nVidia release new all singing all dancing dets.

[Grall]

"Quite frankly, the individual "game" tests in 3DMark are just as valid as any other 3D benchmark. Does a timedemo in Q3A tell you anything about how many frames per second you'll get while playing online?"

Uh, no dude, they're absolutely NOT just as valid. In fact, they're hardly valid at all.

While a timedemo will not directly relate to fps during online play (as that depends on position/viewing angle of the player, amount of players visible, amount of explosions etc on-screen and so on), any driver optimization that raises timedemo FPS in Q3A will at LEAST raise the POTENTIAL for raised FPS in online play as well, while driver optimizations to raise 3dmark scores won't result in anything other than an inctreased 3dmark score (which is USELESS, since we don't play 3dmark2k1), and any general optimization to the driver that raises 3dmarks is just as likely to raise the speed of other apps by just as much on average.


*G*

 

[Sabastian]

"Nvidia Playing the 3DMark Game? - Thursday, August 29 | Solomon

Reports are coming in that the supposed, "Up to a 25% increase in performance" is only showing in 3DMark2001 benchmark. While playing games, Quake 3 is appearing darker then usual. More and more reports will be coming in regarding the new Detonator 40 driver. From the looks of it, it's not what people where really interested in as you can't play 3DMark2001.

More reports to come... "

http://www.3dchipset.com/news.shtml#newsitem1030639337,5718,

EDIT: Darn I was looking forward to my "free" Det40 driver d/l after my $ 1000 upgrade. Oh well... Maybe I will just buy a Radeon 9000

 

[alexsok]

Aniso 8x Perfomance:

Note: Unfortunately, I have no results of Aniso 8 perfomance with 30.82 detonators, so I'll just compare the 40.41 perfomance with 40.41 Aniso 8 perfomance.

Score: 40.41: 11400 - 40.41 Aniso 8: 8298

Game 1 LOW: 40.41: 162.4 - 40.41 Aniso 8: 136.9
Game 1 HIGH: 40.41: 49.3 - 40.41 Aniso 8: 48.5

Game 2 LOW: 40.41: 215.8 - 40.41 Aniso 8: 120.6
Game 2 HIGH: 40.41: 116.8 - 40.41 Aniso 8: 76.7

Game 3 LOW: 40.41: 146.1 - 40.41 Aniso 8: 109.8
Game 3 HIGH: 40.41: 65.2 - 40.41: 56.6

Game 4 Nature: 40.41: 76.6 - 40.41 Aniso 8: 49.5

Fillrate (Single-Texturing): 40.41: 1063.7 MTexels/s - 40.41 Aniso 8: 611.2MTexels/s

Fillrate (Multi-Texturing): 40.41: 2326.6 MTexels/s - 40.41 Aniso 8: 611.2 Mtexels/s (what i was talking about - that hardware bug).

High Polygon Count (1 Light): 40.41 55.0 MTriangles/s - 40.41 Aniso 8: 52.8 MTriangles/s

High Polygon Count (8 Lights): 40.41: 12.6 MTriangles/s - 40.41 Aniso 8: 12.4 MTriangles/s

EMBM: 40.41: 152.6 - 40.41 Aniso 8: 97.2

DOT3: 40.41: 163.3 - 40.41 Aniso 8: 81.3

VS: 40.41: 98.8 - 40.41 Aniso 8: 79.6

PS: 40.41: 123.5 - 40.41 Aniso 8: 71.2

ADV PS: 40.41: 98,8 - 40.41 Aniso 8: 48.9

Point Sprites: 40.41: 33.0 MSprites/s - 40.41 Aniso 8: 33.0 MSprites/s

 

[TempestX]

New Drivers

One thing I have noticed is that there is a new button to check under the Antialiasing tab called Texture Sharpening. I clicked this and was using Quincunx AA and the textures were sharp while playing Madden 2003. Still exploring.

Corrected spelling of (sharpening).

 

[alexsok]

Re: New Drivers

One thing I have noticed is that there is a new button to check under the Antialiasing tab called Texture Shrpening. I clicked this and was using Quincunx AA and the textures were sharp while playing Madden 2003. Still exploring.

Yep, this was also available on RivaTuner for ages!

Also, as Cass (Nvidia guy) promised me, GL_DEPTH_CLAMP is now included in the drivers (I suppose as part of the OpenGL 1.4.0 support) and u can enable it OpenGL, although I noticed artifacts while using it...

I tried that Tenebrae mod for Quake 1 with the new drivers and everything seems smoother.

Gonna test some other games later today.

 

[Neeyik]

Grall - in essense you're saying that even if the tests in 3DMark all showed a massive increase, there would be no potential increase in any DX8-based game. Besides, you've contradicted yourself here...

"Uh, no dude, they're absolutely NOT just as valid. In fact, they're hardly valid at all."

"any general optimization to the driver that raises 3dmarks is just as likely to raise the speed of other apps by just as much on average."

That makes looking at the 3DMark tests just as valid as looking at a Q3A timedemo and seeing if that has improved too.

 

[Xmas]

Anyone noticed that "0x" AF setting in Galilee's dirver panel shots? There are five ticks on that slider, so does this mean 0x, 1x, 2x, 4x, 8x? What's 0x supposed to do then?

 

[pxc]

Anyone noticed that "0x" AF setting in Galilee's dirver panel shots? There are five ticks on that slider, so does this mean 0x, 1x, 2x, 4x, 8x? What's 0x supposed to do then?

Isn't 0x reverting to bilinear filtering?

 

[alexsok]

Anyone noticed that "0x" AF setting in Galilee's dirver panel shots? There are five ticks on that slider, so does this mean 0x, 1x, 2x, 4x, 8x? What's 0x supposed to do then?

0x means off m8!

 

[Doomtrooper]

Grall - in essense you're saying that even if the tests in 3DMark all showed a massive increase, there would be no potential increase in any DX8-based game. Besides, you've contradicted yourself here...

"Uh, no dude, they're absolutely NOT just as valid. In fact, they're hardly valid at all."

"any general optimization to the driver that raises 3dmarks is just as likely to raise the speed of other apps by just as much on average."

That makes looking at the 3DMark tests just as valid as looking at a Q3A timedemo and seeing if that has improved too.

Neeyik,

3Dmark is not a accurate way of judging performance ...I'm sorry I would rather see:

Some leaked UT 2003 Build tests
Nolf2 Demo with fraps
Serious Sam 2:
SOF2

etc...

3Dmark is not a game...I don't play it and we all know lots of Optimizations go into it.

 

[Xmas]

0x means off m8!

Off, 2x, 4x, 8x - that's four options. What's the fifth setting? That's why I'm asking.
In RivaTuner, setting 0 means force point sampling.

 

[Basic]

0x means off m8!

Yes, most likely.
But I must say that it sounds rather stupid. 1x would be better for "off".

 

[alexsok]

[quote[Off, 2x, 4x, 8x - that's four options. What's the fifth setting? That's why I'm asking.
In RivaTuner, setting 0 means force point sampling.[/quote]

I'm not sure I understand...

In Direct3D u have: 0x (which is off), 1x, 2x, 4x & 8x

In OpenGL u have: Disabled, 2x, 4x, 8x

 

[Bambers]

Either 0x or 1x is a pointless option. 1x anisotropic filtering is isotropic filtering.

 

[Galilee]

I did something with the aniso setting and suddenly got point sampling or something, looked like crap.
Wen't to Rivatuner and changed to "Application controlled" and it was fine again.
0X is Use nearest point, compared to Rivatuner.

 

[Neeyik]

I find the sheer, blind refusal to see 2 sides of the coin rather odd. Let's take these hypothetical statements:

Benchmark figures from Quake 3 are more valid than 3DMark
How so? Test figures from Q3A only tell you one thing - how fast Q3A is running in a timedemo. It certainly doesn't translate to how fast other OpenGL applications run - certainly you've got many games out there that are based on the same engine but if you got a 20fps increase in Q3 would that mean anything in MOHAA or JKII? What about NWN?

3DMark isn't a game, so the results aren't valid
Is a timedemo from SS or Q3 a "game"? Certainly they run using the actual game engine that you use when playing it and an increase in timedemo speed probably will reflect itself within the actual game - but, by how much?

All benchmark results are meaningless when compared to another benchmark or "real life" situation - by their very nature, a decent benchmark is nothing like a "real life" game as the test conditions must be repeatable to a close degree for it, which is not how games invariably turn out. Benchmarks, like any investigation, are good for comparisons on the same basis - e.g. same PC, just different video cards.

Somebody just posting figures from benchmarks of the latest drivers, in ANY game/3DMark/whatever, without prior comparisons aren't worth considering, as they contain no reference points. However, I don't see anything wrong with somebody using 3DMark to investigate if a new set of drivers have improved something like antialiasing performance or anything similar.

 

[Galilee]

About anisotropic filtering: Use 1X. 0X looks like total crap (near point sampling), while 1X is normal "no aniso".

 

[Reverend]

As mentioned, some quickie benchmarks before I crash :

GLQuake - Tenebrae mod
1024x768x32 - NoAA, NoAniso

Maps vis'ed - used Tenebrae's demo1

Relevant settings that are identical for both driver sets :
gl_picmip 0
gl_playermip 0
gl_texturemode GL_LINEAR_MIPMAP_LINEAR
gl_finish 0
gl_flashblend 0
gl_polyblend 1
gl_triplebuffer 1
r_shadows 1
r_wateralpha 0.5
sh_entityshadows 1
sh_lightmapbright 0.8
sh_fps 1
sh_infinitevolumes 0
sh_visiblevolumes 0
sh_worldshadows 1
gl_watershader 1
sh_playershadow 0
sh_noscissor 0

30.82 = 43.1fps
40.41 = 43.4fps

Tiger Woods 2002
Using saved replay of hole-in-one at Augusta, using "ball cam", using FRAPS, average framerate

1024x768x32bit - NoAA, NoAniso
30.82 = 63.045
40.41 = 63.125

1280x1024x32bit - NoAA, NoAniso
30.82 = 48.137
40.41 = 48.109

F1 2002
One lap at Monza with "F1 2001 driving help" (which means all I need to do was hit the accelerator w/o needing to steer, pretty much identical steering line everytime this way), using FRAPS, average framerate

1024x768x32bit - NoAA, NoAniso
30.82 = 97.872
40.41 = 98.965

1600x1200x32bit - NoAA, NoAniso
30.82 = 65.880
40.41 = 66.524

Looks like I wasted time benchmarking For all intents and purposes, in the games tested above, performance is similar. As for UT2003, I thought better of it and decided not to benchmark.

Obviously there should've been tests involving AA and/or aniso but like I said, I'm sleepy

I would also like to try benchmarking any game featuring both texture and vertex programs (like 3DMark's Nature) to see if the "optimizations" may be about texture/vertex programs or just Nature alone Tomorrow perhaps...

 

[Basic]

OK, makes more sense.

0x - point sampling
1x - isotropic filter (bilinear or trilinear)
2x, 4x, 8x - anisotropic filter at given degree of anisotropy

Only thing that doesn't make sense is why anybody would want to force point sampling.

 

[Nexiss]

The reason I think it is somewhat infuriating to see (almsot) nothing but 3dmark scores is that they are meaningless if the increase is due to optimizations specifically for 3dMark that may or may not show up in real games. We need some benchmarks of actual games to actually see what kind of increase (or decrease) we may see during gameplay.

That said, the Nature improvement is impressive (although, not if they have been holding back performance, then it is just maddening).