nVIDIA's Cheating Drivers Have Been Ranked WHQL
- Thread starter David G.
- Start date
There are plenty of ongoing threads in which this could've been posted.
Written in a PM to me by David G."
Thread re-opened.
Thread re-opened.
http://www.microsoft.com/whdc/hwtest/device/default.mspx?area=displayadapter
There's a starting point, though I think people like OpenGl_Guy or andypski could help us all by boiling it down.
That being said, my understanding is Microsoft is much more concerned about system stability than poor rendering.
There's a starting point, though I think people like OpenGl_Guy or andypski could help us all by boiling it down.
That being said, my understanding is Microsoft is much more concerned about system stability than poor rendering.
Look at the following 5900 review...
http://www.digit-life.com/articles2/gffx/5900u.html
Notice how rightmark shows the 5900 being slaughtered by the 9800Pro, but when common benchmarks are used the 5900 excels.
http://www.digit-life.com/articles2/gffx/5900u.html
Notice how rightmark shows the 5900 being slaughtered by the 9800Pro, but when common benchmarks are used the 5900 excels.
First of all , thanks John .
Now ... wait a minute , of course stability should be , probably , the main issue for an OS maker along with compatibility BUT , IQ isn't comprised in that certification suite ?!
Then , why do they call it "High Quality" ? "WHSL - Windows High Stability Labs" would sound more natural ...
Now ... wait a minute , of course stability should be , probably , the main issue for an OS maker along with compatibility BUT , IQ isn't comprised in that certification suite ?!
Then , why do they call it "High Quality" ? "WHSL - Windows High Stability Labs" would sound more natural ...
From their standpoint: OS that doesn't crash == high quality. Video card that renders poorly == your bad luck.
crystalcube
Newcomer
David G. said:
I dont think IQ is any concern for an OS developer/vendor because its a subjective thing but functinal accuracy must be tested. Imagine a video driver rendering colour green instead of blue, we will have a beautiful green sky
and we will have extremely stable but highly inaccurate driver. So the question we should be asking is how does this driver which is not as accurate as it should be became WHQL certified.
crystalcube said:
And the answer would be: the same way every driver becomes WHQL certified: by passing the test.
I am very confused by this WHQL certification...
In the past, other IHV's have failed WHQL certification by using same-mipmap sampling for trilinear as Microsoft has a number of kludged mipmap tests that put actual data into different mipmap levels, then inspect for "true" trilinear implementation by looking for this data scaled in each mipmap level.
An example was- early trilinear implementations used a 4x4 block in the current mipmap to signify a 2x2 block in the next mipmap... but because Microsoft actually kludged different data in the different mipmaps, it wasnt able to find this data. I would think any form of texture mapping "cheating" would perturb their data to the degree of non-recognition and thus fail the WHQL process, much like the failed IHVs in the past... Unless they have suddenly opted to remove this process for whatever reason.
In the past, other IHV's have failed WHQL certification by using same-mipmap sampling for trilinear as Microsoft has a number of kludged mipmap tests that put actual data into different mipmap levels, then inspect for "true" trilinear implementation by looking for this data scaled in each mipmap level.
An example was- early trilinear implementations used a 4x4 block in the current mipmap to signify a 2x2 block in the next mipmap... but because Microsoft actually kludged different data in the different mipmaps, it wasnt able to find this data. I would think any form of texture mapping "cheating" would perturb their data to the degree of non-recognition and thus fail the WHQL process, much like the failed IHVs in the past... Unless they have suddenly opted to remove this process for whatever reason.
David G. said:
How is it possible to write a test for a 3DMark specific "optimization" without that test being 3DMark?
nVidia will have some trigger which tells them they're running 3DMark. In the past people have used the filename of the exe (Quack), now I expect thay use thing like (off the top of my head) a checksum of the vertex data. Unless you know that trigger, and how to fool it, you can't test that path through the code.
WHQL tests both stability and conformance to rasterisation methodology as defined in the API specification and implemented in teh reference rasteriser, its not necessarily perfect at this but it still covers a lot of ground. However, it can not check for application specific optimisations, it would have include teh app itself to do that.
John.
John.
