3D Mark 05 released

[Elie]

Downloading at 285KB from Japan

Regards,
Elie

[Helevitia]

My first run:

4814

Apparently if I upgrade to the beta 8.07 drivers I will gain 1000 points.

[Jon P. Inghram]

I've got this unhealthy temptation to try using 3dAnalyzer to try to trick it into running on my 8500, but I'm resisting.

[Novdid]

Maggi I got it from Sunet, there's nothing wrong with that file. But this one requires ps2.0. So the P is out.

My slightly OC'ed R9700Pro(360/340) scores: 2399

[Maggi]

I figured that much, Novdid, but thanx for clarification ...

[edit]
I found out, the forum software wouldn't let me upload the same pic into two separate threads.
Therefore, I'll post my score over here
[/edit]

[Paddy]

upgrading to pro doesn't add much.

The complex vertex shader test is nice... we as nice as a patch of turf can get.. but you can see all the individual blades of grass. The rest is fairly crappy. There is a 'graph' option that allows you to see how your frame rate varies throughout the tests, but that about it.

[Mehen]

alright... well... check the scores on the FX series... very VERY bad :/
Check out who has the highest 5900ultra score though (atm)

[Maggi]

Anybody noticed their new scoring scheme ?

According to the readme, they now use the following base formula:

(GT1 * GT2 * GT3) ^ (1/3)

My first thought was, "why not simply use the arithmetic mean", ie. why not simply add all scores and divide them by three ?

I started comparing both schemes and pretty quickly, I found out that the new way gives higher scores, the more even all three test ran.

A simple example could be:

GT1 = 30fps
GT2 = 30fps
GT3 = 30fps

the according results would be
new scheme: 30
old scheme: 30

so there's no difference at all between both approaches


The difference pops in, as soon as the single results start varying, eg.

GT1 = 20fps
GT3 = 30fps
GT3 = 40fps

the according results would be
new scheme: ~28,845
old scheme: 30

an even worse scenario could be

GT1 = 10fps
GT3 = 30fps
GT3 = 50fps

the according results would be
new scheme: ~24,662
old scheme: 30

As you can see, the more the single scores differ from each other, the lower the resulting score gets.

Pretty interesting, if you'd ask me.

Further investigations revealed, the new approach, indeed rewards consistent results accross the three game tests and sort of penalizes uneven results.

Drawing a conclusion here, I really appreciate the new way they chose, because the end-user will benefit most from a system that doesn't show any particular weaknesses, but instead performs evenly accross all tests.

In other words, what are 500fps in Q3 worth, if Doom3 crawls at 2fps ...

Maybe this is also the reason why those FX boards that Mehen mentioned are scoring comparably bad ?

[mmp121]

wow, there's already a patch/hotfix for 3DMark05...

http://www.futuremark.com/download/?3dmark05hotfix110.shtml

3DMark05 Hotfix 110

* A possible exploit to inflate results in 3DMark05 version 1.0.0 has been fixed
* Updated Entech DLL
* One incorrect unit in the MS Excel importer has been fixed

Note: Hotfix 110 updates your 3DMark05 (100) to 3DMark05 (110). This hotfix is required to submit your result to the ORB.

[bibble]

god and i thought 3dmark 2003 was big...

[leech]

I ran it on my setup and scored 3789 or something like that (P4 3.06ghz 533FSB, 512mb of ram at 355Mhz, eVGA 6800GT at stock speeds) but then, I had other crap running in the background, and supposedly one of the newly leaked nVidia drivers add quite a lot of performance to it.

I didn't even get to watch it all! I had to leave for work in the morning, and I just let it run. I'll try running it next time I'm in windows, and I don't have anything loaded up, and will try overclocking my system as well (I normally don't overclock anything unless I'm trying to benchmark stuff, or if something runs a bit too slow. Also to see where my system is being held back.... )

Leech

[Mehen]

Well, for the FX5xxx series the new NV drivers do nothing; I'm beating other 5900ultras with my 56.72s, and some of them have more than 1ghz cpu clock speed over me. 3dmark05 seems to be anti Nvidia. (i get beaten by 9500pros wtf?!?!)

[Novdid]

Originally posted by |Mehen|
Well, for the FX5xxx series the new NV drivers do nothing; I'm beating other 5900ultras with my 56.72s, and some of them have more than 1ghz cpu clock speed over me. 3dmark05 seems to be anti Nvidia. (i get beaten by 9500pros wtf?!?!)

There's nothing strange about getting beaten by 9500Pros, as soon as it gets shader heavy that is.

[leech]

I read somewhere else that this 3dmark was supposed to support SM3.0. Well, there are reasons why I didn't get the FX 5 series... But the 6x series is quite nice. I'll give the different driver versions a try when I get home from work, since I have to leave very shortly for that horrible place....

Leech

[TransformX]

429 3DMarks

AthlonXP 1800+
1GB Ram
Geforce 5600 Ultra (128MB)