Even more NV20 specs

[NFS4]

That other NV20 thread is getting quite large. Here's even more details on this Uber chip:


<< NV20 announcement is very likely to take place during Comdex/Fall, which will be arranged in mid November this year. That?s why the first graphics cards should supposedly start selling in January-February already.

As far as the official name of the upcoming NV20 is concerned, we hear that NVIDIA allegedly decided to retain a well-known GeForce brand for its next offspring and to call it GeForce III. And now a few words about 4800Mpixels/sec fillrate. Most of you were pretty skeptical about this number, however, we got more info proving that it is absolutely correct. NV20 will support 300MHz working frequency and will process 4 pixels per clock, so that the fillrate will make 1200Mpixels per second. However, NV20 will feature HSR (Hidden Surface Removal) technology, which implies that the invisible surfaces won?t be textured at all and each pixel will be drawn just once. As a result, if we suppose that the average overdraw coefficient equals 4, we will get 4800Mpixels/sec. As for the T&amp;L unit, its performance is expected to be twice the performance of this unit in GeForce2 GTS. >>



Even more NV20 specs

 

[Jonny]

Wow, that cards starting to look impressive. Any idea how much it will go for?

 

[NFS4]

<< Any idea how much it will go for? >>


Too much

 

[Doctorweir]

HSR sounds much to me like a technology similar to Kyro's Tile Based Rendering...

 

[LXi]

My speculation was correct, I saw the rising trend of the video prices. The GeForceIII will cost no less than half a grand.

Doctorweir: The HSR technology is suppose to have that idea in mind, however I think nVidia feels that the Tile-base rendering isnt effective enough, so they developed their own.

 

[Finality]

4 pipelines is not enought. It should be a lot higher than that.
The X-box (NV25) is supposed to have 4.8GP/sec, so how will NV20 have that much fillrate?. The source is BS NFS4.

X-box specs

 

[XMan]

Consoles are programmed to display games at a fixed frame rate . . . why wouldn't NV20 be faster than what's going in the X-Box?

 

[Rudee]

At estimates of $500 to $600 US for this card, it will be at least a year from now before this card becomes mainstream.

My GeForce MX will serve me quite nicely for the next year or so until the price of this behemoth drops.

 

[interchange]

Finality:


<< The X-box (NV25) is supposed to have 4.8GP/sec, so how will NV20 have that much fillrate?. The source is BS NFS4. >>



The quoted figure for the NV20 is 1.2GP/s, which is a far cry from 4.8GP/s. I seriously doubt that their tile-based rendering techniques are anything more than an extra few passes spent at determining overlaps, which will of course speed things up. However, if there aren't any overlaps going on, or the geometry is very complex and takes a lot of cycles to compute overlaps with, then this could slow things down.

Keep in mind that 1.2GP/s is always going to be your max fill rate, but it will supposedly feel like 4.8GP/s because you're knocking out textures that don't have to be rendered. So, in effect, you're taking the same fill rate and cutting the amount of pixels by 4, which gives the effect of 4 times the fill rate.

So, claiming a 4.8GP/s fill rate on the NV20 is preposterous. On the NV25, it seems like much. But, I don't really follow the X-Box, and we all know that nothing is what it seems.

I don't believe what they are doing is going to be similar to say the PowerVR2 tile-based rendering. That would take a major architecture change. Actually, I'd like to see a *gasp* software based solution to this problem, especially on a T&amp;L based card. Especially now, there are a lot of CPU cycles to burn in many games. Using them to take load off the video card could drastically increase fps at higher revolutions where we are bandwidth limited. Although, this would be a major scaling problem and a bitch to implement. It would have to be more of a read-ahead type of thing because you couldn't design the logic to do X number of calculations in 1 clock cycle.

 

[BenSkywalker]

&quot;However, if there aren't any overlaps going on, or the geometry is very complex and takes a lot of cycles to compute overlaps with, then this could slow things down.&quot;

If there was no overdraw then a GF1 could push 100FPS+ at 1600x1200 32bit color, the assumption that complex geometry will slow it down is thinking in a tiling method which nVidia isn't doing.

&quot;I don't believe what they are doing is going to be similar to say the PowerVR2 tile-based rendering. That would take a major architecture change. Actually, I'd like to see a *gasp* software based solution to this problem, especially on a T&amp;L based card.&quot;

Why software? Use the hardware T&amp;L unit. Most everyone has seen the Unreal2 shots, that was being pushed in real time by a GF2, with a significantly more powerful T&amp;L unit the NV20 should have plenty of spare time to utilize its' T&amp;L unit for other things(much as the NV10/11/15 do now for FSAA).

For &quot;HSR&quot; effectiveness, Carmack has spoken of ten pass texturing being used for Doom3, the slightest amount of overdraw is going to crush your fillrate. 640x480 x2 x4= 2,457,600, enough to push nearly 200FPS with a GeForce DDR, no problems there. We up that to ten texture passes and we are looking at 12,288,000, under 40FPS if we don't do something about overdraw, at 640x480. Reducing the overdraw to say 1.5 in that situation and we are back in the 104FPS range. Effective fillrate will become much more of an issue then it is now, outside of the Kyro anyway

 

[RagingGuardian]

Actually NFS4's source may be correct. How do I know? Well those of us following the console race knows it's been confirmed by Microsoft's J. Allard that the X-box chip is actually a stripped down NV20, codenamed NV2A. Here's the quote:




<< JA: I can't really go into a ton of detail about the Nvidia relationship and our approach to the engineering of it. But I can say the NV20 is the core of the Xbox graphics. The guys who are working on the NV2A - we call it the NV2A inside of Xbox - are working in parallel with the NV20 folks, so they sort of marry back upafter the NV20 hits the shelves. Our developers should be getting NV20s real soon. >>

 

[NFS4]

<< The X-box (NV25) is supposed to have 4.8GP/sec, so how will NV20 have that much fillrate?. The source is BS NFS4. >>


Maybe YOUR source is BS The X-Box is based on the NV20, NOT THE NV25, and that's straight from the horse's mouth:

http://www.zdnet.com/gamespot/stories/news/0,10870,2631155-3,00.html

GS: So what are some of the key technical challenges you see to hitting the specs you've released? It's been said that the graphics are based on the Nvidia NV25 chip that's two generations away. With the NV20 said to be slipping a bit, will it be hard to make your schedule?

JA: I can't really go into a ton of detail about the Nvidia relationship and our approach to the engineering of it. But I can say the NV20 is the core of the Xbox graphics. The guys who are working on the NV2A - we call it the NV2A inside of Xbox - are working in parallel with the NV20 folks, so they sort of marry back upafter the NV20 hits the shelves. Our developers should be getting NV20s real soon.


Me luv you long time Buahaha

 

[Pocatello]

Let's place some bet. Who in here think which one comes first, the G800 or the NV20? Seen nvidia usually delivers, I going with the NV20.

 

[Fozzie]

@_@ The other possibilty for the 4.8GP number is that it is actually 4.8 Giga-SAMPLES. Which MS has indicated several times as the true measurement. This is a result of the NV20's &quot;true&quot; multisample AA(I won't get into that arguement here), which supposedly can do upto 4X AA without any extra fillrate overhead, just increased memory bandwidth usage.

 

[NFS4]

<< Let's place some bet. Who in here think which one comes first, the G800 or the NV20? >>


Definitely NV20. Matrox will have a quality card, but IMHO it won't have anywhere near the volume production as the NV20, will be priced higher, and will come a few months later.

 

[Rudee]

NFS4 - I'm curious, are you planning on purchasing a NV20 when it comes out or are you going to wait until prices drop after it's been out for six months or so?

 

[NFS4]

I'll be broke by the time the NV20 comes out (after upgrading to AMD760MP, dual Durons, DDR SDRAM). I might pick up an el cheapo GTS

 

[Sephiroth_IX]

I am happy with my Voodoo5 at the moment, but there may be an upgrade in my future. I'd sure as hell like to have a g800 if it competes...

 

[Shagga]

NV20 for me. Never thought I'd wish my life away till next February!

 

[KarlHungus]

This is a result of the NV20's &quot;true&quot; multisample AA(I won't get into that arguement here), which supposedly can do upto 4X AA without any extra fillrate overhead, just increased memory bandwidth usage.

If it's multisample AA then it has to have extra fillrate overhead (and the corresponding increase in fillrate usage).

 

[Shagga]

<< The lads at MSXBox took it upon their own to verify which NVIDIA chip will be in the XBox. General Manager of XBox operations, J. Allard told them that the XBox GPU will not be the NV20 or NV25, but rather a custom GPU:

&quot;We're working with Nvidia to develop a graphics chip which is a step beyond the NV20, it's called the NV2A. In early design discussions, we did loosely refer to it as an 'NV25', but decided to change it to NV2A to avoid, not incite confusion. it's a custom chip which is designed especially for the Xbox and some very specific requirements for the console that makes it a non-candidate for a pc card and will be faster than the NV20.&quot;
Here's some good topic for comments: Is the XBox really a console, or is it more PC-like with it's storage and upgrade capabilities, and ability to do &quot;work&quot;? Discuss. >>



Got this on VoodooExtreme

[Edit]

Sorry missed the earlier post!....Doh!

 

[Deeko]

Hmm...the T&amp;L unit will only be twice as fast? That's 50 million triangles/sec, right? That's really good, but Sage chips in the Rampage are supposedly 60, right? And the mainstream version is two....so doesn't that mean that 3dfx's will be almost 3x as fast? And to think Bubba was just telling us a few weeks ago that software is better.

 

[Finality]

I would expect so NFS4 what you thought the TNT1 and TNT2 worked on two different solutions and came up with exactly the same chip design? GTS teams would have worked on parallel with GeForce 1 teams.

 

[NFS4]

Umm, what are you talking about? And 2, most people in this thread appear to backup the fillrate claims...

 

[Goi]

If the 4.8Mpixels/s is a result of the assumption of 4X overdraw and the base 1.2Mpixels/s actual fillrate, then that has got to be the most misleading and just flat out untrue claim I've ever heard(similar to what Videologic did with their PVR series)