Nvidia Kepler performance driver fix released.

[hawtdawg]

So the question here is why is it in neutral games, Kepler still reigns up the top, with 780Ti beating the 970 and close to 980, but in recent NV sponsored games, its well below that?

Is NV sponsoring devs to push more "innovative features" that run better on Maxwell's uarch?

No, they just aren't improving Kepler drivers for newer games. It makes Maxwell look more "efficient"

 

[Silverforce11]

No, they just aren't improving Kepler drivers for newer games. It makes Maxwell look more "efficient"

It's not that, because newer neutral or AMD games run fine on Kepler. It only tanks in NV sponsored titles.

 

[ocre]

It would be awesome if we were really asking that question but reality is most people aren't asking what is going on, they are telling us.
You know there are countless things that I could think of, some more logical than others. Any number of them could be at play here, all at once.

Yet many here have picked just one. That is all they can see. And anything that goes against it, they just put their fingers in the ears and keep repeating it.
There are holes all in this theory but it doesn't matter.

For starters, if people really believe it is some sort of sabotage to Kepler, then how? the nvidia sponsered game has encrypted code that detects kepler cards and slows down? why don't they just reflash their 780ti with a custom bios and change its identity marker to gtx980. Then the game will be tricked and the penalty code not applied. Does that sound crazy to anyone else? Cause it sure does to me.

So, if it isn't some identity seaking code that runs slower, what is being suggested here? Are you suggesting that maxwell and Kepler architectures are different, that the architecture differences at play?
This means that the game engine itself involved. So are we saying that nvidia wrote these game engines just to run bad on Kepler? Doesn't this sound a little far fetched? Why does AMD have better performance than Kepler in some if these games, that doesn't add up. Surely that is not what anyone really thinks here.

Is there any chance the performance might have something to do with the architectures being different? If not, then everyone flash their Kepler cards with a maxwell ID and their performance will magically be better

 

[Silverforce11]

@Ocre
Why isn't the better Maxwell uarch showing up in other recent games, it seems to only have a massive performance gap when NV is actively involved in that game's development. So what possible reason would it be that NV sponsor a game and it tanks for Kepler?

ps. AMD does run crap in GW titles, until its patched up, ie. Dying Light, FC4, ACU. It still runs awful in Project Cars. It runs awful in Witcher 3 with HairWorks (which can be disabled, fair enough).

Edit: Ofc the uarch of Kepler to Maxwell differs. The question is why that difference doesn't result in a massive performance gap like we see in NV sponsored games when it comes to non-NV titles? Just look at Dirt Rally still in development, built for GCN/console (AMD GE) and it runs excellent on Kepler.

 

[bystander36]

@Ocre
Why isn't the better Maxwell uarch showing up in other recent games, it seems to only have a massive performance gap when NV is actively involved in that game's development. So what possible reason would it be that NV sponsor a game and it tanks for Kepler?

ps. AMD does run crap in GW titles, until its patched up, ie. Dying Light, FC4, ACU. It still runs awful in Project Cars. It runs awful in Witcher 3 with HairWorks (which can be disabled, fair enough).

Edit: Ofc the uarch of Kepler to Maxwell differs. The question is why that difference doesn't result in a massive performance gap like we see in NV sponsored games when it comes to non-NV titles? Just look at Dirt Rally still in development, built for GCN/console (AMD GE) and it runs excellent on Kepler.

All these NV sponsored games are pushing their GameWorks which adds in PhysX, particle physics, hairworks and all these other compute heavy things. The same things that synthetic benchmarks show Kepler tanking in. Much like the Metro games, which were heavy in compute, which cause AMD to shine while Nvidia tanked.

 

[Silverforce11]

I tried this out. Instead of dipping down to 36 FPS, I'm dipping down to 39 FPS. I'll take what I can get with this game.

http://www.geforce.com/whats-new/gu...-hunt-graphics-performance-and-tweaking-guide

I looked through that, they mention specifically only CPU PhysX is used in Witcher 3, GameWorks cloth & destruction. If its actually GPU accelerated, it means NV's CPU PhysX libraries can be selectively GPU accelerated. In this case, its a detriment since W3 is so light on the CPU and heavy on the GPU.

 

[Silverforce11]

All these NV sponsored games are pushing their GameWorks which adds in PhysX, particle physics, hairworks and all these other compute heavy things. The same things that synthetic benchmarks show Kepler tanking in. Much like the Metro games, which were heavy in compute, which cause AMD to shine while Nvidia tanked.

That's a fair analysis. There has to be a reason why its GW specific because Kepler is very strong otherwise.

 

[n0x1ous]

That's a fair analysis. There has to be a reason why its GW specific because Kepler is very strong otherwise.

but kepler performs poorly in GW games even when gameworks is turned off. thats where the sabotage theorys come in.....

 

[bystander36]

but kepler performs poorly in GW games even when gameworks is turned off. thats where the sabotage theorys come in.....

Just because much of the GameWorks flash is turned off, doesn't mean there isn't a lot of compute still in use. The Metro games were similar in that way. You could turn off accelerated PhysX and Advanced DoF, but it was still compute heavy and worked a lot better on AMD.

 

[Flapdrol1337]

http://www.geforce.com/whats-new/gu...-hunt-graphics-performance-and-tweaking-guide

I looked through that, they mention specifically only CPU PhysX is used in Witcher 3, GameWorks cloth & destruction. If its actually GPU accelerated, it means NV's CPU PhysX libraries can be selectively GPU accelerated. In this case, its a detriment since W3 is so light on the CPU and heavy on the GPU.

They say it's not selectable/tweakable in game. Not if it runs on the cpu or gpu.

But these effects run on the consoles on the (very weak) cpu's, they may have upped the effects a little bit for pc, but still no problem for any cpu, even on my pentium cpu seems the best choice.

 

[ocre]

@Ocre
Why isn't the better Maxwell uarch showing up in other recent games, it seems to only have a massive performance gap when NV is actively involved in that game's development. So what possible reason would it be that NV sponsor a game and it tanks for Kepler?

ps. AMD does run crap in GW titles, until its patched up, ie. Dying Light, FC4, ACU. It still runs awful in Project Cars. It runs awful in Witcher 3 with HairWorks (which can be disabled, fair enough).

Edit: Ofc the uarch of Kepler to Maxwell differs. The question is why that difference doesn't result in a massive performance gap like we see in NV sponsored games when it comes to non-NV titles? Just look at Dirt Rally still in development, built for GCN/console (AMD GE) and it runs excellent on Kepler.

I think the older game engines do fine on kepler. Break it down on a game by game basis. There are new games coming out that are built off older engines. Then there are games that come out on a brand new engine built entirely around GCN consoles in mind. This is where the maxwell SM of 2 blocks of 64 cuda cores that share L1/cache is much better suited than the 192 SMX of kepler. That entire 192 sp SMX has to share the same L1/cache/resources. I think that it is a scheduling issue where kepler SMX is drastically different than GCN and Maxwell.

the architectures are the difference here.
Surely if much of the game engines are overhauled, the performance of kepler can be improved. It is not that the 790ti got weaker, it is just that it is starving and chocking out.

Rewriting the game engine is not something easy or that nvidia does. If you can say that the architectures have something to do with the performance penalty on kepler, then look at maxwell vs GCN vs kepler and really really think about it

 

[richiegore]

Thinking about sidegrading to a 290x from my 780ti... I know its not worth it but so annoyed with the performance issues and worried it will get worse...

 

[AnandThenMan]

So when is this fix coming?

 

[Hitman928]

http://www.nvidia.com/download/driverResults.aspx/85823/en-us

"optimizations and bug fixes which can provide increased performance for Kepler-based GPUs."

 

[happy medium]

http://www.nvidia.com/download/driverResults.aspx/85823/en-us

"optimizations and bug fixes which can provide increased performance for Kepler-based GPUs."

Looking at the gtx980ti review on Anandtech, the gtx780ti did fairly well with this driver.

 

[Hitman928]

Looking at the gtx980ti review on Anandtech, the gtx780ti did fairly well with this driver.

Anandtech didn't use this driver for the 980 Ti review. . .

 

[jj109]

I installed the driver, restarted, and booted up Witcher 3.

Looks like a decent 20% frame rate bump.

 

[happy medium]

Anandtech didn't use this driver for the 980 Ti review. . .

I automatically assumed this seeing how the performance of the gtx780ti was above the 290x and below the gtx980, that's where its supposed to be right.


Well I'm happy for the Keplar owners.
good luck

 

[Hitman928]

I automatically assumed this seeing how the performance of the gtx780ti was above the 290x and below the gtx980, that's where its supposed to be right.

The 780 Ti should be above the 290x, but it more recent games, it had fallen behind. Nothing has changed in the Anandtech review. Anandtech also still uses a reference 290x which usually has worse performance than the 3rd party (non-oc) cards that other sites use.

 

[darckhart]

I've been running 347.25 for the longest time. I sure hope these do something for Kepler and older games too not just the brand new AAA titles.

 

[happy medium]

The 780 Ti should be above the 290x, but it more recent games, it had fallen behind. Nothing has changed in the Anandtech review. Anandtech also still uses a reference 290x which usually has worse performance than the 3rd party (non-oc) cards that other sites use.

I'm sure some review site will do a Keplar re-review with this new driver to shut up all the AMD conspirator theorist.

This thread has served its purpose, good luck guys.

 

[iiiankiii]

I'm sure some review site will do a Keplar re-review with this new driver to shut up all the AMD conspirator theorist.

This thread has served its purpose, good luck guys.

I actually think an improvement in Kepler will strengthen the suggestion that Nvidia has been neglecting Kepler (which is what people are speculating). Atleast the outrage lit a fire under Nvdia and forced them to do what they should have been doing all along; keeping their customers happy.

 

[Hitman928]

I'm sure some review site will do a Keplar re-review with this new driver to shut up all the AMD conspirator theorist.

This thread has served its purpose, good luck guys.

I don't know why the vitriol towards "AMD conspirator theorists". There are plenty of people from Nvidia's "camp" that are crying foul over this too. Some of us just care about getting the best for customers, not about trying to protect a brand.

 

[hawtdawg]

Whoa, these new drivers have my 780's staying locked at 60@1440p in Witcher 3 at every situation i threw at them with everything maxed except for hairworks being turned off. previously there were even situations at 1080p that would occasionally drop a few frames below 60. Now I can throw some DSR at it when Im playing on my TV!

 

[ocre]

And i trully believe that there is something that can be done in every case where kepler is way far behind. The question is, how much time and effort would it take?

See, I believe this is a per game basis. Nvidia has very clever and capable software/driver teams. But there is always the matter of resources. See, i never said that nvidia couldnt do more but i have fought hard against the idea that nvidia purposefully sabotaged kepler by malicious acts. I have always thought this to be a matter of resources. That keplers 192ccore SMX layout was at the heart of the issue. That didnt mean that nothing could be done to improve the situation. There is always something that can be done. It is just a matter of how much time and effort would it take and how feasible the task is.

It is great that this performance improvement for kepler driver has come out. I wonder what the results will be. I would put money on this not being a universal kepler driver that fixes every single game. This is because i believe that the issue is a combination and that the game engines are at play here.

In that case, there is only so much that can be done to work around it. And, tricks to gain performance in witcher, will not work in one of the other games.

Lets see the results now because i have speculated enough.

My expectations are witcher 3 improvements for sure but i dont think this one driver will universally fix every game that has had issues.