View Full Version : 3dmark03 On Nvidia Cards Redux
StealthHawk
04-14-03, 06:02 AM
first a little background. as is well known, after 3dmark03 came out nvidia bashed it, citing that it was a bad representation of reality, and that it was easy to optimize for.
towards this purpose they released the 42.72 infamous cheat drivers which dramatically raised 3dmark03 scores for gfFX users, but also reduced the precision of the shaders and lowered IQ in GT4.
this was fixed in later drivers, but 3dmark03 scores decreased.
nvidia recently released the 43.45 drivers, which again raised scores, but apparently had the same lower precision of 42.72. you can see the differences between the "cheat" and what the scene should normally look like in a screenshot by Dark Crow here (http://www.darkcrow.co.kr/image/tip&tech/4345tech/iq/game4.jpg)
with the newly leaked 43.51 drivers, people in our own forums (http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=10056) have reported this when using a gfFX5800:Originally posted by skypx
2k3 (http://service.futuremark.com/compare?2k3=591794)
Score is on par with the 43.45 dets.
and Dark Crow has posted a screenshot of GT4, Mother Nature, which shows that image quality is perfectly fine now: http://www.darkcrow.co.kr/image/news/0304/4351.jpg
so there you have it. it appears that nvidia was able to retain the same high 3dmark03 score, without any noticeable reduced precision :)
Hanners
04-14-03, 09:07 AM
Nothing is free in 3D.... ;)
Originally posted by Hanners
Nothing is free in 3D.... ;)
Unless you enjoy watching a blank screen being rendered. :lol:
Typedef Enum
04-14-03, 10:09 AM
I knew this was going to happen...
nVidia comes out of the gate with very subsandard drivers, across the board. In order to satisfy the reviewers/readers, they release a drivers that clearly cheated, if you will, to raise performance...UNTIL they had enough time to get their drivers up to scratch.
So, a bunch of time passes, and they eventually have a drivers that's capable of delivering higher performance without having to resort to cheating. The expectation is that (from their POV), people will consider this whole thing acceptable because nVidia eventually fixed/addressed the problem. I'm not one of them. Under no circumstances should ANY outfit be allowed to force special drivers to "cheat" until their driver team can get the software up to scratch. I mean, what kind of integrity is that?
I hope this is the end of the "special" driver...The next time I see either ATI/nVidia providing such a drivers to websites, I'm going to make sure the issue gets raised.
I knew this was going to happen...
Who didn't John?
Originally posted by StealthHawk
first a little background. as is well known, after 3dmark03 came out nvidia bashed it, citing that it was a bad representation of reality, and that it was easy to optimize for.
towards this purpose they released the 42.72 infamous cheat drivers which dramatically raised 3dmark03 scores for gfFX users, but also reduced the precision of the shaders and lowered IQ in GT4.
this was fixed in later drivers, but 3dmark03 scores decreased.
nvidia recently released the 43.45 drivers, which again raised scores, but apparently had the same lower precision of 42.72. you can see the differences between the "cheat" and what the scene should normally look like in a screenshot by Dark Crow here (http://www.darkcrow.co.kr/image/tip&tech/4345tech/iq/game4.jpg)
with the newly leaked 43.51 drivers, people in our own forums (http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=10056) have reported this when using a gfFX5800:
and Dark Crow has posted a screenshot of GT4, Mother Nature, which shows that image quality is perfectly fine now: http://www.darkcrow.co.kr/image/news/0304/4351.jpg
so there you have it. it appears that nvidia was able to retain the same high 3dmark03 score, without any noticeable reduced precision :)
that link provided continues to show 'discoloration' in the sky compared to other driver sets..
though it is less so than before...
nVidia is STILL cheating. But they're cheating in a way every end user should accept, IMO: they're cheating and doing their best not to reduce IQ too much.
This was pretty much to be expected: before, nVidia didn't have the time to cheat well. They were simply using INT12 pretty much everywhere.
Now, they're using FP16 where it increases IQ, too. But only where it increases IQ. And yes, it does look a little worse than older drivers - I'd guess that's because those used FP32. But still, it's a lot better, and I think reviewers should begin to think using those drivers is becoming quite fair indeed :)
Remember that in most cases where you need dynamic range, you couldn't see the difference between FP16, FP24 and FP32. 3DMark 2003 is an exagerated case - and that's a good thing, IMO, because that truly enables you to see the difference, which is a good thing for reviewers and people examining what's really going on at forums like Beyond3D.
Uttar
Hanners
04-14-03, 01:08 PM
Regardless of the quality of the output, 'optimising' for a benchmark in that way still gives me a bad feeling inside. :(
I know, I know, all IHVs do it to some extent, but this is still much more blatant and 'major' than most.
Still, it's a step in the right direction I suppose.
John Reynolds
04-14-03, 01:11 PM
Originally posted by K.I.L.E.R
Who didn't John?
Scott, not John.
nVidia is STILL cheating. But they're cheating in a way every end user should accept, IMO: they're cheating and doing their best not to reduce IQ too much.
:lol: :rolleyes:
You sound like you bought all the PR that's been thrown around lately.
Some of us don't even want "a little" bit of IQ downgraded even if it is for a massive performance increase.
I can't even begin to critique your statement.
Every end user WILL NOT accept this, every end user SHOULD NOT accept this. You do not talk for everyone, remember that.
You make it sound like cheating SHOULD be acceptable.
Scott, not John.
My apologies Scott.
Fair? No. Maybe tolerable.
And we can't be sure if this improvement(well exactly speaking...less retarded version) would work for all the other DX9 gamings and benchmarks or just for 3DMark03 alone.
Let's wait for Aquamark3? ;)
Originally posted by Uttar
and I think reviewers should begin to think using those drivers is becoming quite fair indeed :)
I just asked Gainward to send back their 5800 Ultra sample to me for some further testing of these drivers specifically. :)
Perhaps I dare to post some 3DMark03 scores this time around. ;)
Originally posted by K.I.L.E.R
:lol: :rolleyes:
You sound like you bought all the PR that's been thrown around lately.
Some of us don't even want "a little" bit of IQ downgraded even if it is for a massive performance increase.
I can't even begin to critique your statement.
Every end user WILL NOT accept this, every end user SHOULD NOT accept this. You do not talk for everyone, remember that.
You make it sound like cheating SHOULD be acceptable.
Actually, who cares if you cheat without getting caught and that you do it nearly everywhere? Heck, I'm pretty sure nVidia will optimize a lot of other games similarly - the GFFX is way too complex to allow games to do things properly:rolleyes:
If nVidia didn't cheat everywhere, then it'd be another matter. But they do. So damn, the problem isn't that they're cheating.
The problem is *only* IQ. What we care about, as gamers, is IQ. Of course, as techies, other things matter, but that's really not the primary concern IMO ( feel free to have a different opinion - but then accept mine too. )
Sorry for saying that it isn't a problem to have slightly lower IQ. Probably was lame there... I think, however, that the users should be given the option of a little bit less IQ to get more performance - I did say *option* here. Reviewers should use their best judgement to use the best setting.
And on the topic of IQ - after rechecking the screenshots, I'm not even too sure it's worse than 43.03 anymore...
http://www.darkcrow.co.kr/image/tip&tech/4345tech/iq/game4.jpg
http://www.darkcrow.co.kr/image/news/0304/4351.jpg
The 43.51 looks just as good as the 43.03, it seems :) Didn't check compared to a R300 yet, though.
Uttar
gokickrocks
04-14-03, 04:23 PM
im going to assume that the pics are taken at different frames because some things are missing and the colors also look saturated (looks like someone turned up the brightness/gamma)
marcocom
04-14-03, 04:41 PM
Originally posted by K.I.L.E.R
:lol: :rolleyes:
You sound like you bought all the PR that's been thrown around lately.
Some of us don't even want "a little" bit of IQ downgraded even if it is for a massive performance increase.
I can't even begin to critique your statement.
Every end user WILL NOT accept this, every end user SHOULD NOT accept this. You do not talk for everyone, remember that.
You make it sound like cheating SHOULD be acceptable.
My apologies Scott.
agreed.
nvidia has been so afraid to be confident in the need to engineer an entirely new product and keep just trying to play the race game with ATi and its making them look bad over and over again.
nvidia, take your time. your not being faster than ATi would be fine with most of us if you showcased feature-set and scaleability and the other things that make the FX special. trying to beat ATi while they use a very mature chip like R3xx is devestating and probably impossible for awhile.
nvidia, stop the insanity and get a grip on your market relations. your cooler than this.
GlowStick
04-14-03, 04:53 PM
Though i dont have the best eyes,
http://www.darkcrow.co.kr/image/news/0304/4351.jpg
looks the same as when i run 3dmark03 on my 9800Pro
thats my 2cents =0
StealthHawk
04-14-03, 06:30 PM
Originally posted by Sazar
that link provided continues to show 'discoloration' in the sky compared to other driver sets..
though it is less so than before...
the sky looks the same to me
old driver with no cheating (http://www.darkcrow.co.kr/image/tip&tech/4345tech/iq/game42.jpg)
new performance optimized driver (http://www.darkcrow.co.kr/image/news/0304/4351.jpg)
on the subject of "cheating." unless image quality is being taken away, I don't consider it cheating.
RobHague
04-14-03, 06:45 PM
Exactley, first they were cheating because the nature demo IQ was a bit dodgy looking. Now the image quality looks fine people are still saying they are cheating because the 3dmark is just as good and thats what nvidia are like :rolleyes:. Im just wondering after the WHQL driver is released what more people will want to convince them nvidia are back on track??
Incidentiall the newer drivers give me a higher score with the Quality setting than they did before. So the IQ is back to normal and i get a few more marks to boot. :D
StealthHawk
04-14-03, 07:01 PM
Originally posted by RobHague
Exactley, first they were cheating because the nature demo IQ was a bit dodgy looking. Now the image quality looks fine people are still saying they are cheating because the 3dmark is just as good and thats what nvidia are like :rolleyes:. Im just wondering after the WHQL driver is released what more people will want to convince them nvidia are back on track??
Incidentiall the newer drivers give me a higher score with the Quality setting than they did before. So the IQ is back to normal and i get a few more marks to boot. :D
I think people are confusing cheating with optimizing. or they just like to nit pick.
as an aside, I would like to know if nvidia fixed the problems with PS2.0 with these drivers.
StealthHawk
04-14-03, 10:56 PM
agreed, if image quality is lowered then it is most definitely cheating.
take the driver challenge. the sky looks the same. see post with the two links labelled "old driver with no cheating
new performance optimized drive"
the trees look different, but I'm not sure which shot is supposed to be "correct." it looks like the leaves are just in different places.
gokickrocks
04-14-03, 11:19 PM
i will post a dx9 sdk reference rasterizer image to compare it to in a few hours (yea it takes that long)...
gokickrocks
04-14-03, 11:19 PM
also, could someone tell me how to make the animated images?
Using GIF is a BAD idea. The GIF format is limited to 256 colours, no way you can properly judge the quality of the screenshots.
A better format would be MNG, the sister format of PNG that has animation support just like GIF and supports 24bit colour like PNG or JPEG. Unfortunately, because of MS lazyness or whatever reason only known to the IE for Windows team, there's no MNG support in IE and poor PNG support, while Mozilla based browsers FULLY support both MNG and PNG features.
gokickrocks
04-15-03, 01:04 AM
i was planning on using MNG...
but i just realized the reference 5800 pic is a GIF, so could someone with a 5800 take a PNG screen cap of frame 800 in 3dmark03's nature...
the rasterizer is still going at it, just a few more HOURS to go
OICAspork
04-15-03, 01:15 AM
The reason I have for continuing to be dissapointed with it, even if the image quality is fine... is that all these cheats and work arounds for a benchmark, it is that much time taken away from optimising for games... and damn, if they have to take every freaking game and 'optimise' for each... those drivers are going to get really bloated... and they'll always be playing catch up. They really need to make the drivers work better over the broad range of all applications.
Dear lord, think about it... if they go through game after game and say "this only needs int12... we'll use fp16 on this... oh that needs fp32" for every effect, it is just rediculous.
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.