My thoughts on "Optimization"

[NightFlight]

Here's a good link with some good points on this issue....


http://www.elitebastards.com/page.php?pageid=1343&head=1&comments=1

[digitalwanderer]

Originally posted by NightFlight
Here's a good link with some good points on this issue....


http://www.elitebastards.com/page.p...ad=1&comments=1

Yeah, listen to it straight from a group of self-named "bastards"... ;) :p

[NightFlight]

Doesn't matter what their name is, still alot of good points in that article. As for me I just use Fraps and call it a day.

[Behemoth]

Originally posted by Sazar
so.. what was the logic of your previous post if you do not intend to go to an ati fansite ?

are you as bored as I am sitting here @ home sick :D

lol

:beer:
i really wanted to see how he would feel if nvidia-lovers do this same thing, which i think is stupid and childish, back to him.

[digitalwanderer]

Originally posted by NightFlight
Doesn't matter what their name is, still alot of good points in that article. As for me I just use Fraps and call it a day.

I was being sarcastic, I was on staff at EB up until yesterday. ;)

[zakelwe]

I'd far rather be a drunkenbastard than an elitebastard to be honest :D

Here's some examples of the GT1 game using 41.09 drivers and 42.68 drivers. This is not a clipping or precision bug, it's a non drawing bug -- 41.09 is really slow ( 128 fps ) and 42.68 is really fast ( 138fps ).



41.09 frame 440 (http://www.aocb77.dsl.pipex.com/250000440.jpg)

42.68 frame 440 (http://www.aocb77.dsl.pipex.com/260000440.jpg)

41.09 frame 830 (http://www.aocb77.dsl.pipex.com/250000830.jpg)

42.68 frame 830 (http://www.aocb77.dsl.pipex.com/260000830.jpg)

As I pointed out earlier Futuremark no longer mind these drivers, so they think of iot as a bug not a cheat.


Regards

Andy

[Sazar]

Originally posted by zakelwe
I'd far rather be a drunkenbastard than an elitebastard to be honest :D

Here's some examples of the GT1 game using 41.09 drivers and 42.68 drivers. This is not a clipping or precision bug, it's a non drawing bug -- 41.09 is really slow ( 128 fps ) and 42.68 is really fast ( 138fps ).



41.09 frame 440 (http://www.aocb77.dsl.pipex.com/250000440.jpg)

42.68 frame 440 (http://www.aocb77.dsl.pipex.com/260000440.jpg)

41.09 frame 830 (http://www.aocb77.dsl.pipex.com/250000830.jpg)

42.68 frame 830 (http://www.aocb77.dsl.pipex.com/260000830.jpg)

As I pointed out earlier Futuremark no longer mind these drivers, so they think of iot as a bug not a cheat.


Regards

Andy

I believed futuremark was going to sort the scores per WHQL and non-WHQL standings in order to separate the two...

dunno... perhaps Worm can enlighten us..

btw... for the record.. the older driver sets IQ is some of the worst I have seen... what exactly is it rendering ??

not much apparently..

[rokzy]

now that nvidia's finally got some DX9 drivers, when will they get on the Hall of Fame?

though the Hall of Fame seems quite useless atm; no FX cards at all, no 9800 or 9600 cards, and no differentiation between 9700 and 9500 cards...

[legion88]

Originally posted by Nv40
what i find really sad ,Joe.. its HOW fraudulent is this PS1.4 benchmark.
i find very hard to believe ,that you believe in the things you say.
you only focus you view in what Nvidia have done ,but cares nothing about the test at all. How unfair it is for Nvidia cards ,but still people like you with some "knowledge" still defend the test. everyones knows this ..
3dmark is not apples vs apples , but you point your finger to Nvidia
for not playing apples vs apples.. you complain for Doom3 benchamrks
but are VERY pleased with 3dmark2003. DONt you see the pattern here?
Take your glasses Dude.. look things from other points of view too..
I understand clearly all your points , but what you dont want to see
is the truth of 3dmark 2003 as a fair comparisons between NVidia and ATI cards ,and as a bench representative of Games of the FUture...

you already knows this ,that the test is already invalid ,
since one participant does not agree with their rules anymore .
BUt as you have said."THE only thing that matters is the scores " :rolleyes:

wait... it also matters that both cards do things in exactly in
the same way.. Ohh.. but thats impossible..
WHo cares.. "THE only thing that matters is the scores "

what is even Funny is that in games of the Future (Doom3)
using the best engine in the following years ,Nvidia have the fastest cards ..and trounce all ATI cards.. ppsss... in the real WOrld.


3DMark2003 does not include 'code' specifically to make selected video cards faster. That is something you want everyone else to believe for quite sometime, like for several months. I have not been paying attention to this forum for about a month or so (been somewhere else) but I see that the 'commentary' has not changed from you in regards to 3DMark.

3DMark2003 was designed from the ground up to be as close to an "apples-to-apples" comparison as it can be. That is why FutureMark did not include code like "if video card is an NVIDIA run this slower code instead or run this 'do more rendering work to make it look slower' code or run this 'do less rendering work to make it run faster' code" instead.

Games will often treat various cards different which is why they are not suitable for benchmarking. It is funny how the "real-world" crowd continue to fail to mention this fact when they propagandize the alleged superiority of "real-world" games over "synthetics" for benchmarking.

I can see websites wanting to use "real-world" games as it makes their reviews longer and, generally, more interesting. In other words, real-world games make the reviews look better (as in a better read) but it does not make the quality of the data that they provide better. The quality of the data is actually worse as the reviewer can not prove at all that the games are not selectively favoring one card over another.

3DMark2003 does not selectively treat NVIDIA cards different from ATI cards. 3DMark2003 treats all the cards the same. By definition (something you ignore with your commentary), this is an apples-to-apples comparison.

Cheating on the other hand can render the results invalid. In other words, cheating can make the results a non-apples-to-apples comparison. That is obviously not the fault of 3DMark2003 or FutureMark but the fault of the accused--in this case, allegedly NVIDIA.

[hithere]

Originally posted by legion88
3DMark2003 does not include 'code' specifically to make selected video cards faster. That is something you want everyone else to believe for quite sometime, like for several months. I have not been paying attention to this forum for about a month or so (been somewhere else) but I see that the 'commentary' has not changed from you in regards to 3DMark.

3DMark2003 was designed from the ground up to be as close to an "apples-to-apples" comparison as it can be. That is why FutureMark did not include code like "if video card is an NVIDIA run this slower code instead or run this 'do more rendering work to make it look slower' code or run this 'do less rendering work to make it run faster' code" instead.

Games will often treat various cards different which is why they are not suitable for benchmarking. It is funny how the "real-world" crowd continue to fail to mention this fact when they propagandize the alleged superiority of "real-world" games over "synthetics" for benchmarking.

I can see websites wanting to use "real-world" games as it makes their reviews longer and, generally, more interesting. In other words, real-world games make the reviews look better (as in a better read) but it does not make the quality of the data that they provide better. The quality of the data is actually worse as the reviewer can not prove at all that the games are not selectively favoring one card over another.

3DMark2003 does not selectively treat NVIDIA cards different from ATI cards. 3DMark2003 treats all the cards the same. By definition (something you ignore with your commentary), this is an apples-to-apples comparison.

Cheating on the other hand can render the results invalid. In other words, cheating can make the results a non-apples-to-apples comparison. That is obviously not the fault of 3DMark2003 or FutureMark but the fault of the accused--in this case, allegedly NVIDIA.

I think there's a misconception here: Users aren't benchmarking a card alone when they use "real-world" benchmarks (games). You contend that games treat different cards differently, therefore there can be no valid comparison. But what if the user doesn't care which card is "superior"? All the user wants to know is which card will deliver the best performance in his/her favorite game, no matter how that game "treats" either manufacturer.

The 3dMark thing is a cheat. The benchmark depends on being run as if it were a normal game in order to facilitate comparison. It is irrelevant that the cheat is invisible, it still destroys the integrity of the test, the benchmark, and that of the driver coder who facilitated this nonsense. It is irrelevant if you think it is an invalid benchmark: Nvidia thinks it's valid enough to set coders to work putting in clipping planes in complete disregard to the integrity of the test. Instead of offering enhancements to drivers that enable a faster run of the test in a legitimate manner, they have endeavored to circumvent the spirit of the benchmark.

I don't care how any of you want to sugar-coat it. The fact is, they turned a fully realized 3-d test environment into a rendering on rails to inflate their score, and that sucks.

Optimization: Enables a higher score/smoother performance while maintaining the credibility of the test/ viability of the gameplay experience.

Cheat: Enables a higher score/smoother performance in a benchmark/timedemo at the cost of the credibility of the test or the viability of the gameplay experience, usually coupled with secrecy; sometimes referred to as an extremely fortunate "bug"
ex: Clipping planes that just so happen to follow the camera around in a rail-guided benchmark, and continue to follow the preset path of the camera even when the user takes the camera off the rails.

[legion88]

Originally posted by hithere
I think there's a misconception here: Users aren't benchmarking a card alone when they use "real-world" benchmarks (games). You contend that games treat different cards differently, therefore there can be no valid comparison. But what if the user doesn't care which card is "superior"? All the user wants to know is which card will deliver the best performance in his/her favorite game, no matter how that game "treats" either manufacturer.

(snip...)


If users don't care which card is "superior", then they are obviously not benchmarking the card. They are looking for which card is best for which game. Such a user must have money. Card A is best for Game X while Card B is best for Game Y (where best means fastest performance). Right off the bat, users need two cards.

Real users don't want to keep switching cards based on what game they happened to be playing at the time. So they need an apples-to-apples comparison, a benchmark that is not biased.

[Zurble]

Optimization is when you can make it faster without changing the specifications of the interface. This is internal saucage. It is usually done within a routine independantly from the program using it. Any program using it will benefit from it, and DX9 specs or opengl specs are not card dependant.

If the output is different (noticiable or not), it is not optimization, it is cheat.
Many people don't see the difference between 1280 or 1600, if your card decide to render 1280 instead of 1600 it is cheat, noticiable or not. There is a lot of thing people don't notice (Aniso level, AA type, ....) but using something else because it is not noticiable (who decide?) is blatant cheat.

In many response i see "3dmark isn't a game...it is not important" but do you all realize that EVERY game that is used as benchmark in reviews are cheated by both ATI and NVIDIA since the begining.
Texture filtering are modified, AA and AF routines are tweaked, timedemos are changed.....Quake 3 is displaying 200 fps, but it is actually runing at 80 but nobody care...it is not noticiable.

EVERY big games usualy used as benchmarks and those using their engines (almost every games) are modified in MANY ways by both ATI and NVIDIA to make you think 1024 8AF 2xAA 80 fps is actually 1280 16AF 4xAA 300 fps, because nobody notice.

ok it is a bit exagerated, but after all the threads i saw....

I don't want any specific optimization from ATI or NVIDIA for ANY games/bench. I wan't my card to run fast on DX9/Opengl CARD INDEPENDANT engines/routines. Specific code for games which is either "bug correction" or "optimization" IS CHEAT by default.

If ATI or NVIDIA think a game can run faster on their hardware path with some a game being coded differently they should release a special EXE on their own or in association with the game company (ex: UT2003_NV35.EXE)

otherwise wouldn't it be a copyright violation?