GeforceFx only 200 Mill Triangles/s

[-=DVS=-]

Lol must be a typo or a very slow Geforce Fx becouse Radeon9700Pro Does 325 MT/s , GeforceFx Ultra model at 500Mhz is suppose to do 350 MT/s

0,13 micron GPU fabrication
125 million transistors (give or take a few, I didn't count)
DDR2 memory clocked at 1 GHz
51 billion floating point operations per second (51 gigaflops) in the pixel shader alone
Advanced Programmability (3rd generation)
High-precision color (64-bit & 128-bit color)
High-level Shading Language
New Vertex and Pixel Shading instructions
Highly efficient architecture (3rd generation Lightspeed Memory Architecture)
High Bandwidth to memory and CPU
Shaders can be 1000's of instructions long.
8 pixels per clockcycle rendering power
200 Million Triangles per second
64-bit & 128-bit color, this is film-quality precision, in fact a higher precision than the movie Toy Story 2 used. 64-bit offers high precision with 2x the performance & half the memory of 128-bit. As it seems developers want both 64 and 128-bit color precision for advanced effects.
AGP 8x (over 2GB/sec bandwidth to the system).
Fully DirectX9 compatible
Pixel Shaders 2.0
0.13 Micron fabrication process.



LINK to Guru3D (http://www.guru3d.com/tech/geforcefx/?PHPSESSID=c22dcdfc7028a1375c670c33e12677ea)

[MuFu]

Doesn't sound right at all. It's probably 300 or 280.

Nice of them to put 0.13u twice just in case we didn't catch it the first time. :D

MuFu.

[Uttar]

Remember: There's a huge different between triangles and vertices.
After doing a few calculations, 75% of AGP 8X & 64MB of memory reserved for vertices would still only allow you to have 165M Vertices/s
If you wanted to get to 350M Vertices/s, you'd have to use about 83% of AGP 8X, slightly over 128MB of Video Memory ( there should be 256MB GFFX, so it's possible ) and 50% of the GFFX bandwidth only for that.

Nobody in his right mind is going to do that.

And that's an optimal case; in real-world scenarios, you probably aren't going to use every single static vertex in the same frame. That could only happen in a technology demo or a synthetic benchmark.

So, if you want that 350M Vertices/s in a real-world application, you may need 512MB of video memory and 2 times the current GFFX memory bandwidth :D

Remember, the GFFX is all about *better* vertices/pixels. And if you want better vertices/pixels, it doesn't cost you much more memory/bandwidth. It simply requires you to have less vertices and pixels. That raw power only enables you to have better pixels. So it's USELESS in current games.

Thus, more than 200M Triangles/s would be wasted triangle setup engines. I think this was already disclosed in a November spanish for-developer document, but it wasn't very clear...

The GFFX is truly future-oriented. Even Doom 3 is way too basic for it :)


Uttar

[GamblerFEXonlin]

Originally posted by Uttar
Remember, the GFFX is all about *better* vertices/pixels. And if you want better vertices/pixels, it doesn't cost you much more memory/bandwidth. It simply requires you to have less vertices and pixels. That raw power only enables you to have better pixels. So it's USELESS in current games.

And as Carmack don't want more polygons, just passes, we can look forward to 5 years of edgy heads and square fingers? At least if the game uses Doom3 tech, i.e realtime shadows and multiple lightening sources? Raw polygon performance is useless if it can't be used with the new revolutionary features enabled. This time its 4 generations old per-pixel lightening and regular T&L calculations.

Sorry to be hars here but I'm very critical to all the hype around a new generation, and doom3s edgyness dissapointed me. Finally we see a game use the incredible features of geforce1 and we end up with edgyness.


Originally posted by Uttar
The GFFX is truly future-oriented. Even Doom 3 is way too basic for it :)

yes, as this quote from Carmack say:
"My current work on Doom is designed around what was made possible on the original GeForce, and reaches an optimal implementation on NVIDIA's next-generation GPU. My next generation of work will be designed around what is made possible on NVIDIA's next-generation GPU."

But he also did not advice us to buy a GF4 to play doom3... that means the game was pretty much intended for a GFFX and Radeon 9700. We have already concluded the GFFX isn't fast enough to use its extreme programmability, does that mean we can forget graphics like the Dawn demo? NVIDIA did release demos with high-poly models with the GF1...

[Uttar]

Originally posted by GamblerFEXonlin
We have already concluded the GFFX isn't fast enough to use its extreme programmability

Actually, I'm not so sure about that. In a perfect 100% front-to-back case, it could be possible to use programs of multiple hundred insturctions on like 10% of the scene ( assuming the rest of the scene uses something like 10 instructions maximum )
I'm very interested in seeing how true that prediction is with a test program I'll make once I have a GFFX.

Please also note that this is obviously only true at 32BPP; and since the programmers can switch to 32/64/128BPP during a program, it might be less than that in practice.


Uttar

[Lezmaka]

Originally posted by GamblerFEXonlin
yes, as this quote from Carmack say:
"My current work on Doom is designed around what was made possible on the original GeForce, and reaches an optimal implementation on NVIDIA's next-generation GPU. My next generation of work will be designed around what is made possible on NVIDIA's next-generation GPU."

But he also did not advice us to buy a GF4 to play doom3...

Um... no. He said don't buy a GF4 MX for Doom 3.
http://www.webdog.org/plans/1/
On the topic of current Nvidia cards:

Do not buy a GeForce4-MX for Doom.

Nvidia has really made a mess of the naming conventions here. I always thought it was bad enough that GF2 was just a speed bumped GF1, while GF3 had significant architectural improvements over GF2. I expected GF4 to be the speed bumped GF3, but calling the NV17 GF4-MX really sucks.

GF4-MX will still run Doom properly, but it will be using the NV10 codepath with only two texture units and no vertex shaders. A GF3 or 8500 will be much better performers. The GF4-MX may still be the card of choice for many people depending on pricing, especially considering that many games won't use four textures and vertex programs, but damn, I wish they had named it something else.

If a GF3 is much better than a GF4MX, and since a GF4 Ti is fast er than a GF3, I think a GF4 will be just fine. Of course if you want to run at high resolution and at max details you'll need a GFFX or a 9700.

[StealthHawk]

Originally posted by GamblerFEXonlin
And as Carmack don't want more polygons, just passes, we can look forward to 5 years of edgy heads and square fingers? At least if the game uses Doom3 tech, i.e realtime shadows and multiple lightening sources? Raw polygon performance is useless if it can't be used with the new revolutionary features enabled. This time its 4 generations old per-pixel lightening and regular T&L calculations.

4 generations? it is using NV10 tech. next gen was NV20. and the current gen is DX9 and NV30, which we all hope will be out before Doom3 :surrender

the various refreshes didn't really add any new tech, and shouldn't be called generations. plus, if you look at ATI cards, you will also get 3 generations. the r256 should be just as capable in running Doom3's feature set as the gf256.

[ricercar]

MSI claims 350 Million Triangles per Second - 3x The Geometry Performance of a GF4 Ti (http://www.msicomputer.com/pressrelease/GeForceFX.asp)

[Uttar]

Originally posted by ricercar
MSI claims 350 Million Triangles per Second - 3x The Geometry Performance of a GF4 Ti (http://www.msicomputer.com/pressrelease/GeForceFX.asp)

*sigh* When will everyone accept vertices as a standard instead of triangles? Triangles is outdated since the day T&L was implemened in the GeForce 256. It's ( most of the time *cough* Parhelia *cough* ) a completely useless figure.


Uttar

[ReDeeMeR]

lol I clearly remember he said that Geforce 4Ti4600 would be just perfect, then I saw Radeon9700 promo trailer and Mr Carmack was telling how r300 will let you enjoy D3 in all its glory, now FX is on the horison and he's saying FX is just perfect :rolleyes:

Who knows maybe he will shutup when Doom3 is finaly released, or the two companies will keep using him as a good promo to their products. :)

[lagadu]

in a few weeks he'll say that you'll need a R350 to run it with all it's glory, if the Parhelia2 is a competing product he'll say it's the one card for D3, and then he'll say that anything other than a NV35 will only be able to play D3 @ 800x600 16bits.
and he'll keep like this until D3 is finally released in 2005 :p

[Greg]

This is starting to get away from the topic but...

Hmm, I think I see the pattern with Carmack... He likes the fastest card available!

[LORD-eX-Bu]

doom 3 on a surround view display would be sweet :D

[PreservedSwine]

All the webites selling the FX have changed their numbers from 200 to 350.

Then *back* to 200.

It looks like the 200mill/sec is accurate after all......:(

[ReDeeMeR]

wow, it's gnna su** badly, I also saw like 3 sites listing it with spec "200t m/s"

[Uttar]

I already explained in this thread that it makes no difference at all.
So please, just quit talking about it unless you can proof my explanation is incorrect.


Uttar