Nvidia cheating in benchmarks?

[Indiana]

Hi,

I think the following links might be of interest:
http://www.3dconcept.ch/cgi-bin/show...how=1411&kt=3D (German)
http://www.matrox.com/mga/press_room..._bypassing.htm

Matrox have released a new tool that detects cheating of drivers of "some graphics card manufacturers" ;-)) by bypassing GDI calls.

NVidias drivers apparently fail on this tool, while 3DFx' don't...

[This message has been edited by Indiana (edited 26 August 2000).]

[rubank]

Cheating in 2D benchmarks!?!
How interesting.

My old 2mb Cirrus 5446 is fast enough for all practical reasons in 2D.

Matrox lends itself to ridicule.

rubank

[Indiana]

Originally posted by rubank:
My old 2mb Cirrus 5446 is fast enough for all practical reasons in 2D.

Matrox lends itself to ridicule.

No. I know the CL5446 myself, and while it was faster then the CL5428 (I think that was the number), I still preferred the 16Bit mode with it, cause in 32Bit-modes moving windows and especially scrolling were too slow for me... Yes, I love FAST scrolling.

You'd have a valid point if you had said that todays chips are fast enough in 2D.
And therefore I think it's not Matrox whose behaviour is ridicule/stupid but NVidias: Causing probable instabilities because of a speed-gain that is not needed and most probably not even noticed!

Anyway, this thread would better fit in the "General Hardware" forum. Sorry.

[PaulS]

That's real clever. Somebody takes the development time to figure out where a bottleneck is in the Windows code, is smart enough to code around it, and their competitors call foul. If they are bypassing parts of the GDI then they are not cheating on the benchmarks, the speed increases are real (though doubtless irrelevant, 2D speed being a non event these days).

I wish the Matrox development team would show this kind of initiative, we would probably have real dual head in W2k by now.

Paul

[mdhome]

Have anyone considered the fact that some companies may have already contemplated this approach but decided against it for *whatever* reasons. I dont know whether this bypassing of the GDI has been incorporated into live drivers but If it has, I cant see to many Nvidia users complaining.

On a different note, I dont really think this is Matrox's concern. Let them survive on the quality of what they produce.

regards MD

[Ashley]

I forget how long ago, but Matrox was "busted" by PC Magazine some time back for this kind of "cheating"...

[Himself]

Wasn't that ATi?

Frankly, when I had an NVIDIA card I would have random problems with 2D and DirectX, the amount of hacking and tweaking is overdone to the extreme, that's why I don't like their drivers, some part of the system was in partially finished hack mode depending on what driver set you were using. Now, if the tweaks were there but you could turn them off, that would be another thing, that's if they could still remember how to code within the lines.

As for Matrox making a big deal of this, they should maintain their silence until they have something other than a year old G400 to flog off. See that Q&A thread, Matrox comes off rather poorly, they are not even in the ballgame. If they had a conference room full of 16 year olds and the PR guy came in wearing a suit, showing slides and talking sales projections, it would be rather similar. "30-40 fps is enough for everybody", the guy is out of touch.

[Joel]

They do that to make their numbers look good in the reviews, thus making ppl think that it is the best card out. When in reality the card is not any better than the rest. But no matter how much GDI bypassing they do it will never solve their inferior graphics quality.

Joel

[Rob M.]

Are you sure it's nvidia who's doing this? When I first heard of it over at JC's page (Greebe, you're a superstar now ) I got my nvidia-empowered friend to try it out. He said everything passed.

[Joel]

Does it really matter who's doing it? Either way it's wrong and misleading.

Joel

[Humus]

That's real clever. Somebody takes the development time to figure out where a bottleneck is in the Windows code, is smart enough to code around it, and their competitors call foul. If they are bypassing parts of the GDI then they are not cheating on the benchmarks, the speed increases are real (though doubtless irrelevant, 2D speed being a non event these days).
I wish the Matrox development team would show this kind of initiative, we would probably have real dual head in W2k by now.

This is bad behaviour. Bypassing GDI causes the system to become unstable since you're confuses windows. The tool tells that the app enter a state many time, but never or only a few times exits this state. This makes windows collecting a lot of trash which never gets cleaned up, meaning system runs out of resources and requently crashes.


[This message has been edited by Humus (edited 27 August 2000).]

[Rags]

GDI hooks are a cheap way of making a driver less cpu intensive. This isn't just for 2D, it can be for 3D as well. Himself is correct, bypassing the GDI can cause resource leaks. GDI hooks are generally bad practice, and is frowned upon by microsoft when doing WHQL certs.

Rags

[Pace]

Is GDI bypassing possible in Win2k? I ask this because I recently installed Detonator 3 drivers (618 final) on my Win2k machine and BOOM! No bootup - I had lots of beta stuff on it etc (inc SP1b) but this was the last straw.

Rather than muck about trying to get it working again I reinstalled. First up was the 618 drivers - Boom! Not as big a boom as before but almost critical I'd say - drivers are an area where I believe Win2k is still very vulnerable - so that if you say install a driver which is very dodgy Win2k will warn you then install it.

I don't think applications etc can cause as much damage as drivers - anyway - back to the question - do the D3 drivers bypass the GDI in Win2k (causing my instability)?

Paul.

PS - Rags or anyone correct me if my beliefs are wrong!

[This message has been edited by Pace (edited 27 August 2000).]

[Salmonius]

I must be a psychic, as soon as I knew abou the test I had the feeling that nvidia would fail.

Although it seems fun nvidia can increase the speed of 2D no matter how, I think that creating going above the specs of the OS like that is bad practice and not very professional. What happens if MS decides to change the way that GDI calls are made internally? You do an innocent Windows Update that updates a bunch of .dlls and then your machine doesn't work anymore and you think it's your fault and that you must have screwed up in some way.

MS assures its clients of the interface to the GDI and what it will do, and the OEMs of the interface to the drivers that the GDI will be using, the rest is up to them and only them.

Besides, the GDI offers a lot of functionality that are not used by many people. I wonder if nvidia breaks some of that functionality. Personally though, I think that the GDI is full of crap, so it makes me laugh that nvidia did that and I wonder how much of a difference it makes.

[kjliew]

I don't know how many of you have gone through the Millenium/Mystique era or simply know about Matrox G-series. If I remember correctly, Matrox is the first to pioneer the GDI bypass technique and later followed by Diamond. Back in 1995-96, PC Magazine reported this trick and it was believed that GDI bypass can give good benchmark numbers on Winbench96 or any similiar benchmarks which time GDI calls. But it doesn't show any real benefits in real world application. Since then, Winbench97 had been designed to based on real world application usage, and hence had negligible impact on GDI bypass. No GDI bypass display driver will get through WHQL certification.

If you still have the old Millenium/Millenium II or Mystique170/220, you can download previous Matrox Powerdesk to have GDI bypass display driver. All Powerdesk prior to version 3.70 have a Powerdesk option called "PowerGDI". It's enabled by default. If you have "ShowMeAllMatrox" set to 1, you can also had options to apply GDI bypass on different GDI drawing calls.