nVidia release new all singing all dancing dets.

[BoardBonobo]

Under the spreading chestnut tree,
the ATI rep sat.
He amused himself by abusing himself.
And catching it in his hat.

Game Over


---------------------------------------------

I started this thread to find out the how and why nVidia could state that the new dets give a 25% speed increase in some applications. Apart from the usual low blows and cheap shots it would appear that some people have givenit some serious time and effort. Thankyou.

The long and the short is that nVidia have given control of the filtering settings to the user, but cocked it up. I guess a new revision will be out soon. The speed inrease in probably in the realms of 2%-3% over the previous driver set when Aniso is used and the FSAA modes have been altered slightly(?). They seem to be of better quality and slightly faster. But that maybe a topic for another forum.

Apart from a slight (real world) speed incease, the only new things they have provided are a jazzed up interface, Macrovison protection is now on by default, and (the most interesting to my mind) support for the entire NV30 OGL extensions. They're all there listed in the nvoglnt.dll. Along with some funny stuff like BoomBoom\BoomBoom2 emulation(?) and Buzz emulation(?).

nVidia were rumoured to be at 95% efficiency with their previous driver sets, a 25% increase would put them at 120% efficiency... That's why I reckon in the real world we are looking at 97-98% now.

Mods:
Is it time for this thread to buy a house in Eastbourne and live happily ever after by the sea side.

 

[mboeller]

looking at all this mess,

I can only come to the conclusion that a lot of people are going nuts, because (for me) it seems Nvidia has transformed into ATi (being late with new cards/chips and producing buggy drivers with cheats(also called "optimisations) for 3DMark2001 and QuakeIII ), whereas ATi now delivers it's chips/cards on schedule and the drivers (magically) have only minor issues, even when they are for an new card like the R9700.

 

[Xmas]

I really hope NV30 is able to do much better than that (eg. some 6x jittered grid AA like R9700)

Only if they have some form of compression for the samples like the Radeon9700 has with HyperZ. Otherwise the Radeon9700 will walk all over it in performance.

Either that or some other algorithm. But I think compression is the obvious thing to do in combination with multisampling, and NVidia should be smart enough to do it right this time.

 

[walkndude]

Just a good example that these boards are just as gay as any other

 

[Galilee]

Atleast we know this is "for fun"
But yeah, it's time to lock this thread.

 

[walkndude]

I wouldnt be so sure it wasnt taken seriously by quite a few.

Lock it? i say DELETE IT

 

[Dave Baumann]

Just a good example that these boards are just as gay as any other

Well then, simple - don't read and don't bother adding more pointlessness.

 

[jb]

Well,

spent most of the weekend playing around with them and it "anit so bad". I have had the same amount of issues on with ATI drivers before and that was very managable. Live and learn I suppose

 

[Doomtrooper]

Like the link above states, unless texture sharpening and Anisotropic 1 is checked the IQ is lower, yet increased frames...

*sighs*

3dchips.net . . . right. You have read these 22 pages here? And you don't realise 3dchips is, like others before, mistaking the true default settings of Detonator40 with "defaulting to aniso 0°"?

Better have a look at the short article on 3DCenter. Be sure to check out the two registry keys updates. /shameless plug

ta,
-Sascha.rb

I read just fine, let me highlight the key points they discovered...Texture Sharpening is disabled by default and until that is checked Plus Aniso 1 IQ is much lower..I think they explain that quite well with screen shots.
When texture sharpening is enabled the score also drops back to the 29.17's range...go figure.

http://babelfish.altavista.com/urlt...file=article&sid=3223&lp=de_en&tt=url

After further bench mark with the new detonator 40 we always approach the thing. Meanwhile I could determine that 3DMark2001 in the standard attitude and Anisotropic Filtering results in auf stage "1" practically the same result as in the case of the old detonator without Anisotropic Filtering, this with same image quality. That it means that "1" really 1 does not represent, the circumstance would have had to be noticeable, there to Anisotropic Filtering in the case of the GeForcekarten for example with 2x, 4x or 8x Samples is adjusted. In the driver thus a nose is contained, which becomes, then it now clearly, in "stage zero" of any Filtering (Bilinear or trilinear) turns off, so that the textures appear rough. The fact that naturally thereby a performance increase develops is natural. With the "Texture Sharpening", as I assumed first erroneously, does not have to do it to anything, although we visually achieve here nearly identical values with the image quality as with the detonator 29,17 in the default value.

NVidia must thus again over the books. To the performance: If we let 3DMark2001 SE run once also and once without "Texture Sharpening", then this results in a difference, which lets aufhorchen: 8981 3DMarks instead of the 10900 or 10300 with the detonator 29.17. With Anisotropic Filtering "1", which none is, we get nearly the same values as with older detonatoren: 10471 3DMarks. This result led on the correct way.


So they are saying texuture sharpening is disabled by default and provide lots of screen shots to show the different modes..including mode 1 which is not point sampling.

 

[Galilee]

Doh, Texture Sharpening has always been off.

 

[walkndude]

I'm trying dave but its like passing a bad wreck on the road, damn hard not to look -especially when all the cars in front of you have jammed up the road doing the same.

Of course you could haul that wreck off and allow things to get rolling again if you wished, but have decided to let it lay.

 

[Doomtrooper]

Doh, Texture Sharpening has always been off.

Obviously not...

29.17 Drivers Default

40.41 Default

 

[Galilee]

The last one looks like Pointsampling. That is not default, that is 0X buggy slider.

Plz stop making a fool of yourself.

 

[nggalai]

Hi Doom,

please read that translation with a grain of salt. here's the original, I'll translate and comment on some passages.

"Nach weiteren Benchmarks mit dem neuen Detonator 40 kommen wir der Sache immer näher. Mittlerweile konnte ich feststellen, dass 3DMark2001 in der Standardeinstellung und Anisotropic Filtering auf Stufe "1" praktisch das gleiche Ergebnis wie beim alten Detonator ohne Anisotropic Filtering ergibt, dies bei gleicher Bildqualität. (1)Dass heisst, dass "1" nicht wirklich 1 darstellt, der Umstand hätte auffallen müssen, da Anisotropic Filtering im Falle der GeForcekarten beispielsweise mit 2x, 4x oder 8x Samples eingestellt wird. In dem Treiber ist also ein Bug enthalten, der, so wird es jetzt klar, in "Stufe Null" jegliches Filtering (Bilinear oder Trilinear) abstellt, sodass die Texturen grob erscheinen. Dass natürlich damit ein Performancezuwachs entsteht, ist selbstredend.(2) Mit dem "Texture Sharpening", wie ich zuerst irrtümlich angenommen habe, hat es nichts zu tun, obwohl wir visuell mit der Bildqualität hier fast identische Werte wie beim Detonator 29.17 in der Defaulteinstellung erreichen.

NVidia muss also noch einmal über die Bücher. Zu der Performance: Lassen wir 3DMark2001 SE einmal mit und einmal ohne "Texture Sharpening" laufen, so ergibt dies ein Unterschied, der aufhorchen lässt: 8981 3DMarks statt den 10900 oder 10300 bei dem Detonator 29.17. Mit Anisotropic Filtering "1", welches keines ist, bekommen wir fast die gleichen Werte wie bei älteren Detonatoren: 10471 3DMarks. Dieses Ergebnis führte auf den richtigen Weg.(3)"

(1) "By now I realised that 3DMark2001 at standard settings and with anisotropic filtering set to the "1" position delivers about the same quality like the older detonators without anisotropic filtering."

In other words, they realised that the "1" setting doesn't enable anisotropic filtering at all, but is--as has been discussed in this thread at length--AF DISABLED with bi- or trilinear filtering ENABLED. Unfortunately, they don't realise that that's to be expected of an AF setting of the first degree.

(2) "There's a bug in the drivers, and now it's finally clear that the 'zero setting' disables any filtering with the result, that the textures appear much coarser. This results, of course, in a performance boost."

Wow. They now learn that there's a bug in the Detonators40 which falls back to point sampling if you manually change to "0". *rolls eyes* That there actually is hardly any to NO performance boost between point sampling and bilinear filtering has been illustrated by Neeyik in this very thread, as well as on many other web sites all over the planet (some even reported a performance boost with bilinear enabled). Why do they guess wrong? because they actually compare point sampling performance to bilinear + texture sharpening performance, i.e. 2° anisotropic filtering.

(3) "With anistropic filtering set to '1' which doesn't appear to be AF at all, really, we achieved about the same results as with older Detonators: 10471. This setting shows the true performance."

They still don't understand what 1° anisotropic filtering means.

In short: 3DChips completely miss the point about the 0-setting bug in the Detonator40s. They disregard the fact that "0" means two things in this drivers release: with a default installation (and when hitting the "Defaults" button), it means "set by application", if you manually chose the 0-setting, it means "force point sampling."

Anyway. This discussion is rather pointless. If you insist on reading German articles, at least chose a source with reliable information.

ta,
-Sascha.rb

 

[Doomtrooper]

Ok..now Slider set to 1 and texture sharpening enabled...40.41

Are they wrong...??

 

[Doomtrooper]

nggalai,

Just using a translator...could you transalate this paragraph please..


NVidia must thus again over the books. To the performance: If we let 3DMark2001 SE run once also and once without "Texture Sharpening", then this results in a difference, which lets aufhorchen: 8981 3DMarks instead of the 10900 or 10300 with the detonator 29.17. With Anisotropic Filtering "1", which none is, we get nearly the same values as with older detonatoren: 10471 3DMarks. This result led on the correct way.

 

[Doomtrooper]

BTW this has nothing to do with Point Sampling, Noko noticed this also about texture sharpening having to be enabled and that is why this caught my eye...I'm more concerned about Texture Sharpening and they explain that in the paragraph..

 

[nggalai]

nggalai,

Just using a translator...could you transalate this paragraph please..


NVidia must thus again over the books. To the performance: If we let 3DMark2001 SE run once also and once without "Texture Sharpening", then this results in a difference, which lets aufhorchen: 8981 3DMarks instead of the 10900 or 10300 with the detonator 29.17. With Anisotropic Filtering "1", which none is, we get nearly the same values as with older detonatoren: 10471 3DMarks. This result led on the correct way.

Sure:

"NVidia has to go back to the drawing board. Regarding performance: if we chose to run 3DMark2001 SE with and withouth 'texture sharpening', the difference makes you point your ears: 8981 3DMarks instead of 10900 or 10300 with Detonator 29.17. With anistropic filtering set to '1' which doesn't appear to be AF at all, really, we achieved about the same results as with older Detonators: 10471. This setting shows the true performance."

The 8981 stem from aniso set to 1, texture sharpening enabled. This is equivalent to aniso set to 2°, no texture sharpening. the 10471 is aniso set to 1, no texture sharpening--i.e. bilinear, exactly what 3DMark2001 defaults to if you don't touch anisotropic settings in the buggy Detonator40 at all.

ta,
-Sascha.rb

 

[Doomtrooper]

Thx..

 

[nggalai]

You're welcome.

ta,
-Sascha.rb