog/transparency effects...

Drool... DOOM III in 64bit color...
Is it supported by DX8 ?

 

[DaveB3D]

Well, IMO, if we use 32-bit color, error difusion dithering and then add a post filtering on top of that, we should be good. With deferred rendering this is especially true.

ah... 4 am.. goodnight.

 

[DaveB3D]

Before I go to bed, I should correct myself so I don't look like an ass. As far as I know, you have to use deferred rendering to use error diffusion because your entire scene needs to be rendered before you start your dither. The only only place I could see it useful with a traditional architecture is if you super-sample a 16-bit image and on the downsample you need to do a dither pass. It would work there as well, though I doubt anyone would take the time to put it in hardware for that purpose.

now goodnight... I think...

 

[jpprod]

Soccerman; can someone like jpprod tell me how much of a hit shading etc, of DX8 gives?

On GeForce the shading is just a combination of DOT3 operations, which each take one extra pass/texturing unit. Example of a shaded surface would be base texture + DOT3 diffuse lighting + DOT3 specular lighting (this could be a scratched shiny metal for example); obviously this requires 300% as much texturing fillrate as the same surface would take if it would be just single-textured.

With DirectX8 things get more complicated. As far as I understand, DX8 only defines an interface for the shaders, it does not specify the hardware implementation. Different hardware may have different fillrate penalty associated with usage of pixel shaders, but baseline performance hit per shading operation is the same what one additional texture would introduce.

One more observation to this matter, having an advenced HSR is an essential feature for any new accelerator claiming full DX8 feature set support. Imagine how costly it is going to be to perform overdraw in a next-generation DX8 game with all these bells&amp;whistles eating fillrate...

 

[DaveB3D]

One thing to keep in mind as well is that fill-rate numbers are going to become less important. It won't matter how much fill-rate you have is your pipeline isn't effecient. So company X could have a really low fill-rate and be much more efficient than company Y. Now in talking about effiency I'm not talking about the core architecture (so tradition, deferred, etc). Rather the effiency of combiners. That will be a big key to things. The next product cycle is going to be increasingly difficult for reviews as well, because if they don't understand this they aren't going to be able to accurately review a product.


There are other issues involved, but those can be addressed at a later date (translation: I don't feel like it right now. ).