NVIDIA GeForce 522.25 Driver Analysis

[Flydommo]

The additional FPS are very welcome on my trusty old GTX 1080Ti.

 

[anachron]

I checked a few games myself before and after updating to these drivers and saw no improvement on my 3070@1440.
Division 2 DX12 showed <1% difference gain is FPS, though the CPU graph was slightly smoother.
Other DX11(Hunt, Steep) and Vulkan(Rage 2, WWZ) games had no gains.
I didn't change any settings or test other resolutions so YMMV.
I've not had any stability issues with these 522.25 drivers.

Other sources are claiming the performance gains are from "shader compilation optimizations, improved Resizable BAR, and less CPU overhead".

Supposedly, these gains were seen on a 3090:
Assassin’s Creed Valhalla: up to 24%
Battlefield 2042: up to 7%
Borderlands 3: Up to 8%
Call of Duty: Vanguard: up to 12%
Control: up to 6%
Cyberpunk 2077: up to 20%
F1Ⓡ 22: up to 17%
Far Cry 6: up to 5%
Forza Horizon 5: up to 8%
Horizon Zero Dawn: Complete Edition: up to 8%
Red Dead Redemption 2: up to 7%
Shadow of the Tomb Raider: up to 5%
Tom Clancy’s The Division 2: up to 5%
Watch Dogs: Legion: up to 9%
They claim higher end cards gain more, with the 3060 only gaining up to 7% on AC: Valhalla

Seems like more gains on games that use resizable BAR, but IDK...
I was hoping to see something, but on my rig, on my settings, on my games, I got nothing...

I got a 7% gain on Cyberpunk 2077 in 1440p with my 2070 Super and i'm still on an old bios with no Resizable BAR support on my motherboard. That's not too far from this review results, so it doesn't seems that Resizable BAR support got a big impact on the performance gain.

 

[solarmystic]

A game that's not in TPU's test suite, Horizon Zero Dawn scored noticeable gains in all metrics on my 12400F/2080 TI system moving from 516.94 to 522.25. Absolutely worth upgrading to, especially if you're still playing this game. I daresay this driver is as good as the previous performance champ, 511.79 for older cards.

For example, at 1080p Ultimate, Average FPS went up from 134 on 516.94 to 141 on 522.25, but more crucially, 1% CPU and GPU low FPS went up by 7 and 15 respectively. Definitely a more smoother experience on this very CPU limited game.

516.94

522.25

 

[Vayra86]

it should be possible to benchmark battlefield 2042 there is a conquest-solo mode
Or a custom portal mode

But why though, franchise is a total crapfest, arguably worse than CoD

I say good riddance for being an always online product.

 

[W1zzard]

Horizon Zero Dawn

Why are people so obsessed with this game? Is terrible programming is now a reference benchmark to represent the typical PC gaming experience?

 

[nguyen]

Nvidia Fine Leather

 

[Red_Machine]

"These are amazing performance gains"
The amazing performance gains: 3-5fps

Am I just jaded, or is this a load of hot air? Call me back when there's a 20fps bump, then I might be impressed. Maybe.

 

[spajdrEX]

Days Gone (UE4) is using DX11 only, not DX12.

 

[qubit]

"These are amazing performance gains"
The amazing performance gains: 3-5fps

Am I just jaded, or is this a load of hot air? Call me back when there's a 20fps bump, then I might be impressed. Maybe.

I'll second that. I remember the sometimes huge performance gains available yesteryear that left these in the dust.

 

[tps3443]

Crazy to see how fast a bone stock 3090 actually is during gameplay. But then again, my overclocked 3090 is putting down 25% higher 3DMark graphics scores than a standard and stock 3090. These cards will hold up for the long haul for sure. Especially when you water cool one, and use the right bios. These GPU‘s are plenty fast! My 3090 Kingpin Is actually 5% faster right out of the box than a 3090Ti FE. I highly considered upgrading to the 4090. But I do play at 2560x1440P. The boost looks to be a solid 50%. That’s not really all that much during real game play when you are already averaging 80-200fps in anything with just a basic stock 3090. Just saying. The 4090 is awesome. But, no rush on upgrading that’s for sure.

 

[RainingTacco]

So it improves the performance in newest DX12 titles, but reduces it in DX11? Why?

 

[Ruru]

So a little performance boost for 1080 Ti, need to check these out. I hope there's no audio issues with these like I had with 517.48 drivers.

 

[wolf]

We keep hearing about "AMD Fine Wine", but it looks like NVIDIA optimizes performance for older drivers, too.

I wonder if NVidia did it, just because we keep hearing about AMD fine wine. Save a bunch of optimizations in one release to maximize media exposure.

And it works.

 

[Space Lynx]

any gains are good gains in my book. well done Nvidia.

@W1zzard

In the future you may want to color code the bars based on the gpu. the dark greens and light greens make it just a smidge hard to read. like take example God of War below, I almost thought for a second that there was a 30 fps gain in the driver, then I took a closer look, but if you were to say use Blue for both 1080 ti graphics, one blue bold one light blue, then light green and dark green for 2080 ti, and then say red bold/light for the 3090, i think it would be easier for the end user to read it. just my two cents, I don't actually care, just trying to help

 

[Ware]

That's the thing, that's why we have reviews on the TPU, and they have found out there were increases from 1 - 2.5%, and that's just using some games. If you included more games, you'd probably see it is a wash, meaning no meaningful gain across the board, but rather some give and take like it usually is with new driver releases.

I'm sorry, but I didn't mean to imply I tested all those games myself - I personally tested only 5.
I wasn't talking about gains from Resizable BAR in general - I was talking about these drivers and those games that were tested(not by me) and showed greater gains, being games that do benefit more than 1-2% from Resizable BAR, and the possibility that it may have been a contributing factor to the huge claimed increases such as 24% in AC.

I got a 7% gain on Cyberpunk 2077 in 1440p with my 2070 Super and i'm still on an old bios with no Resizable BAR support on my motherboard. That's not too far from this review results, so it doesn't seems that Resizable BAR support got a big impact on the performance gain.

Thanks. This is what I was speculating about. Your right, that does seem roughly in line with the results here...

I didn't see any info on the test system used to get a 24% gain in AC, and I trust the 17% figure from TPU more - still a huge gain.
The consistent bit I'm seeing is that the biggest gains are from higher end cards at lower resolutions(3090@1080).

 

[OneMoar]

Why are people so obsessed with this game? Is terrible programming is now a reference benchmark to represent the typical PC gaming experience?

Have you Not played a video game recently?
Terrible Programing is the Default

 

[LifeOnMars]

I'll second that. I remember the sometimes huge performance gains available yesteryear that left these in the dust.

I suppose we could argue maybe that the drivers are more optimised from the get-go nowadays leaving less wiggle room. Also, most modern cards are pushed to the max at stock (undervolting helps efficiency) whereas overclocking in the past meant significant gains (Lovingly remembers the old pencil mod on my HD 4850 lol)

 

[tps3443]

I suppose we could argue maybe that the drivers are more optimised from the get-go nowadays leaving less wiggle room. Also, most modern cards are pushed to the max at stock (undervolting helps efficiency) whereas overclocking in the past meant significant gains (Lovingly remembers the old pencil mod on my HD 4850 lol)

Overclocking today still offers MASSIVE gains. I get huge numbers from my hardware. Some more than others. I’ve been overclocking since well before the ATI 4850. But yeah you can get great performance out of GPU’s and CPU’s. My last generation 2080Ti was a whopping 30% faster than a RTX3070 in rasterization performance benchmarks like Firestrike or Timespy. That was a GPU that matches and even beats the 2080Ti. The same goes for RTX3090’s. You can squeeze a solid 18-23%. My 11900K CPU scores 25% higher Cinebench multithreaded scores than the PC Gamer magazine review of it. Overclocking is alive and well, it just changes. now a days it’s all about having low temps and good cooling.

 

[Dr. Dro]

Interesting, good to see free performance improvements on offer, however slight. I've already installed it.

My venerable 780 Ti supports DX12, so it would have also been interesting to see if that had any performance improvements too. Alas, the card is no longer supported with driver updates so that's not happening.

I don't believe many gains would show up, Kepler doesn't really support DirectX 12, it's restricted to feature level 11_0 so only the subset of hardware accelerated features supported by DirectX 11.0 (the original Windows 7 DX11) are available, its DDI was simply updated to support D3D12 instances, the same goes for Fermi... and they are rather limited at that.

The first-generation GCN would support 11_1 features and from GCN 2 onwards, basic true DX12 support is available with 12_0 feature level. NVIDIA went from 11_0 straight to 12_1 with Maxwell 2 (GM200 and GM204), even though AMD fans would joke NVIDIA did not support async compute (they did; the hardware was just hilariously bad at it) - it was GCN that was slightly behind in features. Alas, even today it is not rare that "DX12" games run on Kepler or the original GCN, just because the hardware features aren't actually required by the engine. Death Stranding and Red Dead Redemption 2 are the only games on the top of my head that do not boot on 11_0 and 11_1 hardware.

Why are people so obsessed with this game? Is terrible programming is now a reference benchmark to represent the typical PC gaming experience?

That game is so good! But at the same time, I suppose people also want to benchmark Cyberpunk 2077 (which is an even worse mess and unlike Horizon, it is not even a good game) ¯\_(ツ)_/¯

 

[nguyen]

That game is so good! But at the same time, I suppose people also want to benchmark Cyberpunk 2077 (which is an even worse mess and unlike Horizon, it is not even a good game) ¯\_(ツ)_/¯

CP2077 is extremely optimized for PC (high end PC to be precise), meanwhile Horizon is a shitty port (Death Stranding was much better port)

 

[Dr. Dro]

CP2077 is extremely optimized for PC (high end PC to be precise), meanwhile Horizon is a shitty port (Death Stranding was much better port)

LOL, we must not have played the same game. The one that would crash every five minutes, with Big Rigs quality animation, tons of script errors and low teens frame rate unless I enabled DLSS with a then brand new 3090... High end PCs being able to power through the mess after almost 2 years of fixes (hey at least they didn't hit and run) doesn't mean it's optimized.

Cyberpunk left an extreme negative impression on me.

 

[nguyen]

LOL, we must not have played the same game. The one that would crash every five minutes, with Big Rigs quality animation, tons of script errors and low teens frame rate unless I enabled DLSS with a then brand new 3090... High end PCs being able to power through the mess after almost 2 years of fixes (hey at least they didn't hit and run) doesn't mean it's optimized.

Cyberpunk left an extreme negative impression on me.

I played like 100h+ of CP2077 a week after it launched, almost flawless experience, the visual more than make up for low but smooth FPS. CP2077 will crash within minutes if you have unstable overclock, it is the best tool to find the stable clock for me ever since.

Horizon had terrible shader compilation stutters and horrible looking TAA (grass shimmering) for a long time before they were fixed, so yeah, high and stuttery FPS with shitty visual is nothing to compliment about.

Anyways there is Death Stranding in the test suit, no need for Horizon to be included

 

[Dr. Dro]

I played like 100h+ of CP2077 a week after it launched, almost flawless experience, the visual more than make up for low but smooth FPS. CP2077 will crash within minutes if you have unstable overclock, it is the best tool to find the stable clock for me ever since.

Horizon had terrible shader compilation stutters and horrible looking TAA (grass shimmering) for a long time before they were fixed, so yeah, high and stuttery FPS with shitty visual is nothing to compliment about.

Anyways there is Death Stranding in the test suit, no need for Horizon to be included

Yeah nah, back at launch at least it wasn't overclocking, just the game's exceptionally poor programming. Crowbcat video on Cyberpunk was exactly the experience I had, and at the time I was running the exact same specs I have today except I had a 3900XT processor. I might give it another chance after the expansion packs release, but to call it bad was being nice

 

[lexluthermiester]

it should be possible to benchmark battlefield 2042 there is a conquest-solo mode
Or a custom portal mode

That requires an online presence. W1zzard doesn't test that way.

Why are people so obsessed with this game? Is terrible programming is now a reference benchmark to represent the typical PC gaming experience?

I'm with them on that one, it's a fun and popular game. Easy to test with. And if it is poorly programed, that would make it an excellent benchmark.

 

[GameZenChill]

Thanks for doing these Driver Performance Analysis benchmarks !

I'm kind of new here. Are these a regular feature or something that happens if there's a major impact ?