53.03 official?

I still wonder how Futuremark missed the PS2.0 test optimization. That scares me.

 

52.16 had for the PS2.0 test something else than just shader identification and replacement. We didn't figure out the optimizations for that test in time for the patch, but we needed to launch the patch. By that time there was no nv driver suitable for 3DMark03 use, so we had to get the patch out rather sooner than later.

We have now added the following note to 52.16 on our driver list.
"The 52.16 drivers have 3DMark03 specific optimization for the Pixel Shader 2.0 test and that score is solely comparable between nvidia cards."

 

But your sure you caught every other detection of 3d mark 03 :?

Coincidentally wouldnt it be better to not allow non futuremark approved drivers to be not visible on the orb? I mean be able to publish them but not able to find them by searching on the orb, and maybe put a small message explaining this when you publish them. The reason for still allowing publsihing would be so you could still compare between your incomparable results which are comparable with each other.

Allowing non futuremark certified drivers onto the ORB gives the wrong impression. Few people know what drivers they have installed and even fewer will read your approved driver list.

EDIT: do you know roughly how fast the fx 5900 series are on the ps2.0 test when the detection is disabled?

 

Of course we can understand how difficult it is for you to deal with this kind of things. But unfortunately, even this note is inaccurate. Are you sure that the PS2.0 score is comparable between NVIDIA cards ? IE : NVIDIA can use optimisations to show a virtual improvement between NV31 and NV36. I don't know if it is the case but I've seen this many times.

 

Or the opposite, use the INT 12 units which you cant possibly use with ps2.0. This would benefit the normally very slow nv30 derived cards. It would effectively put them only a bit behind the newer cards as the nv30 cards INT 12 is about as fast as the nv35 cards FP16. Which also shouldnt be being used....

 

Nah, I figure that NVIDIA realise that since they are going to be doing this procedure very often, IE everytime a new game / patch comes out. Then it makes sense to only re-optimise the stuff that counts in the final score.

 

Excepting that since the PS2.0 test itself wasn't changed between version 330 and 340, there must have been a change in the 53.03 drivers themselves. (And early-on changes, since the loss was seen on the earliest beta release.)

 

Preliminary results of UT2003 testing: No change. Forgot to made a 52.16 control panel AF image to compare to, but the 53.03 difference is exactly the same as between the other 52.16 application AF/control panel AF shots that I've seen.

 

the following bit is taken from nvidia's release notes fro driver 53.03

whatever they were doing with 3dmark2k3 b4 the 340 patch
they r doing it again...

[complaint on]
all of the "optimisations" seem 2 address exclusively the gffx
lineup.... wot is happening with users of earlier gf products?
[complaint off]

 

Users of earlier products are generally being ignored. The GfFX cards were the first ones with performance issues such that really blatant cheats had to be employed to bring the numbers up.

 

And....why Futuremark will always be considered Nvidias playground until they change their practices. Which include no uploading to the ORB with drivers that contain application detection, shader replacement.

 

More BS from Nvidia. By insisting that their hand tuned shader replacements are part of their "compiler", they can keep saying that all they are doing is "re-enabling the compiler".

FM needs to act now to regain their credibility and ban all results from cheating drivers from the ORB. And that includes the previously approved version that they have now admitted has shader cheats they were not able to disable. Despite their big words from a month ago, FM is not being seen to police their benchmark and ORB from cheating drivers.

 

At least they aren't too shy too admit they're doing it.

 

Now wait a tick, didn't NV state *officially* before that the 340 patch did not disable the compiler?

 

I can assure you that older nVidia cards are NOT being ignored, and benefit greatly from the new drivers (I should know, I have a GF4 Ti4600 and upgraded from a GF3 Ti200).

If you read the pdf release notes on the drivers page you will find the TNT-1 is still being supported and coded for with the latest 53.03 drivers... When was the last time ATi updated their Rage/Mach series drivers???

 

Oh yes. UT2003 results.

No change. Waiting to hear what the driver fixes.

 

Luciano Alibrandi retracted the statement about the "Unified Compiler" (AKA "All our cheats are in the same place") being disabled, IIRC...

Date was 11/14

http://www.xbitlabs.com/news/video/display/20031114041519.html

 

Aw, you guys beat me to it.

Anyway, I don't want me having strained my eyes the last hour in vain, so here it is, with quotes & stuff:

The accusation:
http://www.xbitlabs.com/news/video/display/20031112031947.html

Credit for the original link goes to madshi from this thread here on B3D: http://www.beyond3d.com/forum/viewtopic.php?t=8952&start=180

NVidia's initial response:
http://www.xbitlabs.com/news/video/display/20031112181114.html

Credit for the original link again goes to madshi, same thread, page 14.

NVidia's revised answer:
http://www.xbitlabs.com/news/video/display/20031114041519.html

Credit for the original link goes to AndY1, same thread, page 20.

Cheers,
Mac