nVidia: Good Drivers finally

[dragonic]

Well Extremetech posted new article of 44.67 Detonators and they bring more better performance in 3dMark03 (1000 Marks)

But here is the nice part

After we tested the two drivers, we fired up the developer version of 3DMark03, ran Game Test 4 ? Mother Nature, and enabled the free-camera mode. We wanted to see whether the same corner-cutting we observed with 44.03 was still present in this new driver. It wasn't. As far as we can tell, nVidia achieved these impressive performance gains without any "on-rail" optimizations.

Apparently nVidia has optimized their drivers properly this time

Link

EDIT: The link works now =)

 

[Regs]

I don't believe in nVidia drivers anymore. nVidia + 3dmarks = bs.

 

[GtPrOjEcTX]

your link is the suck

 

[dragonic]

well the link is fixed now

 

[Jeff7]

Great. Do the optimizations apply elsewhere, or are they limited to just that game? I wonder if they did something like what...was it ATI that did it? - turn off anisotropic filtering to get higher framerates in certain games/apps?

 

[1ManArmY]

What 3D Mark 03 a GAME! Say it isn't so

 

[Twista]

They suck! All they do is make your gfx quality lower so you can get higher fps which i got higher fps, but my game looked fu*k up. In quake 3 arena 125fps is all you need and with theses drivers i got 125 steady but than again my game looked differnt. Now that im on the older ones i get 85fps and my good looks better. Oh well i should took a screenshot so i could so you the differnces.

/edit. Sorry, im in a rush so i had to type fast.

 

[Blastman]

Nvidia has already come out with the 44.65?s which re-enabled the cheats and defeat Futuremarks anti-cheat patch 330.

I see where this is going ? .

Nvidia ? ? ? ? cheat driver 43.03
Futuremark ? ? anti-cheat patch 330
Nvidia ? ? ? ? cheat re-enabler driver 44.65
Futuremark ? ? anti-anti-cheat patch 340
Nvidia ? ? ? ? cheat re-re-enabler driver 46.0
Futuremark ? ? anti-anti-anti-cheat patch 350
Nvidia ? ? ? ? cheat re-re-re-enabler driver 45.0
Futuremark ? ? anti-anti-anti-anti-cheat patch 360
????
????..
?????

 

[JBT]

Originally posted by: Blastman
Nvidia has already come out with the 44.65?s which re-enabled the cheats and defeat Futuremarks anti-cheat patch 330.

I see where this is going ? .

Nvidia ? ? ? ? cheat driver 43.03
Futuremark ? ? anti-cheat patch 330
Nvidia ? ? ? ? cheat re-enabler driver 44.65
Futuremark ? ? anti-anti-cheat patch 340
Nvidia ? ? ? ? cheat re-re-enabler driver 46.0
Futuremark ? ? anti-anti-anti-cheat patch 350
Nvidia ? ? ? ? cheat re-re-re-enabler driver 45.0
Futuremark ? ? anti-anti-anti-anti-cheat patch 360
????
????..
?????

lol so true

 

[Rand]

Originally posted by: dragonic
Well Extremetech posted new article of 44.67 Detonators and they bring more better performance in 3dMark03 (1000 Marks)

But here is the nice part

After we tested the two drivers, we fired up the developer version of 3DMark03, ran Game Test 4 ? Mother Nature, and enabled the free-camera mode. We wanted to see whether the same corner-cutting we observed with 44.03 was still present in this new driver. It wasn't. As far as we can tell, nVidia achieved these impressive performance gains without any "on-rail" optimizations.

Apparently nVidia has optimized their drivers properly this time

Link

EDIT: The link works now =)

The absence of any graphical issues when diviating from the set camera path only implies a lack of static clipping planes, which was but one of a large number of issues.
Naturally I can't say for certain whether nVidia's optimizations are legitimate or not, by ExtremeTech's article does little to answer the question, they indicate no in-depth tests at all besides looking into the clipping issues.
Image quality was not looked into, nor were any of the other questionable optimizations specifically addressed.

The lack of a performance gain in any real world games, or any recorded benefit in the synthetic PS/VS 2.0 tests recorded on B3D's forum via ShaderMark and D3D Rightmark would seem to make the gain slightly questionable.

Personally I would tend to hold any nVidia 3DM03 results with or without the 330 3DM patch until we see a specific analysis of the drivers and their impact upon 3DM in respect to all known cheats.
Formerly anti-detector could be applied to accomplish this task, but that's no longer viable as the 44.67's have incorporated run-time decryption of the driver to prevent any outside application from easily specifically disabling any application detection techniques within the driver.

 

[Blastman]

If you compare the scores with the 44.67 with the 44.03, the scores have jumped back to what they were test for test before Futuremark?s patch. Which basically means NV re-enabled the cheats.



extremetech

extremetech

 

[AtomicDude512]

Originally posted by: JBTele

Originally posted by: Blastman
Nvidia has already come out with the 44.65?s which re-enabled the cheats and defeat Futuremarks anti-cheat patch 330.

I see where this is going ? .

Nvidia ? ? ? ? cheat driver 43.03
Futuremark ? ? anti-cheat patch 330
Nvidia ? ? ? ? cheat re-enabler driver 44.65
Futuremark ? ? anti-anti-cheat patch 340
Nvidia ? ? ? ? cheat re-re-enabler driver 46.0
Futuremark ? ? anti-anti-anti-cheat patch 350
Nvidia ? ? ? ? cheat re-re-re-enabler driver 45.0
Futuremark ? ? anti-anti-anti-anti-cheat patch 360
????
????..
?????

lol so true

nVidia stinks lately, In the article about them bullying the Omega guy about his awesome drivers it makes me sick. They need to exist for one reason, to bring ATI card prices down.

 

[NEVERwinter]

actually... i prefer to see the improvements on games fps rather than in 3dmark only

 

[nemesismk2]

Originally posted by: Blastman
If you compare the scores with the 44.67 with the 44.03, the scores have jumped back to what they were test for test before Futuremark?s patch. Which basically means NV re-enabled the cheats.

extremetech

extremetech

Where is your proof that Nvidia is cheating with the 44.67 drivers?

I want to see hard facts not some as$ talk from you, anyway even Futuremark has said it's not cheating it's optimisation.

In the Extremetech article it said "After we tested the two drivers, we fired up the developer version of 3DMark03, ran Game Test 4 ? Mother Nature, and enabled the free-camera mode. We wanted to see whether the same corner-cutting we observed with 44.03 was still present in this new driver. It wasn't. As far as we can tell, nVidia achieved these impressive performance gains without any "on-rail" optimizations."

 

[Rankor]

Has anyone heard of problems with these drivers, the FX 5900 Ultra on Star Wars Galaxies?

 

[Blastman]

I want to see hard facts ?

nemesismk2 are you going let NV dupe you again so easily? There is evidence ?

1) scores have jumped back to what they were test for test before Futuremark?s patch. Let me get this straight. We?re suppose to think these are legitimate optimizations when test for test the scores are back to what the cheats were!!!???? Reality check here.

2) In the other 3 games Extremetech tested there was no performance gain whatsoever. Nothing.

anyway even Futuremark has said it's not cheating it's optimisation.

Basically FM was threatened with a lawsuit by Nvidia and they unfortunately capitulated in their wording about using the .... ?cheat? ... word under threat. This does not change the facts of the situation. Read this thread and WaltC?s post ? beyond3d

? I disagree with your opinions about "cheating" versus "optimization." The differences in the English language are clear and profuse. When nVidia chopped down the benchmark workload by implementing clip planes, unrendered frame segments, etc., they were clearly "cheating" and not "optimizing" at all. I would not characterize everything nVidia did as a cheat relative to the benchmark, but these things are clearly cheats--the word "optimize" is a gross misdefinition when applied to them?.

I think we have a right to view these gains with extreme skepticism. Just because Extremetech didn?t find the cheats doesn?t mean they aren?t there. Nvidia is just hiding the cheats a little better this time.

 

[nemesismk2]

If Futuremark had optimised their software maybe Nvidia wouldn't have to find ways to optimise their drivers to improve performance. Let's be honest here Futuremark aren't exactly well known for their highly optimised code, 3dmark03 is badly written, inefficent and is nothing more than a collection of poorly designed benchmarks with no real value.

IMO what Nvidia should do is buy Futuremark and then concentrate on optimising their drivers for games and not 3dmark03!

Hey Nvidia have I done enough to get a FX5900 Ultra yet?

 

[nRollo]

LOL
All this wailing and gnashing of teeth over meaningless cheats on a meaningless synthetic benchmark.
"Waaaaaaaaa! Whiiiiiiiiinnne! nVidia is CHEATING! ATI has in the past as well, but I forgive that because I have an ATI card now!Waaaahhhhhh! I CERTAINLY wouldn't have bought a card by the CHEATER company!"
LMFAO

 

[Blastman]

Originally posted by: nemesismk2

Hey Nvidia have I done enough to get a FX5900 Ultra yet?

ROFL.

 

[Cat]

Do any of you do graphics programming for a living? Probably not. I do, however, and I can say one thing for certain:

Imagine you had Doom 3 running a timedemo, and NVidia enabled user clip planes for the demo. Moving off the demo's path would produce visual errors. Then everybody complained, because this was clearly not fair in terms of benchmarking. Then let's suppose NVidia's next drivers let you move around the level with the same FPS, only without error. Do you complain? No. However, if the NVidia programmers specifically provided a special case for that one Doom level, you'll be disappointed when the results don't translate to e1m2 But, if the drivers are able to analyze arbitrary levels beforehand, you'll be happy, and so will I


Take what you will from this statement.

 

[Blastman]

Originally posted by: Cat

But, if the drivers are able to analyze arbitrary levels beforehand, you'll be happy, and so will I

You can?t, that?s what makes them ? ?arbitrary?. Since I?m running around shooting bad guys/monsters in a completely arbitrary fashion, there is no way to know beforehand exactly what explosions, blood walls ?etc. need to be rendered.

 

[Pete]

Originally posted by: nemesismk2
If Futuremark had optimised their software maybe Nvidia wouldn't have to find ways to optimise their drivers to improve performance. Let's be honest here Futuremark aren't exactly well known for their highly optimised code, 3dmark03 is badly written, inefficent and is nothing more than a collection of poorly designed benchmarks with no real value.

Proof, please.

 

[Cat]

Originally posted by: Blastman

Originally posted by: Cat

But, if the drivers are able to analyze arbitrary levels beforehand, you'll be happy, and so will I

You can?t, that?s what makes them ? ?arbitrary?. Since I?m running around shooting bad guys/monsters in a completely arbitrary fashion, there is no way to know beforehand exactly what explosions, blood walls ?etc. need to be rendered.

You can, if the levels contain enough information. Doom's shaders are not hard coded, they're simple text files parsed on level loads. The models are loaded into vertex buffer objects, which have known size on creation, etc.

The Nature test apparently lets you run around and do what you want, so either NVidia has very impressive cheats going on, or they've actually optimized some stuff, perhaps the way ATi did, by rewriting shaders such that they produce identical output but at a lower cost. Just like compilers reorder instructions based on pipeline stalls/pairing, etc. Like I said, you don't do graphics programming for a living, but I do.

 

[beserkfury15]

perhaps before everyone starts accusing/defending nvidia, how bout we wait a bit. in the last week, they've realeased like 4nonofficial drivers. i'm guessin an official version is coming soon, seeing how the 44.03 is like 2months old. if nvidia doesn't improve then fine, if they do then...

 

[Blastman]

Originally posted by: Cat

The Nature test apparently lets you run around and do what you want,

Not in the benchmark. Or, you can?t change what the benchmark runs, otherwise how would it be a comparable benchmark to the last run? You have to render the same things every time for it to be proper benchmark.

Originally posted by: Cat

Doom's shaders are not hard coded, they're simple text files parsed on level loads. The models are loaded into vertex buffer objects, which have known size on creation, etc.

So. Knowing what shader routines you?re going to run doesn?t mean you know what has to rendered ie ? that arbitrary information coming down the pike that?s going to be rendered. To use an analogy, you may know the general type of stuff that will be rendered (like say some numbers) so you can optimize for that, but you still don?t know what numbers exactly will have to be rendered.