New NV20 Specs !!!! (semi confimed by near nvidia!)

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

_Silk_

Junior Member
Mar 2, 2000
22
0
0
Who ever made those specs up, is laughing his a$$ off right now watching this. A 70+ thread based on lies...

The 256bit bus is enough to proof its false. nVidia themselves said in several interviews that it would not be economically feasible.

You would have to add 128pins for the data path alone. Plus extra lines for ground and control lines. You?re looking at adding an extra 150pins or so. This would increase the size of the chip (thus cost), and greatly increase the complexity of the PCB layout. In order to route that many pins onto one chip, they would have to increase the numbers of layers on the PCB which would greatly increase cost.

We've already seen the true NV20 specs? They are the same as the X-Box.
 

BenSkywalker

Diamond Member
Oct 9, 1999
9,140
67
91
"I dont know if the NV20 will be 4 times faster then the NV15. If it will be, your theory is correct."

I was figuring for NV10:)
 

Blackhawk2

Senior member
May 1, 2000
455
0
0
The real question should be: is the NV20 a gamers video card? Lets take a look.

Most software to be considered popular must sell a bare minimum of 100,000 copies, the really successful games sell at least 1,000,000 copies. In order to sell a game it must be compatible with the buyers hardware. Taking a look at history will tell us that games have always scaled with the hardware of the masses.

Now lets take a look at the NV20. Rumored to cost $800, lets say that number is bogus and it really costs $500. Is this a masses video card? No. Why? Well it costs the price of two complete console systems, with revolutionary graphics. Hmm, tough choice, two consoles(or one console and 4 plus games) or one video card?

In conclusion, it looks like Carmack is going to have to scale back his ten pass rendering for Doom 3, because at $500+ not many people are going to own an NV20 and therefor in order for Doom 3 to be successful(1,000,000+ copies sold) it must be compatible with what people have at the time. Due to high prices of current video cards, that is likely to still be a Voodoo3 or TNT2.
 

Hardware

Golden Member
Oct 9, 1999
1,580
0
0
Some confirmation from near nvidia: (in German)

Aus Nvidia nahen Quellen stammen detaillierte Informationen zum GeForce-2-Nachfolger, dem NV20-Chip. Neben der verbesserten Geschwindigkeit von 300 MHz Chiptakt und 200 MHz Speichertakt, also einem DDR-RAM-Takt von 400 MHz, wurde vor allem der GPU-Kern aufpoliert.

So soll der NV20-Chip sechs Rendering-Pipelines besitzen, die pro Rechendurchgang 3 Texturen erzeugen können. Dies ermöglicht den Einsatz von überlagerten Texturen in kommenden Spiele-Generationen, die mehr Realismus, Tiefe der Details und flüssiger animierte Charaktere bieten werden.

Die Füllrate eines NV20-Chip soll bei maximal 1,8 Gigapixel liegen, was 5,4 Gigatexel pro Sekunde entspricht. Zum Vergleich: Der GeForce 2-GTS-Chip schafft 800 Megapixel, beziehungsweise 1,6 Gigatexel pro Sekunde.

Nvidia will den NV20-Chip im Herbst auf der Comdex vorstellen. Erste Karten werden im Frühjahr 2001 erwartet, wobei spekuliert wird, dass der GeForce-2-Nachfolger mit 64 MByte DDR-RAM zu einem Preis von etwa 1.600 Mark zu haben sein wird.

Source
 

Blackhawk2

Senior member
May 1, 2000
455
0
0
<<Carmack knows whats up in development, and he will get what he wants sooner or later.>>

Sure he can have what he wants, but the question is will people buy it? He could make Doom 3 for an uber NV20 system that he might own. But if it doesn't sell Carmack is left in the dust. Fantasy and reality are far apart.

 

NFS4

No Lifer
Oct 9, 1999
72,636
47
91
Here you go Hardware:



<< From Nvidia close sources originate detailed information to the GeForce-2-Nachfolger, the NV20-Chip. Apart from the improved rate of 300 mc/s chip clock and 200 mc/s storing ATS, a GDR RAM clock of 400 mc/s, above all the GPU core was thus up-polished. So the NV20-Chip is to possess six Rendering pipelines, which can produce 3 textures per computing passage. This enables the use of overlaid textures in coming play generations, which more realism, depth of the details and liquid animated characters will offer. The filling rate of a NV20-Chip is to be with max. 1.8 Gigapixel, which corresponds to 5.4 Gigatexel per second. To the comparison: The GeForce 2-GTS-Chip creates 800 megapixels, or 1.6 Gigatexel per second. Nvidia wants to present the NV20-Chip in the autumn on the Comdex. First cards are expected in the spring 2001, whereby it is speculated that the GeForce-2-Nachfolger with 64 MByte will have to be had GDR RAM at a price of approximately 1,600 Marks. >>


 

Marty

Banned
Oct 11, 1999
1,534
0
0
Those German specs are invented as well. Go to this Beyond3D thread to read up on some specs for the 3dfx Rampage. Those Rampage specs, as well as these NV20 specs, are simply fabricated.

Marty
 

Sunner

Elite Member
Oct 9, 1999
11,641
0
76
Blackhawk, ever heard of scalable LOD?
Have a NV20 Ultra Turbo GT+, boost detail levels.
Have a GF DDR, run in low detail mode.
Have a TNT2 Ultra, run in really low detail mode.

As for 10 pass rendering and 64 bit color, he will surely get it sooner or later, when every card has it, it will be inexpensive for everyone to buy.
 

Blackhawk2

Senior member
May 1, 2000
455
0
0
<<As for 10 pass rendering and 64 bit color, he will surely get it sooner or later, when every card has it, it will be inexpensive for everyone to buy>>

True, but when will these cards be inexpensive enough for everyone to buy? 5 years from now?

 

VladTrishkin

Senior member
Sep 11, 2000
421
0
0
10 pass rendering, 64 bit color 1600x1200 FSAA 8x @ 100FPS = NV50; can you say $10,000? Then get the Pentium 7 for $15,000 more! I think i'll go buy a BMW M5 for that $$$ now. :)
 

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,395
8,558
126
you found an m5 for $25,000?!? we're not talking about one of those they made way back when, are we?
 

Blackhawk2

Senior member
May 1, 2000
455
0
0
Will Nvidia products be reserved for CAD designers? Is this the end of mass market gaming or will 3dfx Matrox and ATI come to the rescue? Could an unknown pop out of nowhere and blow Nvidia's 800+ engineers out of the water? Stay tuned to Comdex to find out...:D
 

VladTrishkin

Senior member
Sep 11, 2000
421
0
0
ElFenix, Ehh add 1GB of RDRAM or DDR-RAM, a nice Intel mobo, X15 Quad SCSI Raid, and you will be at $50,000 :D

Of course i am not 100% accurate here, but i am just making a point that the &quot;new&quot; prices are very high. Take the $500 Ultra for example.
 

Blackhawk2

Senior member
May 1, 2000
455
0
0
What has more power? 8 Duron 750MHz(they are now $78 each:)) chips or one $800 NV20? I could have 1 Duron chip dedicated to rendering, 5 Duron chips dedicated to physics, 1 Duron chip controlling the flow and 1 Duron chip doing cleanup or didly squat :D

Doable?
 

Dark4ng3l

Diamond Member
Sep 17, 2000
5,061
1
0
how the hell are you going to make all those durons work in one system. And a geforce 2 would be faster in rendering than all those durons.
 

Dark4ng3l

Diamond Member
Sep 17, 2000
5,061
1
0
actualy it was 8 I dont know where that guy got a mobo that can use 8 duron/t-bird cpu's
 

The Wildcard

Platinum Member
Oct 31, 1999
2,743
0
0
Those pipeline specs are pretty damm kick ass, but that price? i might have to donate blood, lol. which is worth it for an nvidia card though, lol.
 

Soccerman

Elite Member
Oct 9, 1999
6,378
0
0
&quot;FSAA- Pretty much, free in terms of memory bandwith. With the method that the NV20 is using it will only require a single texture read from memory, and at the native resolution. The fillrate hit will be an issue, but that is something that the GF2 boards already have loads of to spare.&quot;

if that were true, we'd see the GTS wasting the Voodoo 5 in terms of performance hit %, becuase as you put it, bandwidth isn't any more an issue, POWER IS. correct me if I'm wrong..

ie, if the GTS is memory limited without FSAA, and has gobs of power to spare (I don't doubt that), then when you enable FSAA, all that extra power goes into use, and therefor you don't have nearly as much slowdown..

second, Vlad, the reason 256 requires ALOT of money, is cause you have to add 128 MORE wires to connect to the CPU. SO, you'd have 256 for MEMORY ALONE, let alone anything else. it costs money to fit more connectors to a chip.

as for price. I sincerely hope they don't get many buyers (anyone who does that, I hope will not get anywhere near as many buyers as before). I don't consider it a steal. I consider it a huge waste of money, when it costs soo much less to make (and even including R&amp;D).

I won't buy anything over $200 STILL. screw u nVidia, and 3dfx. if your price does NOT meet my price, you don't get my money.

those ATi cards are starting to look nice..
 

VladTrishkin

Senior member
Sep 11, 2000
421
0
0
Soc;



<< if that were true, we'd see the GTS wasting the Voodoo 5 in terms of performance hit %, because as you put it, bandwidth isn't any more an issue, POWER IS. correct me if I'm wrong..

ie, if the GTS is memory limited without FSAA, and has gobs of power to spare (I don't doubt that), then when you enable FSAA, all that extra power goes into use, and therefor you don't have nearly as much slowdown..
>>



-First of all, we are not talking about cars, what do you mean by &quot;Power&quot;? You mean peak 3D rendering capability of the GPU? MHz Clock? Instructions per pass/clock?

No one is 100% correct here. FSAA does NOT take too much bandwidth, (IE: 4x FSAA does not take 4 times more bandwidth at same resolution), but the GPU has a bit of a struggle to keep up (Processes 4 new textures per clock @ FSAA 4x instead of 1). This problem would be solved if NV20 were say 2-4 times faster than NV15.




<< second, Vlad, the reason 256 requires ALOT of money, is cause you have to add 128 MORE wires to connect to the CPU. SO, you'd have 256 for MEMORY ALONE, let alone anything else. it costs money to fit more connectors to a chip. >>



-I see that NO ONE understands this. 128 more interconnections are NOT needed (in most cases), there are MANY techniques used, one is simply making the gate paths wider and using fewer interconnections. This is a CMOS topic.

PM, back me up here, join us;)