NVIDIA's detailed slam of 3DMark03

[John Reynolds]

You know, I've argued in circles with certain webmasters on the very heavy use of 3DMark over the years. I've sat back and been quite puzzled watching people switching driver versions, looking for the one that'll give them the highest 3DMark score. I agree with what Nvidia has written on the newest version, too.

But, as others have said, the timing of this stinks. And which sites fall in line with Nvidia's new tack on "synthetic" benchmarks (cough, Treemark, cough). . .that's going to be more interesting to me than anything else. I expect a company to act in whatever will be its best interests, but I don't expect so-called web journalists to tow whatever line is dictated to them from a certain IHV.

[Nemesis77]

In short:

Previous 3DMarks supported NVIDIA-features, so they supported them

3DMark03 supports features that mainstream NV-cards (GF3, GF4 etc) do not support (PS 1.4), therefore new 3Dmark is "not accurate display of gaming performance"

If NV doesn't support PS 1.4 (like Ati does), then it really is NV's problem, not Futuremarks! And the fact that NV started whining about it, is just pathetic. Should Futuremark cripple their benchmark just so it would fit more nicely to NV's feature-set?

If NV supports using real games for benchmarking instead of 3DMark, why didn't they whine when previous 3DMarks were released?

[creedamd]

Originally posted by Nemesis77
In short:

Previous 3DMarks supported NVIDIA-features, so they supported them

3DMark03 supports features that mainstream NV-cards (GF3, GF4 etc) do not support (PS 1.4), therefore new 3Dmark is "not accurate display of gaming performance"

If NV doesn't support PS 1.4 (like Ati does), then it really is NV's problem, not Futuremarks! And the fact that NV started whining about it, is just pathetic. Should Futuremark cripple their benchmark just so it would fit more nicely to NV's feature-set?

If NV supports using real games for benchmarking instead of 3DMark, why didn't they whine when previous 3DMarks were released?

awesome summary, how others spin this is unbelievable. Just blind I guess.

[digitalwanderer]

Originally posted by scott123
Nvidia may not have market share this time next year, but when the leading graphics card maker bows out, and doesn't sanction your benchmark (along with some leading web-sites), you have problems. I think Futuremark has problems right now.

Scott

Very nicely put, and very truthful. nVidia really does have a point in their argument, but their credibility is pretty shot right now.

[silence]

Originally posted by digitalwanderer
Very nicely put, and very truthful. nVidia really does have a point in their argument, but their credibility is pretty shot right now.

yup.....there are few points in what nvidia says,their main problem with this is not futuremark itself.....it's all PR BS lately coming from their direction which made ppl discard everything coming from nvidia.

the hole they digged themselves......and still digging.

[MikeC]

For what it's worth...

When I discussed 3DMark03 with NVIDIA, their main concern was with the method games 2 and 3 used to calculate shadows. They documented the program code used in 3DMark03 and discussed its inefficiency.

They also provide examples of graphics artifacts that occurred as a result of the shadow algorithm.

Since I'm not a developer, I can't comment on if the code is inefficient or not. But it is documented.

There was no discussion concerning pixel shaders.

[Nemesis77]

Originally posted by digitalwanderer
Very nicely put, and very truthful. nVidia really does have a point in their argument, but their credibility is pretty shot right now.

So people who make 3D-benchmarks should only make benchmarks that are solely targeted at NV-hardware? If some "other" 3D-companies have more advanced features, they must not be supported so that almighty NV doesn't get shown in bad light?

"We could use PS 1.4, it is better choice after all..."
"Forget it! ati supports it, but NV doesn't so that's a big no-no!"

Why should Futuremark optimize for NV? If they do that, they are screwing Ati? isn't it just smart to use the best tech available? With Ati, 3DMark uses PS 1.4. NV doesn't support it, so it defaults to lesser PS (and takes the performance-hit with it).

[saturnotaku]

Originally posted by PsychoSy
I agree.

Since saturnotaku's not around at the moment, I'll go ahead and recite "The Script" in his absense: You can't FRAG with 3DMark!.

3DMark is something that I use once in blue moon just to get a general ballpark figure. I certainly as hell don't accept their scores as if it were gospel. It's a benchmark untility - not a game. Too many people, however, have a fetish for the darned thing and the moment they lose 200+ points, they're ready to pop a few Prozacs and check themselves into a damned 12-step program.

That's it - 3DMark buffs need their own official brick and mortar support group.

Let's call it "L33t|\|355 4|\|0|\|y|\|\0u5" (that's "Leetness Anonymous" for all you folks that require a translation - I'm not a leetspeek wiz by sense of the word so I probably butchered the hell of that attempt anyway).

I can see it now...

""Hi...name's Dave...I used to be supaleet..."

HI, DAAAAAAAVE!!!

"Thanks...uh, I lost 748 points in 3DMark today and I'm depressed. Hey Carmack, pass me one of those Ativans and a Zoloft, too, bro!"

Good grief!

Couldn't have said it any better myself. I won't argue with NVIDIA's credibility going down the crapper, but I'm not about to get my undies in a bunch over what some people say about a synthetic benchmark.

[creedamd]

I don't understand the fuss, the Gffx is neck and neck with the 9700pro, just as most other benchmarks, I think the problem is that the GFFX is a pipedream now, and the separation between the g4ti4600 and the 9700pro is really starting to show, not just benchmarks but real games as well. The graphics on 3dmark03 is simply unbelievable, I can only hope that a game comes out soon that is half as good.

It is meant to bring your system to a crawl. I tweaked my system quite a bit using it and it showed in real gaming, for what it's worth.

[jbirney]

Originally posted by MikeC
For what it's worth...

When I discussed 3DMark03 with NVIDIA, their main concern was with the method games 2 and 3 used to calculate shadows. They documented the program code used in 3DMark03 and discussed its inefficiency.

FYI both games 2/3 make the most use of PS1.4

But HardOCP offers a reason why the shadows are slower:

This test does use the stencil buffer to perform shadow volumes "kind of like" Doom]|[. However, its method is not exactly the best way to do things. Their algorithm is very redundant, there is a portion called “Skin Object in Vertex Shader” which basically does the exact same skinning calculations over and over for each object. It is an inefficiency that would not be translated to a real world game such as Doom]|[. Because of the way their algorithm works the vertex pipeline becomes a huge bottleneck.

[John Reynolds]

Originally posted by jbirney
FYI both games 2/3 make the most use of PS1.4

But HardOCP offers a reason why the shadows are slower:

I"m sorry but the writers are [H] aren't technical enough to offer these explanations up on their own (which, yes, implies they're being spoon-fed from someone).

[creedamd]

Originally posted by John Reynolds
I"m sorry but the writers are [H] aren't technical enough to offer these explanations up on their own (which, yes, implies they're being spoon-fed from someone).

government/hangar18/conpiracy++