Nvidia Kepler performance driver fix released.

[Gryz]

Not bad.
Seems like a almost 10% fps increase in The Witcher 3 on my gtx680.

I loaded my last 10 savefiles in TW3. Kept standing still after the savegame loaded. Made a note of my fps. Loaded the next savefile. Repeat 10 times. Then I installed the new drivers. And did it again. Very crude benchmarking. But in all those 10 cases my fps was ~10% higher. I had cranked up my settings up until my fps was 35-50 in most cases. Usually running around 40 fps. Now, with the new driver, it's more like 44 fps average. So ~10%. Maybe a little less.

I was planning on buying a new gtx980ti. That card seemed to be ~2.2x faster than my gtx680. That's now down from 2.2x to 2x. 750 Euro was supposed to give me 120% more framerates. Now it'll be only 100% more framerates. 1/6th of my gain evaporated. That means 750/6 = 125 euros already down the drain.

 

[n0x1ous]

a good 15% improvement in Witcher 3 on 780 SLI for me.....enough to keep me locked at 60 with everything ultra hairworks off @1440p

On the previous driver couldn't hit 60 consistently.

Nice job Nvidia driver team.

 

[Silverforce11]

So the theory that Kepler tanks due to usage of modern game engines focus on compute... doesn't hold. It just tanks cos it wasn't optimized. The focus shifted to Maxwell. There's plenty of rage from Kepler owners, so NV has to comply. Gamers win.

 

[happy medium]

So the theory that Kepler tanks due to usage of modern game engines focus on compute... doesn't hold. It just tanks cos it wasn't optimized. The focus shifted to Maxwell. There's plenty of rage from Kepler owners, so NV has to comply. Gamers win.

Kepler release date March22, 2012. Its been over 3 years, you wonder why they optimize for Maxwell first. Maxwell has been out almost a year now.

GDC 1.0, AKA 7970 series release date Jan 2012. When AMD releases their new architexture AKA Fury I'm sure they will optimize for it first and foremost also. DO you think they still will be optimizing the 7 series a year from now first?

 

[Silverforce11]

How old are the 780Ti and Titan Z?

When AMD's new stack come out, I fully expect them to still optimize for R280/X, R290/X etc. We certainly won't see 360X or 370X beating R290X like we've seen 960 spanking Titan and 780s.

 

[Rvenger]

Until AMD moves off of the GCN Architecture all GCN gpus will still see optimizations.

 

[Silverforce11]

Until AMD moves off of the GCN Architecture all GCN gpus will still see optimizations.

Yes, and until new gen of consoles launch with a new uarch, GCN will still be the focus of game engines due to cross-platform development being the norms.

 

[Rvenger]

Nvidia did Kepler users a good service by optimizing the drivers for recent games though, I am not certain what it will be like when we move onto DX12 and how much Kepler will support DX12.

 

[96Firebird]

So the theory that Kepler tanks due to usage of modern game engines focus on compute... doesn't hold. It just tanks cos it wasn't optimized. The focus shifted to Maxwell. There's plenty of rage from Kepler owners, so NV has to comply. Gamers win.

A 10% increase doesn't even put the 780 Ti above a 290...

http://gamegpu.ru/rpg/rollevye/the-witcher-3-wild-hunt-v-1-04-test-gpu.html

 

[boozzer]

A 10% increase doesn't even put the 780 Ti above a 290...

http://gamegpu.ru/rpg/rollevye/the-witcher-3-wild-hunt-v-1-04-test-gpu.html

damn, 780 ti is still below a 290 even with the new driver? that is just damn sad. D:

 

[96Firebird]

damn, 780 ti is still below a 290 even with the new driver? that is just damn sad. D:

That test wasn't run with the new driver, but it is the latest patch. I'm just saying that even adding the 10% gain seen by our fellow forum members, Kepler would still end up sub-par to what it used to be.

It's good that progress has been made, however...

 

[Silverforce11]

That test wasn't run with the new driver, but it is the latest patch. I'm just saying that even adding the 10% gain seen by our fellow forum members, Kepler would still end up sub-par to what it used to be.

It's good that progress has been made, however...

It's not too bad when you turn on HairWorks, it runs close to the 970.

Notice how AMD with the Tessellation override, can enjoy HairWorks with little performance loss.

NV's Control Panel should have such an option to give gamers a choice and more tweaking options.

 

[ocre]

So the theory that Kepler tanks due to usage of modern game engines focus on compute... doesn't hold. It just tanks cos it wasn't optimized. The focus shifted to Maxwell. There's plenty of rage from Kepler owners, so NV has to comply. Gamers win.

if the 780ti still isn't punching at the 290x levels, even after the patch and driver, I just don't see how you concluded that at all.

Regardless, I think the issue is a little larger than differed lighting or compute.
But even if it was just one thing, its naive to think that there is absolutely no way to improve performance.

There is almost always room for improvement. It may just be in tiny bits at a time. it is just a matter of the resources and trade offs. When the low hanging fruit is picked, the task just becomes harder and harder. How many man hours do you have and at what cost will the returns be too small to matter.

I can't say that nvidia sabotaged Kepler on purpose, I really think that is far fetched. But obviously the complaints got them to focus on Kepler performance. They got results in witcher 3. But I am not sure they will easily fix the larger issue of Kepler falling behind. The good thing is, the work done for witcher 3 should help with any future game that uses that engine.

I am glad to see them listen but I just don't see there being a magic fix for all of Kepler woes.

 

[RussianSensation]

Regardless, I think the issue is a little larger than differed lighting or compute.

It shouldn't matter what the reason is. As a consumer, if one spends $699 for a flagship card in November 2013, he/she should expect it to mop the floor with a $399 R9 290 in 95% of games and remain so for the next 2 years. Today, $650 780 is getting killed by a $400 R9 290, the card it was meant to compete with, while 780Ti is slower than R9 290X. It's so bad that 980Ti at $649 is 75-80% faster than a 780, which I guess is a good upgrade for 780 owners!

GTA V
$399 R9 290 > $699 780Ti
$299 R9 280X > $650 780
$649 980Ti is 65% faster than 780Ti, 83% faster than 780! WOW

TW3
980Ti is 91% faster than 780Ti, and 110X (2.1X) faster than the 780! Shocking.
R9 290 > 780Ti

Kepler flops -- I am glad I didn't buy into the Kepler perf/watt marketing hype 3 years ago. Today $1000 690 < single $550 7970 in many games cuz it's choking on VRAM too. Against HD7970Ghz CF, 690 flops so hard, it's hard to believe really. Waste of $1000. In hindsight, the OG Titan comes out looking good compared to the awful performance we now see in 690/680 2GB SLI.

HD7990 56% faster than GTX690 in SoM

Even a single 925mhz 7970 is smoother than 690 in SoM.

HD7970 > 690 in Civilization Beyond Earth

HD7990 is 30% faster in FC4, a game known to have major issues with CF for months....

HD7990 is 2.39X faster in Total War: Attila

HD7990 is 31% faster in Grid AutoSport

GTA V performance on 690 is a disaster.

I guess TPU takes it easy on the GTX690 since they test games with lower levels of MSAA compared to a site like AT.

Kepler went from a darling of last generation to a major fail imo. There is no way to declare Kepler the winner of last generation anymore after the pounding 680/770 receive from 280X, 690 from 7990 and 780/780Ti from 290/290 given all their respective prices and all.

Kepler in retrospect was:
- Overpriced in price/performance - check
- VRAM gimped SKUs - check
- Inferior SLI scaling vs. HD7990 or GTX780TI vs. R9 290X CF - check. Once AMD fixed smoothness, HD7990 is also smoother than SLI
- Performance that falls off a cliff in the last 8 months - check

 

[RussianSensation]

damn, 780 ti is still below a 290 even with the new driver? that is just damn sad. D:

$1400 780Ti SLI < $800 R9 290 (reference models!) CF

It's amazing how certain people just keep buying the same brand over and over, and constantly discuss how SLI is "superior" to CF, even after seeing stuff like this. I guess it's easy to ignore now since well Maxwell is the it thing for the moment. Imagine if in 2 years from now a $1300 GTX980Ti SLI setup gets beaten by $800 R9 390X CF? Wouldn't that be a shocker? It would be to me.

Luckily Maxwell seems to be a lot better on paper, both in compute, DX12 support, 980Ti has 6GB of VRAM which is much closer to the current consoles than 780TI's 3GB was. Maxwell should be a lot better prepared for UE4 and DX12 games. Once DX12 games start coming out, I expect Kepler to get even worse vs. R9 290X, although by that point I have a feeling either of those cards will be too slow, but R9 295X2's lead over 780TI SLI should continue to grow with time.

 

[Silverforce11]

$1400 780Ti SLI < $800 R9 290 (reference models!) CF

It's amazing how certain people just keep buying the same brand over and over, and constantly discuss how SLI is "superior" to CF, even after seeing stuff like this. I guess it's easy to ignore now since well Maxwell is the it thing for the moment. Imagine if in 2 years from now a $1300 GTX980Ti SLI setup gets beaten by $800 R9 390X CF? Wouldn't that be a shocker? It would be to me.

Those would move onto Pascal, they won't have to worry about lack of optimizations or game engines not designed for older tech etc.

They are happy with their purchase. You don't need to try to convince them otherwise.

 

[RussianSensation]

Those would move onto Pascal, they won't have to worry about lack of optimizations or game engines not designed for older tech etc.

They are happy with their purchase. You don't need to try to convince them otherwise.

Ya, I am starting to think the business model of 2-year-GPU architecture life-cycles suits NV very well. It's like the smartphone 2-year contract model, without the official contract. Either way, looks like NV is converting an awful lot of gamers to Maxwell. If they cripple Kepler cards, they know that their gamers will just move on to Maxwell and not even consider any R9 200 series cards. The types of gamers who pay attention to how fast a GPU architecture falls apart or how the driver optimizations have stopped are a tiny minority of the entire GPU buying market. This could be very dangerous for the GPU industry since a lot of those gamers will just keep rolling over to the next generation of NV cards which means AMD could be stuck in the 20% market share or less for the next 5 years.

Knowing how competitive Maxwell cards are, it gives NV almost no incentive at all to optimize the drivers for Kepler. What's the point when there is so little risk that your loyal customers will abandon you? It's a win-win for NV since they make more $ crippling older generations and once their loyal customers upgrade and spread word-of-mouth experience of how amazing Kepler is, it gets brand agnostic gamers to upgrade too.

 

[ocre]

It shouldn't matter what the reason is. As a consumer,

The reasons shouldnt matter? Conspiracies and people making up and inventing evil plots as reasoning?

I guess that is what makes you and me very different.

I think we should have deeper conversations that dive into the architectures.

To me it matters. Spreading misinformation and twisting conspiracy theories out of nothing but will, i dont think they have much place.
The fact that kepler is falling behind, it is a result of many things. When we try to talk about some of them individually, people seem to put great effort in trying to force the conversation back into a jumbled mess.
All the while, there are people who keep insisting that nvidia purposefully is gimping kepler performance in order to force people to buy maxwell cards.

I find it a little bizarre that you would think that last part is just fine and you are cool with that. You are one of the few that might actually be able to see some of the deeper reasons underneath kepler falling behind.

In your ranting, especially since you bring up the 290(x), it is odd that you dont seem to be aware of your total disservice to AMD. You completely ignore the work and huge gains they achieved. The omega drivers alone gave them gains up to 20%. That is huge and completely ignored.

Of course, it may not matter to you. But as an fan of PC, i think it is really really impressive. So, why dont we take the time out of the nvidia (& customer) bashing and give AMD a hand. They did a great job and we eventually start to see GCN performing to its full potential. You may insist we all should ignore that, I would rather applaud AMD for their great achievements. Ignoring it is a disservice to them.

just a note, my opinion:
Bashing nvidia and their customers will not win over mindshare. It hasnt worked and i dont think it ever will. It is just messy and ineffective.

Compared to you, I see things different i guess....

 

[Pariah]

Ya, I am starting to think the business model of 2-year-GPU architecture life-cycles suits NV very well. It's like the smartphone 2-year contract model, without the official contract. Either way, looks like NV is converting an awful lot of gamers to Maxwell. If they cripple Kepler cards, they know that their gamers will just move on to Maxwell and not even consider any R9 200 series cards. The types of gamers who pay attention to how fast a GPU architecture falls apart or how the driver optimizations have stopped are a tiny minority of the entire GPU buying market. This could be very dangerous for the GPU industry since a lot of those gamers will just keep rolling over to the next generation of NV cards which means AMD could be stuck in the 20% market share or less for the next 5 years.

Knowing how competitive Maxwell cards are, it gives NV almost no incentive at all to optimize the drivers for Kepler. What's the point when there is so little risk that your loyal customers will abandon you? It's a win-win for NV since they make more $ crippling older generations and once their loyal customers upgrade and spread word-of-mouth experience of how amazing Kepler is, it gets brand agnostic gamers to upgrade too.

You're not making any sense. If these users are on a 2 year upgrade cycle, why would they waste their time researching the aging arch of an architecture or how well it is supported 3 years down the line, when they have already moved on? That's like expecting a person who habitually replaces their car every 4 years, to not consider a car with a 7 year powertrain warranty, because the competition has a 10 year warranty.

 

[3DVagabond]

The reasons shouldnt matter? Conspiracies and people making up and inventing evil plots as reasoning?

If it walks like a duck and quacks like a duck, it's probably a duck. How many times is stuff like this going to happen before people accept that it's SOP for nVidia? It's not a coincidence or circumstances beyond their control. It's a determined effort.

 

[Flapdrol1337]

Just tested this new driver in the witcher 3 with my 670

ran around in the nilfgaard camp, old driver dropped to 30, new to 34. median 35 ish to 39. Seems like an improvement.

Looks like the internet outrage worked.

 

[SirPauly]

I expect Kepler to get even worse vs. R9 290X, although by that point I have a feeling either of those cards will be too slow, but R9 295X2's lead over 780TI SLI should continue to grow with time.

Probably, and has happened before with ATI/AMD's x19xx generation compared to nVidia's 79xx generation.

 

[Face2Face]

I'll have the try out the new driver tonight before I fire up TW3. Regardless, with Kepler performance not looking so hot in future titles, I may retire my GTX 780 and give it to my son. I'll be looking really hard at Fiji when it's released. I just don't want to spend $600+ on a video card. Hopefully, there will be a compelling option at $400-450 USD.

EDIT:

Here is a YouTube video showing the driver performance with a single GTX 780Ti.

https://www.youtube.com/watch?t=93&v=aIG791SQdIw

 

[DarkKnightDude]

Certainly improved FPS for me in Witcher 3.

 

[Gikaseixas]

Any improvements on Project CARS?????
Getting ready to sell my laptop