nVidia release new all singing all dancing dets.

[nggalai]

Hi Hellbinder,

]So i guess no one seems to care that they achieved their *25% performance increase* by hacking their IQ to hell??? You actually think there is some other *magical* way they could do it as optomized as their drivers already were?... I wonder if well see Kyle make a big deal out of this. Or claim that they are cheating.... hmmmm....

Look close at nature..... a 30 FPS increase in that test... See what your eyes tell you. Morrowind. Or other games. They are disabling Filtering and tanking LOD in distant objects and odd angles.

Thats no victory in the driver department IMHO.... 8) But then I am only one small voice in a sea of many.

Err . . . that was sarcasm, wasn't it? Or did you really miss Neeyik's aniso 3DMark comparison on the last page?

Well, these drivers are quite obviously "beta". Thank goodness I can't install them on my home system for lack of WinXP.

ta,
-Sascha.rb

 

[R300King!]

I haven't seen blocky textures like that since... oh... since I had a Commodore 64!

You had textures on your C64? How'd you fit them into the 64 k of memory?

..it wasn't easy!

 

[Doomtrooper]

http://www.pc-extreme.com/artikel/detonator-test/detonator4041/

 

[Reverend]

Here's a bit of what NVIDIA wrote me :

Performance
In many popular applications and benchmarks, we see a substantial increase
in performance using Detonator 40. Antialiasing performance and anisotropic
filtering performance will also increase with the Detonator 40 driver.
A few highlights include:
* Quake 3 goes up around 5% across the board, peaking at over
10%
* 3D Mark 2001SE goes up around 7% across the board, peaking
at over 10%
* Aquanox goes up to 28% faster in some modes
* 3D WinBench goes up to 13% in some tests
* Serious Sam goes up around 4% across the board
* Operation Flashpoint: Cold War Crisis goes up to 8%

For Quadro users, the performance bump is even more impressive:
* ranging from a minimum of 14% in one Viewperf 7.0J test all
the way up to 19% in another.
* SPECapc 3DStudioMax v4.2 R4 graphic mean score increases
20%,
* SPECapc Solidworks 2001plus graphic mean benchmark goes up
over 30%.

 

[Doomtrooper]

I have a question here, why is Quake 3 still a target for speed bumps (ummm its 3 years old and there is maybe 1200 people online playing it), wouldn't it make more sense for MOHAA,Wolfenstein etc ??

3Dmark ??
3Dwinbench ??

 

[Galilee]

I'm working on a small article/test myself. Will be up sometime tomorrow:

What I have tested:
Aquamark, UT, SS, Mafia, 3dmark and Vulpine GLMark.
And image comparations from UT2003.

 

[KimB]

Because it's still the latest id software engine we have, Doomtrooper.

It seems pretty obvious that id's games will be major benchmarks for some time to come, and there will also be times like this again where the latest from id is actually quite old.

 

[Doomtrooper]

Hardly anybody plays it Chalnoth...Wolfenstein and MOHAA players which outnumber the Quake 3 players by about 20,000 people, these players would appreciate some 'speed improvements'.

BTW ID had alot to do with Wolfenstein (Quake 3 engine) and MOHAA certainly could use some peformance bumps (Quake 3 engine again)..yet alot more modern engines for todays high end cards.

 

[martrox]

SteveG, sorry, but these kind of half truths go way back.....remember when all the T & L lighting games were coming out Christmas.....3 years ago? Remember when nVidia claimed Diablo II had T & L? These are the kind of things I'm talking about. As far as ATI goes, I'm sure they do similar things...and get called on it also. As I own only nVidia, I can only speak on what I have.....although R300's will be in my & my wife's gaming boxes in the near future. Of course, IF NV30 improves my gaming experience....then they will replace the R300's. I have no allegience to any company.....I buy what best serves me. I don't defend any of these companys.....but I will call em as I sees em......

 

[Galilee]

In those games: Buy a new cpu

Here are some Aquamark scores:
(not doublechecked)

I know they are hard to read this way

40.41 (no aniso) 30.82 (no aniso) 40.41 (8X aniso) 30.82 (8X aniso)
1024*768 69,1 62,9 26,7 14,5
1600*1200 36,6 30,3 10,9 6,5

2X 1024 48,4 41,6
Quincunx 1024 48,4 41,7
4X 1024 28,2 25,6
4XS 1024 26,2 24,2

Around 45% gain in 1024 with 8X aniso. Atleast there is the 25% (no make that 45% )

 

[Doomtrooper]

Wolfenstein is not CPU limited...

Do you have screen shots comparing the Aquanox shots...is their IQ degradation between the two sets ??
Of course a shot of the sub along the ocean floor would be ideal.

 

[Galilee]

No screens from that (but I have UT2003 screens soon, will be posted). It looked the same in Aquanox, sharp and nice both drivers.

Ut2003, 1024 max

New drivers no aniso:
http://www.stud.ntnu.no/~vidaralm/UT2003/4041_1X.jpg

Old drivers no aniso:
http://www.stud.ntnu.no/~vidaralm/UT2003/3082_1X.jpg

New drivers 8X aniso:
http://www.stud.ntnu.no/~vidaralm/UT2003/4041_8X.jpg

Old drivers 8X aniso:
http://www.stud.ntnu.no/~vidaralm/UT2003/3082_8X.jpg

The new drivers are actually slightly more blurry faar faar away on the 8X pictures. Switch back and forth. Without aniso they are similar.

 

[Nagorak]

So basically these drivers boost pixel shader scores... Current games will basically not be affected, but next generation games (and the few current games with PS) will benefit from it.

 

[Doomtrooper]

Looking at your anisotropic shots, the filtering is affected, I'm not sure but it also looks like textures are missing, darker or something around the door compared to the old driver...

 

[Doomtrooper]

..

 

[cellarboy]

* Quake 3 goes up around 5% across the board, peaking at over
10%
* 3D Mark 2001SE goes up around 7% across the board, peaking
at over 10%
* Aquanox goes up to 28% faster in some modes
* 3D WinBench goes up to 13% in some tests
* Serious Sam goes up around 4% across the board
* Operation Flashpoint: Cold War Crisis goes up to 8%

Does it strike anyone as funny that with the exception of Flashpoint, all the other software mentioned are popular benchmarking titles??

 

[noko]

Using SeriousSam SE I see no signigicant difference in performance or IQ using the new drivers in anisotropic filtering performance. Here are the results using a static screen shot in the Techdemo of SS-SE. GF3 Ti200 at 240/500 core/mem. This is without MSAA. Below images you can compare the FPS (upper right corner) and IQ from the different in game AF settings. System specs below.

NoAF

2xAF

4xAF

8xAF

In Summary

From:

1. NoAF - 2xAF, 58 - 42 FPS -> 28% drop
   NoAF - 4xAF, 58 - 29 FPS -> 50% deep drop
   NoAF - 8xAF, 58 - 20 FPS -> 66% Plunder!

A pretty much non T&L game like SeriousSam SE shows no significant difference between the new and old drivers, at least for me. Another note is that 3dMark2001se does not predict nor indicate the performance change or lack of performance change in this program. To believe that a change in 3dMark2001se means a change will happen in other programs is obviously a fallacy. You must know what caused the change in 3dMark2001se to be able to determine if another program will reflect similar result changes. To say that 3dMark2001se predicts 3d performance I would say is streching the truth.

 

[RussSchultz]

You're right. There's a big gray circle on the new one. I'd never buy a card that did that.

 

[Doomtrooper]

Free hand circle I might add

 

[Babel-17]

I'm looking forward to seeing what Unwinder, creator of RivaTuner, can do with these drivers. The possibility that nVidia is currently exploiting by default a bit of the opportunities for optomizations/tuning for speed he's managed to exploit has occured to me. It would of course be excellent if in fact his tweaks were something that nVidia hasn't tapped into and if they were to be enabled for these new drivers.