ALL NVidia GTX 670 drivers compared (2012 --> 2015)

[Keystone Nyan Cat]

[sCROLL DOWN TO SEE DRIVER COMPARISON GRAPHS]

 

I have been an owner of the GTX 670 since 2012. I’ve read numerous claims from people stating that new drivers have led to reduced performance, but all of them were one off cases. Nonetheless, I remained curious, and over the past few days, have put these claims to the test by comparing five AAA games, each using a different game engine.

 

(Please note the data values on the Y-axis)

The data collected is the average frames per second for a given play through. I urge you to draw your own conclusions with the given data; below is my interpretation.

 

All the tested games have had a net improvement in performance over time, even after their end of life and introduction of a new series of cards.
 

1. In Dirt 3, I was concerned with 353.06’s performance, which is almost 10 FPS lower than the preceding driver, 350.12; this is corrected with 353.06 though.
2. Just Cause 2 had a consistent improvement in performance after 344.89, which may have to do with the release of the multiplayer mod during that time (see table), increasing the player base. Just Cause 2 has had an almost 25 FPS increase from its lowest performance in 301.42 and its best performance in 347.88.
3. Battlefield 3 had a ~5 FPS boost from its first driver to the time leading to the 700 series launch
4. Sleeping Dogs, and Crysis 2 have had no major performance improvement or degradation

The data does not account for added features such as ShadowPlay, or for support for newer games and performance enhancing improvements such as the capability for running SLI when using older hardware which may improve the driver in people's view

.

Below are the settings used for each of the tests, along with the in game mission/ benchmark which I used for gathering the average frames per second.

1. Battlefield 3
              Preset: Ultra
              Mission: Going Hunting
              FOV: 70
              Motion Blur: 0
              Stop benchmark 5-6 seconds after first enemy fires at you
2. Crysis 2
              Preset: Ultra     
              Mission: A Walk in the Park        
              Motion Blur: Low            
              When spawned; walk on the pathway, looking only straight on. Use armor mode with   the rifle equipped to take down the first   enemy; and invisibility with a shotgun to take down the other two. Walk up to the garden with the cloak, and pan from left to right    and then end the benchmark
3. Dirt 3
         Preset: Ultra (no AA)     
         Mission: Benchmark (8 people rally in Aspen, 3^rd person view)
4. Just Cause 2
              Preset: Ultra (no AA; Bohek Filter Off)
              Mission: Benchmark (Concrete Jungle)
              Motion Blur: 0
5. Sleeping Dogs
              Preset: Extreme
              Mission: Benchmark
              Motion Blur: 0

Here’s the procedure I followed for removing and installing a driver

1. Uninstall the driver in “Programs and Features”
2. Restart the system as prompted
3. Delete the NVidia folder in the c:// folder
4. Install the new driver (deselect everything else such as PhysX, HD Audio, GeForce Experience)
5. Restart the system as prompted
6. In NVidia Control Panel,
   change only the following V-Sync option à “Use the 3D application setting” à “Off”
   If the time taken for the drop down menu under V-Sync takes over 40 seconds to open during which time the NVidia control panel freezes, the menu is denoted “laggy”, if it responds instantly, it’s denoted “responsive”

Miscellaneous

1. During testing, the GTX 670 stayed under 70 C in a temperature controlled environment of 21 C
2. All tests were conducted offline
3. 3DMark could not be tested because it cannot start offline
4. Crysis 3 was omitted since during cut scenes, the frame rate exceeded 1000 frames per second
5. Dying Light and Battlefield 4 are fairly new, the range of drivers to test them from would be small, hence these games were omitted
6. I was initially testing two 670s in SLI under maximize 3D performance on a resolution of 5760x1080. However, I soon realized that not many people would have a setup consisting of three screens, hence opted for a single 1080p panel. However, SLI on a single screen caused purple flickering in all tested games; something which I could not find a solution to. Unfortunately, I ruled out several games such as BioShock Infinite and Arkham City to this; a problem solved by using only a single GPU. I did not revisit these games for testing when using a single card. Should I do this test in the future again with another card, I will bear this in mind.
7. Certain games such as Dying Light crashed when FRAPS was being used to record the frames per second.
8. At the time of running the tests, the only applications running in the background were Microsoft Security Essentials (for all), MSI Afterburner (for all), and, FRAPS (when testing Crysis 2 and Battlefield 3)
    

All tests were run on the following setup-

• Intel i7 3770K
• ASUS P8Z77-V-Pro
• Zotac GTX 670 (Reference)
• Corsair HX850
• Corsair 8GB (2X4GB) CMZ8GX3M2A1600C9
• Corsair Force 3 SSD for the OS
• Windows 7 Home Premium
• 2x WD1003ABYX RE4 HDDs in RAID 0 for the games
• Acer S221HQL (1920X1080)         

 [i appear to have forgotten to make the table; I'll add that tomorrow morning (it's 3 in the morning right now!) ]

Edited August 7, 2017 by Keystone Nyan Cat
Hmm... For some reason, the same graphs had been added twice. Removed duplicate graphs.

[Swndlr]

Good post

[Guest Philosophy]

Wow! Awesome! 

[FatPenguin]

wow! nice job on the post! very clean  

[CydiaDash]

Wow, a lot of effort went into this

[ForsakenLive]

Really nice post!

 

It looks very similar to the GTX 480 results linus showed on the video, where in some games there is little to no difference, but on many titles there is an actual improvement over time with driver maturity!

[Senzelian]

Well done.

How long did it take to do all those graphs and testing?

Btw.: NVidia Fanboy. You didn't include an AMD card...   

[TomvanWijnen]

Great post!

 

Yay, so this means I can safely update the drivers of my GTX 680 (something I haven't done since I built this computer, 1 and a half years ago...   ).

[MahtXL]

Well done.

How long did it take to do all those graphs and testing?

Btw.: NVidia Fanboy. You didn't include an AMD card...   

Neither did linus to be fair.

[Senzelian]

Neither did linus to be fair.

Oh god damn... Someone had to ruin the joke...   

Some people were crying about Linus' review, because they sad it was biased because he didn't include an AMD card.

That's why I was referring to it as a joke.

Obviously non of them are fanboys or anything like that. 

[MahtXL]

Oh god damn... Someone had to ruin the joke...   

Some people were crying about Linus' review, because they sad it was biased because he didn't include an AMD card.

That's why I was referring to it as a joke.

Obviously non of them are fanboys or anything like that. 

oh im an amd fanboy alright just not their gpus as ive only ever had one and it was a cheapo low end card in a stock hp pc.

[Senzelian]

oh im an amd fanboy alright just not their gpus as ive only ever had one and it was a cheapo low end card in a stock hp pc.

Do you even read? Do you even have eyes?  :blink: 

How the hell can you even think that I called you an AMD fanboy? WTF QQ 

[MahtXL]

Do you even read? Do you even have eyes?  :blink: 

How the hell can you even think that I called you an AMD fanboy? WTF QQ 

never claimed you did . just pointed out that its not entirely fanboys crying over the amd card not being shown as well, because im one and couldnt care less .

[Keystone Nyan Cat]

@Swndlr @ @FatPenguin @CydiaDash @ForsakenLive @Senzelian Thanks for all the kind words!

 

@Senzelian Around 4 hours before testing trying to get things working; then 25 minutes per test (around 14 hours testing); and then 4-5 hours making the graphs (this does not include breaks) Haha, yup, sorry about that; I'll be sure to include one if I do something like this again.

 

@TomvanWijnen Thanks! I'm glad I could help 

 

Edit- I've added the table as well

[Senzelian]

never claimed you did . just pointed out that its not entirely fanboys crying over the amd card not being shown as well, because im one and couldnt care less .

Alright...  :wacko:

[Hulk]

This is my first time tagging someone to look at something, but I feel this definitely deserves the attention @LinusTech and @Slick 

 

To the OP- I've been speculating that Nvidias been sabotaging newer drivers for older cards. This puts these thoughts to a rest, and is really detailed. Thank you for sharing it!   

[alpenwasser]

I don't really game anymore, so this isn't very relevant for me. But I really like how

you did the plots (full version to see the changes in relation to the absolute values,

then cropped with emphasis on the differences between the chances), and the documentation

of methodology. Nice work IMHO!

[Keystone Nyan Cat]

@alpenwasser and @Hulk Thank you for the kind words!

[Curufinwe_wins]

Gratz on the sticky!

[Picky]

Still rocking a 670 FTW. Loud whining from the fan, very high pitched electrical whine. Goes away when I put my finger on the fan to stop it. I think it's time to upgrade to a 980 TI but waiting for a nice deal.

[Andrew Sh]

Misunderstood the graphs a little at first. Lots of work done here. Well done.

Though funny enough, there are quite a few performance dips throughout the testing.

Wonder if its just how things turned out

[Mahigan]

Next up, GameWorks titles please:)

There's a similar post to yours on Overclock.net with the same conclusion. There is also an article over at hardware canucks with a similar conclusion.

If you test an AMD card, you will see a continual boost in performance over time. That's why an R9 290x, which lost handily to a GTX 780 Ti now, is within 6% of it on avg (winning and losing): http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/70125-gtx-780-ti-vs-r9-290x-rematch.html

It's not so much a driver thing as it is an architecture thing. As new games are released, AMD cards mature better. This has to do with the gradual shift from Graphics to Compute, over time, in titles. We'll see this even more clearly with DX12 titles.

I think most Kepler users are comparing themselves to AMD GCN users and that this is where these claims stem from.

[abdrad]

I can see you spent a lot of time on this, nice & useful!

[Thomas CG]

Well done, nice to include a close up graph of each test.