Discussion Nvidia once again cheating benchmarks

Prefix_NA

Seasoned Member
Messages
3,817
Reaction score
962
There seems to be a major problem with NV's default settings in the control panel. Under default, it's supposed to "Let Application Decide" on AF. Yet it overrides in-game settings to **** quality for texture filtering.

It started with a guy on OCUK who benched Titan X & Fury X, people noticed worse IQ on Titan X:



http://forums.overclockers.co.uk/showthread.php?t=18679713

A full PNG comparison screenshot with NV on default driver setting vs forced on Quality: http://i.imgur.com/0UzUR3A.jpg

There is a significant performance (10% perf hit) difference to leaving it on default and forcing it to Quality (to match AMD's default image quality).

A lot of review sites don't touch it, they leave it on default, as per [H]: http://hardforum.com/showpost.php?p=1041709145&postcount=81

One reason I don't think it's a recent bug, is because its been happening for a long time, yet its undisclosed by review sites, they don't even do IQ testings these days.

From a few years ago, PCPer's video to compare performance. You can see AF is rubbish on NV.

http://imagizer.imageshack.us/a/img834/8836/example4i.jpg


Source (PCPER video review):
Basically its either a bug where the default settings are supposed to let the app decide on quality, so when you set in-game to Ultra you expect it to be Ultra, but its actually running Low. If review sites test with settings on default, it leads to an inflated benchmark numbers for nvidia GPUs.

This kind of cheating is so early 2000s (http://www.geek.com/games/futuremark-confirms-nvidia-is-cheating-in-benchmark-553361/ ), as a gamer, I would have thought it no longer happens!

Edit 1: Project Cars IQ comparison, same settings in-game, default drivers: http://i.imgur.com/7lJhNsU.jpg

(no AF on Titan X, road textures blurred) source:

Changing in Nvidia control panel to prefer max quality actually fixes this "bug" and stops Nvidia quality degradation
http://forums.overclockers.co.uk/showpost.php?p=28267475&postcount=197

Source: https://www.reddit.com/r/pcmasterrace/comments/3c2kiw/bug_or_nv_cheating_with_optimized_image_quality/

Wonder if Nvidia will admit this in a similar way to the Kepler gimp where they also claim its another "Driver bug"


Video comparisons here.
https://www.youtube.com/user/gregsvidchan/videos
 
Last edited:

Prefix_NA

Seasoned Member
Messages
3,817
Reaction score
962
One of the videos is dead.
Fixed with a link to see other video's

Also look how much of an impact changing nvidia cp to make it the same as AMD's makes an impact.

Taken from about every 30 seconds (I couldn't do 0 seconds because menu so I took when menu ended)
start = 109 to 95
:30 = 110 to 95
1:00 = 90 to 84
1:30 = 95 to 88
2:00 = 84 to 76
2:30 = 68 to 66 (smoke blocks most objects)
3:00 = 114 to 108

58 FPS difference from the 7 interval frames or a 8.286 frame rate loss overall with some area's being more impacted while others were less impacting. The issue is not Nvidia settings to change the game the issue is they OVERRIDE application settings by default and it says use application defaults yet it does not.
 

Selena Gomez

Twitch.tv/kingsintv
Retired
Messages
16,886
Reaction score
9,558
I haven't touched any of the nvidia control panel settings and yet my games look amazing so... not sure what this is going on about.
 

Z61

Some times our saints are sinners
Retired
Messages
5,532
Reaction score
3,423
I haven't touched any of the nvidia control panel settings and yet my games look amazing so... not sure what this is going on about.
If you looked at any of the screenshots you would notice that there's a difference in quality with the NCP settings and game settings.
 

Prefix_NA

Seasoned Member
Messages
3,817
Reaction score
962
I haven't touched any of the nvidia control panel settings and yet my games look amazing so... not sure what this is going on about.
Anisotropic Filtering and a few other stuff are incredibly low capped for Maxwell cards (Doesn't seem to be happening on Kepler). Even when turning them up in game it gives no change as they are capped in driver despite the nvidia cp saying let applications decide.

Its about 8% drop in performance (in BF4) to fix the Nvidia CP to give same picture as the Fury. This does also explain why the 960 does so much above the 750ti in recent games despite both being Maxwell v1 architecture because the 700 series did not seek this "driver bug" as nvidia will probably call it.


Draw distance even seems lower. (Video at 1080p I believe.)


Ignore the smoke in this notice how less objects on the ground far away appear and how blurry background is?
 

Selena Gomez

Twitch.tv/kingsintv
Retired
Messages
16,886
Reaction score
9,558
Anisotropic Filtering and a few other stuff are incredibly low capped for Maxwell cards (Doesn't seem to be happening on Kepler). Even when turning them up in game it gives no change as they are capped in driver despite the nvidia cp saying let applications decide.

Its about 8% drop in performance (in BF4) to fix the Nvidia CP to give same picture as the Fury. This does also explain why the 960 does so much above the 750ti in recent games despite both being Maxwell v1 architecture because the 700 series did not seek this "driver bug" as nvidia will probably call it.


Draw distance even seems lower. (Video at 1080p I believe.)


Ignore the smoke in this notice how less objects on the ground far away appear and how blurry background is?

So what should I be changing?


 

Prefix_NA

Seasoned Member
Messages
3,817
Reaction score
962
Adjust image settings with preview tab in Nvidia CP I believe its called.

Select best Quality wth the slider. The "Let application decide" is actually capping many things at driver level despite what game says.
So what should I be changing?
 

Selena Gomez

Twitch.tv/kingsintv
Retired
Messages
16,886
Reaction score
9,558
Adjust image settings with preview tab in Nvidia CP I believe its called.

Select best Quality wth the slider. The "Let application decide" is actually capping many things at driver level despite what game says.


That was already set like that by default. Should I set "Use my preference emphasizing"?
 

Prefix_NA

Seasoned Member
Messages
3,817
Reaction score
962


That was already set like that by default. Should I set "Use my preference emphasizing"?
Change it to preferred and put slider on quality.

The top 2 options have some limitations.

You could also change anisotropic manually in the 3d settings to specific levels and you can see how its still not using 16x even on quality in games even if it says 16x in game. But I digress. I personally have tessellation capped at 16x on my AMD CCC but no other global changes.
 

Prefix_NA

Seasoned Member
Messages
3,817
Reaction score
962
Done.

Will this screw any current in game settings? Like it won't automatically change them.
No this doesn't affect your game settings. It does however make your game settings actually take effect.

People on reddit looking into the Kepler performance issue actually found that only Maxwell cards seemed to be noticeably lower draw distance and anisotropic filtering than AMD's.


Here is some BF4 on Nvidia to show there are difference and its intentionally misleading to benchmarks when nvidia is rendering lower stuff then saying FPS similar to AMD's.


Look at the road and see less things render & quality is lower by default despite having game on fully max settings.



By default AMD's looks like the quality nvidia setting. So seems awfully fishy that Nvidia always has these "Driver bugs" they call them that make Nvidia appear better in benchmarks.
 

Jet1337

Enthusiast
Messages
845
Reaction score
271
This doesn't affect me atm because the only game I'm playing right now is Skyrim and the mods I use require me to disable AA and all the other graphic settings.
 

Prefix_NA

Seasoned Member
Messages
3,817
Reaction score
962
This doesn't affect me atm because the only game I'm playing right now is Skyrim and the mods I use require me to disable AA and all the other graphic settings.
Well draw distance would be affecting you but most likely not anisotropic filtering.
 
Top Bottom