Warp wrote:
>   And this isn't illegal? Doesn't it fall into something like
> http://en.wikipedia.org/wiki/False_advertising

Depends what you advertise, I suppose. "Passed the WHDL tests" is 
correct, after all. :-)

But yah, I guess if you really wanted to sue someone, you'd have a case. 
You might not win, but you'd have a case.

-- 
   Darren New / San Diego, CA, USA (PST)
     Remember the good old days, when we
     used to complain about cryptography
     being export-restricted?

Post a reply to this message

"Warp" <war### [at] tagpovrayorg> wrote

>   And this isn't illegal? Doesn't it fall into something like
> http://en.wikipedia.org/wiki/False_advertising

It's a complicated matter. Systems tailoring their behaviour to running
applications can have a perfectly valid reason. Sometimes, OS's and drivers
will do that to enable older applications that used, let's say,
unconventional or undocumented behaviour, to keep functioning. MS goes to
great lengths to accomodate badly written applications to maintain backwards
compatibility. Graphics card drivers work closely with major games and CAD
application vendors to maximize their performance, which benefits both
parties. So, a driver reconfiguring itself to maximize performance of CATIA
is not much different than a driver doing the same for 3DMark. Why should a
graphics driver optimized for many such applications let itself be
underspec'd through a synthetic benchmark, if it can help it?

Post a reply to this message

somebody <x### [at] ycom> wrote:
> So, a driver reconfiguring itself to maximize performance of CATIA
> is not much different than a driver doing the same for 3DMark. Why should a
> graphics driver optimized for many such applications let itself be
> underspec'd through a synthetic benchmark, if it can help it?

  Performing application-specific optimizations which result in the exact
same images being produced faster is completely different from producing
images of lower quality faster.

  You'd agree that if the driver detects that 3DMark is being run and
then just presents a black screen and gives a score of 10000 frames per
second, that's not an acceptable result. The speed is staggering, but
obviously unacceptable because the visual result is not acceptable.

  Why would the driver artificially reducing the number of polygons (thus
resulting in a poorer-quality image) just to get a higher benchmark score
be any more acceptable than the black-screen extreme example?

  Or think about it like this: Imagine that a benchmark tests how fast
the graphics card can render using 4x antialiasing. The driver detects
this and artificially turns on 2x antialiasing instead, in order to score
better. This is obviously unacceptable.

-- 
                                                          - Warp

Post a reply to this message

Warp nous apporta ses lumieres en ce 2007/10/29 19:16:
> Alain <ele### [at] netscapenet> wrote:
>> To get the real results, you need to use demos and benchmarks by independant
>> devlopers, and hope that your drivers don't have special codes that lowers the
>> settings like resolution, polygon count and colour depth, to artificialy make
>> the benchmark run faster. nVidia DID use that dirty trick, and there is no proof
>> that ATI used it or not at this time.
> 
>   You mean that nvidia deliberately put some code in their display driver
> that detected that you are running, for example, 3DMark, and if so it has
> a special code which converts the triangle meshes given by 3DMark to the
> driver to ones with less polygons so that the graphics card will appear
> to run faster than it really would?
> 
>   I suppose there's no way for a program to detect such trickery...
> 
What it does is reset the benchmark parameters to some lower ones when the 
actual benchmark starts, then set them back to the selected values on exit.
Result: you get a display that say that you used parameter set X, but the test 
was performed with parameters X/2.
Things that where done:
- changing the colour depth from 32 to 16 bits.
- changing the detail level from "High" to "Medium" or even "Low".
- turning fog off.
- reducing the Z-buffer depth.
- reducing the particles count.
- turning antialiasing off.
- reducing textures quality from "High" to "Medium", "Normal" or "Low".
- turning 3linear interpolation off.
- reduce the lights number.
...and some others...

You mention 3DMark, and it's exactly the first they targetted. They got caught. 
They got sued, and LOST. 3DMark coutered by using name randomization of the 
programm and all of it's components, folders and registry entries. Then every 
other benchmarks makers followed. Net result: nVidia's results droped, sometimes 
by an order of magnitude, around 10 to 30 times slower in some cases!
Think about boasting a 240+ FPS in a mark, then seeing that drop to around 12~17 
FPS after your hack got negated!

-- 
Alain
-------------------------------------------------
WARNING: The consumption of alcohol may make you think you are whispering when 
you are not.

Post a reply to this message

Alain wrote:

> They will probably install and run, but also, probably, with deliberately
> cripled performances.
> A dirty trick by those making the demos (independent from where they 
> come from):
> Let install on the concurent maker's card, but disable some features and 
> replace
> others with some DEoptimised versions that deliberately run slower.

Yeah, that wouldn't surprise me.

> To get the real results, you need to use demos and benchmarks by 
> independant
> devlopers, and hope that your drivers don't have special codes that 
> lowers the
> settings like resolution, polygon count and colour depth, to artificialy 
> make
> the benchmark run faster. nVidia DID use that dirty trick, and there is 
> no proof
> that ATI used it or not at this time.

Um... lower the polygon count? That sounds like a *highly* nontrivial 
task. Arguably more work than actually rendering all the polygons. :-P

Post a reply to this message

Orchid XP v7 <voi### [at] devnull> wrote:
> Um... lower the polygon count? That sounds like a *highly* nontrivial 
> task. Arguably more work than actually rendering all the polygons. :-P

  Not if the driver has code specifically designed for the specific
triangle mesh the benchmark program send it. It can simply have some
kind of lookup table or something which substitutes some of the triangles
with other (fewer) triangles.

-- 
                                                          - Warp

Post a reply to this message

Warp wrote:
> Orchid XP v7 <voi### [at] devnull> wrote:
>> Um... lower the polygon count? That sounds like a *highly* nontrivial 
>> task. Arguably more work than actually rendering all the polygons. :-P
> 
>   Not if the driver has code specifically designed for the specific
> triangle mesh the benchmark program send it. It can simply have some
> kind of lookup table or something which substitutes some of the triangles
> with other (fewer) triangles.

...hmm, so *that* is why the nVidia video driver is over 200 MB...

(Seriously, are you NUTS?! Most drivers are about 20 KB.)

Post a reply to this message

Darren New wrote:

> Yep!  And you wonder why video drivers under Windows crash the system so 
> often.

Uh... do they?

Post a reply to this message

Orchid XP v7 <voi### [at] devnull> wrote:
> (Seriously, are you NUTS?! Most drivers are about 20 KB.)

  You can fit quite a lot of things into 20 kB. Just take a look at
some good 4 kB demos.

-- 
                                                          - Warp

Post a reply to this message

Orchid XP v7 wrote:
> Darren New wrote:
> 
>> Yep!  And you wonder why video drivers under Windows crash the system 
>> so often.
> 
> Uh... do they?

Well, relatively often, yes, compared to most other stuff.

-- 
   Darren New / San Diego, CA, USA (PST)
     Remember the good old days, when we
     used to complain about cryptography
     being export-restricted?

Post a reply to this message