• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Is Nv just stopped optimizing Kepler GPU's in their drivers?

pawel86ck

Banned
Picture tells more than 1000 words, so I will just post hardware unboxed benchmark results, and you guys will see for yourself what I'm talking about.
https://www.techspot.com/review/1795-metro-exodus-benchmarks/

a2233.png


For many years 780ti was on the same level as GTX 970, but in Metro Exodus GTX 970 almost doubled minimal framerate compared to 780ti (it was kepler top dog). What's next Nv, you will forget about 980ti users soon, and then about 1080ti users? What's interesting are also AMD 7970 results, it's still holds up very well despite it's age and in this particular game it's even faster than kepler top dog 780ti! So AMD even after so many years still support their old GPU correctly! It's not nice Nv, not only you make your cards much more expensive and not future proof (just 8GB and 11GB in their "Turing" Top Dogs? That's a joke if people will consider PS5 launch next year and VRAM requirements will skyrocket), but now we also have a proof they dont care supporting their old cards. I feel bad because I bought only Nvidia GPU's for so many years, but next time I will think twice before buying my next GPU.

And BTW- metro exodus is game made with current gen consoles in mind first, PC version besides raytracing still looks like the same game. So I wonder why even good GPU like 1070 is suddenly crushed even at 1080p why even xbox X runs that game at 4K?
 
Last edited:

ethomaz

Banned
I replied in another thread... there is too much variables.

It could be memory, it could the game using features that speed up 970 not present in 780, it could be better DX12 drivers on 970, could be a lot of things.
 
Last edited:

pawel86ck

Banned
I replied in another thread... there is too much variables.

It could be memory, it could the game using features that speed up 970 not present in 780, it could be better DX12 drivers on 970, could be a lot of things.
OK, can you think about certain feature missing in 780ti kepler compared to 970 Maxwell that could explain so MASSIVE difference? For so many years we could see very similar performance between these two cards, and suddenly BAM there's metro and 780ti looks like a 100 years old grandpa compared to 970 :D.

I'm aware tesselation performance was massively improved on Maxwell GPU's compared to Kepler, but that's exactly why I have linked results without tesselation. It also cant be just memory difference, because 7970 also has 3GB vram, yet it was also faster than 780ti (I would never thought 7970 will be faster than 780ti, because back in the days Nv have said even GTX 680 was on 7970 performance level).
 
Last edited:

ethomaz

Banned
I need to check the difference between the Archs but I have a question first... is the nVidia drivers already optimized to Metro?
 
Picture tells more than 1000 words, so I will just post hardware unboxed benchmark results, and you guys will see for yourself what I'm talking about.
https://www.techspot.com/review/1795-metro-exodus-benchmarks/

a2233.png


For many years 780ti was on the same level as GTX 970, but in Metro Exodus GTX 970 almost doubled minimal framerate compared to 780ti (it was kepler top dog). What's next Nv, you will forget about 980ti users soon, and then about 1080ti users? What's interesting are also AMD 7970 results, it's still holds up very well despite it's age and in this particular game it's even faster than kepler top dog 780ti! So AMD even after so many years still support their old GPU correctly! It's not nice Nv, not only you make your cards much more expensive and not future proof (just 8GB and 11GB in their "Turing" Top Dogs? That's a joke if people will consider PS5 launch next year and VRAM requirements will skyrocket), but now we also have a proof they dont care supporting their old cards. I feel bad because I bought only Nvidia GPU's for so many years, but next time I will think twice before buying my next GPU.


nvidia be like that. they stop optimizng new games for their older GPUs and the games run as if the GPU deteriorated.
 
Last edited:

dogen

Member
OK, can you think about certain feature missing in 780ti kepler compared to 970 Maxwell that could explain so MASSIVE difference? For so many years we could see very similar performance between these two cards, and suddenly BAM there's metro and 780ti looks like a 100 years old grandpa compared to 970 :D.

I'm aware tesselation performance was massively improved on Maxwell GPU's compared to Kepler, but that's exactly why I have linked results without tesselation. It also cant be just memory difference, because 7970 also has 3GB vram, yet it was also faster than 780ti (I would never thought 7970 will be faster than 780ti, because back in the days Nv have said even GTX 680 was on 7970 performance level).


Check these compute benches.
https://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/20

Maxwell was hugely improved in this area, and Metro likely uses lots of compute shaders.
 

Leonidas

Member
OK, can you think about certain feature missing in 780ti kepler compared to 970 Maxwell that could explain so MASSIVE difference? For so many years we could see very similar performance between these two cards, and suddenly...

It's not sudden, this has been known for many years. But it only affects certain games, in many games 780 Ti is still a bit faster than 7970 as it used to be.
 

pawel86ck

Banned
I need to check the difference between the Archs but I have a question first... is the nVidia drivers already optimized to Metro?
In drivers release highlights I can read they have provided optimal gaming experience when using DLSS and Ray Tracing in Battlefield V and Metro Exodus, so I'm guessing they have also optimized everything related to this particular game.
 

Pagusas

Elden Member
If they did I see no problem with it, they shouldn’t support legacy hardware with resources past a certain date, the 780’s time is more than up. It’s been over 5 years.
 
Last edited:

pawel86ck

Banned
It's not sudden, this has been known for many years. But it only affects certain games, in many games 780 Ti is still a bit faster than 7970 as it used to be.
So what certain games exactly, because until now I wasnt aware about games like that

Check these compute benches.
https://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/20

Maxwell was hugely improved in this area, and Metro likely uses lots of compute shaders.
Interesting, but none of these benchmark shows so massive difference as presented in Metro Exodus.

f they did I see no problem with it, they shouldn’t support legacy hardware with resources past a certain date, the 780’s time is more than up. It’s been over 5 years.
The thing is, AMD still optimized their ancient 7970, so why Nv cant do the same?
 
Last edited:

dogen

Member
So what certain games, because until now I wasnt aware about games like that


Interesting, but none of these benchmark shows so massive difference as presented in Metro Exodus.


The thing is, AMD still optimized their ancient 7970, so why Nv cant do the same?

Because the 7970 is essentially the same architecture as their newer GPUs. Kepler isn't really.

I wouldn't expect those benchmarks to behave exactly like metro anyway. Maybe the game hits those bottlenecks harder, or entirely different ones. No way to know really.
 
Last edited:

Leonidas

Member
Odd how TechSpot tested Metro with the old cards only with tessellation disabled...

So what certain games, because until now I wasnt aware about games like that

It sounds like you also weren't aware of Kepler aging badly in certain games. The fact is that most old cards simply aren't tested that often today, but if you did a little research you could easily find that 780 Ti bests 7970 in many modern games.

Some of those games include:

Battlefield 1
Dirt 4
For Honor
Ghost Recon Wildlands
Mass Effect Andromeda
Prey
Resident Evil 7
Gears of War 4

Nobody ever posts about that though, people only post when 780 Ti looks bad.
 
Last edited:

pawel86ck

Banned
It sounds like you also weren't aware of Kepler aging badly in certain games. The fact is that most old cards simply aren't tested that often today, but if you did a little research you could easily find that 780 Ti bests 7970 in many modern games.

Some of those games include:

Battlefield 1
Dirt 4
For Honor
Ghost Recon Wildlands
Mass Effect Andromeda
Prey
Resident Evil 7

And many others, it's just something that's not tested often as these cards are very old...
Battlefield 1? According to results from the same site (techspot/hardwareunboxed), 970 scores in 1440p 56fps, while 780ti 55fps, so that's only 1 fps difference.

It's harder to find results from other games on 780ti, but I have found mass effect andromeda benchmark with 780ti results on GAMEGPU site and in 1440p there was only 3 fps difference between these two cards.

Maybe it will be faster if I will just ask you where we can see a massive difference similar like in Metro Exodus because certainly battlefield and mass effect andromeda are not games like that. In metro exodus minimal fps is cut down by half, that's massive difference, not just 3fps difference.
 

Leonidas

Member
Battlefield 1? According to results from the same site (techspot/hardwareunboxed), 970 scores in 1440p 56fps, while 780ti 55fps, so that's only 1 fps difference.

It's harder to find results from other games on 780ti, but I have found mass effect andromeda benchmark with 780ti results on GAMEGPU site and in 1440p there was only 3 fps difference between these two cards.

Maybe it will be faster if I will just ask you where we can see a massive difference similar like in Metro Exodus because certainly battlefield and mass effect andromeda are not games like that. In metro exodus minimal fps is cut down by half, that's massive difference, not just 3fps difference.

First of all I'm not comparing the 970 vs 780 Ti at all. I stated "780 Ti bests 7970 in many modern games." and that is still a true statement even to this day.

If you saw those results at GameGPU you either were comparing different cards or don't know which cards are comparable.

280x is on par with a 7970 GHz Edition(overclocked 7970). 780 Ti is much faster in that case.

In this example(which you pointed out as being a 3 fps difference) 780 Ti is 50-60% faster than 7970. Where did you get 3 FPS difference from?

me4_1920.png
me4_2560.png
 
Last edited:

pawel86ck

Banned
First of all I'm not comparing the 970 vs 780 Ti at all. I stated "780 Ti bests 7970 in many modern games." and that is still a true statement even to this day.

If you saw those results at GameGPU you either were comparing different cards or don't know which cards are comparable.

280x is on par with a 7970 GHz Edition(overclocked 7970). 780 Ti is much faster in that case.

In this example(which you pointed out) 780 Ti is 50-60% faster than 7970.
Ok I have read your post for a 2'nd time and I can see now you were talking about 7970 vs 780ti performance difference. But you still wrote Kepler aged badly in certain games so I would like to know if you are aware of any particular game that would show similar difference as in "metro exodus" (where GTX 970 almost doubled 780ti results). There are other games like that as well, or metro exodus is really the first as I suspected?
 
Last edited:

Leonidas

Member
Ok I have read your post for a 2'nd time and I can see now you were talking about 7970 vs 780ti performance difference. But you still wrote Kepler aged badly in certain games so I would like to know if you are aware of any particular game that would show similar difference as in "metro exodus" (where GTX 970 almost doubled 780ti results). There are other games like that as well, or metro exodus is really the first as I suspected?

Kepler performing badly compared to modern GPUs in certain games is well known. The example in the OP is strange to me, disabled tessellation and the 970 is a bit faster than 390. That's just a very good result for 970, a below average result for 390 and a very bad result for 780 Ti given the game and settings chosen.
 
Last edited:
Are there tests with older drivers? Nothing stops one from going back a few years and utilizing 780's best drivers vs current drivers to get to the bottom of what is probably a nefarious activity from nvidia. I wouldn't put it past them. Like Apple, it pays to cripple old hardware so your zombie consumers buy the next iteration.
 
Last edited:

pawel86ck

Banned
Kepler performing badly compared to modern GPUs in certain games is well known. The example in the OP is strange to me, disabled tessellation and the 970 is a bit faster than 390. That's just a very good result for 970, a below average result for 390 and a very bad result for 780 Ti given the game and settings chosen.
Well, I can agree kepler was no longer performing good next to modern GPU's, but as you can see in techspot performance analysis Metro Exodus it's a totally different beast. When GTX 970 almost doubles 780ti then it's obviously something wrong. Because not even the difference in compute power suggest such drastic difference, so I really belive Nv simply no longer optimize games for kepler at all.

Are there tests with older drivers? Nothing stops one from going back a few years and utilizing 780's best drivers vs current drivers to get to the bottom of what is probably a nefarious activity from nvidia. I wouldn't put it past them. Like Apple, it pays to cripple old hardware so your zombie consumers buy the next iteration.
Techspot Metro Exodus performance article mention all Nvidia cards were tested with 418.91 drivers.
 

Leonidas

Member
Well, I can agree kepler was no longer performing good next to modern GPU's, but as you can see in techspot performance analysis Metro Exodus it's a totally different beast.

Why does it matter so much to you that a seven year old GPU architecture is not performing well in some cases today?

You're talking about a 6 year old card that no one uses today.
 
Last edited:

JohnnyFootball

GerAlt-Right. Ciriously.
Are there tests with older drivers? Nothing stops one from going back a few years and utilizing 780's best drivers vs current drivers to get to the bottom of what is probably a nefarious activity from nvidia. I wouldn't put it past them. Like Apple, it pays to cripple old hardware so your zombie consumers buy the next iteration.
You realize that such a test wouldnt work for Metro Exodus right? There is simply no way Kepler era drivers would function properly on a recent game like Metro. Now such a test could work using an older game like say Crysis 3. But on Metro Exodus forget about it.
 
You realize that such a test wouldnt work for Metro Exodus right? There is simply no way Kepler era drivers would function properly on a recent game like Metro. Now such a test could work using an older game like say Crysis 3. But on Metro Exodus forget about it.


No guarantee that they wouldn't work with Metro. May even be able to modify them to make them support the game properly if they run into some problems.
 

pawel86ck

Banned
Why does it matter so much to you that a seven year old GPU architecture is not performing well in some cases today?

You're talking about a 6 year old card that no one uses today.

7970 still performs well because AMD sold 7970 under a different name just a few years ago and AMD's architecture has not changed anywhere near the amount that Nvidia's has in the same time frame...
Then look at steam stats, and you will see people still use kepler GPU's. Years ago (especially in early 2000) when PC technology was advancing much faster 7 years GPU would be indeed a long time, but "thanks" to consoles requirements in all multiplatform games literally remain on similar level since PS4 launch. For that reason I have still used kepler GPU 1.5 year ago before I bought 1080ti and I still have nostalgia for that card (the same for every other card I had and I'm always interested how well my GPU's fare even few years later on).

Also based on kepler example I can see, Nv may do the same thing with Maxwell GPU's next, and MANY people still use maxwell GPU's (970 and 980ti especially) . Although I've always bought Nvidia GPU's I'm not a Nvidia fanboy and I can say objectively speaking what they are doing is not fair. Not only they make overpriced cards with small vram amount (next year PS5 will launch and I wonder what Turing owners will do then, especially on 6GB and 8GB cards? They will be forced to lower texture settings very soon), but now it looks like they also gimp their old cards performance. What's even worse, some of you guys dont even consider Nv have done it on purpose. 970 is twice as fast as 780ti? That's must be because of that famous compute difference, there's no other possible explanation because of course Nv care about people and they would never do something like that :D. We live in a strange times, companies gimp their products on purpose and people not only see no problems with that but even defend companies like Nvidia.
 
Last edited:

JohnnyFootball

GerAlt-Right. Ciriously.
No guarantee that they wouldn't work with Metro. May even be able to modify them to make them support the game properly if they run into some problems.
Dude, there is no way that drivers from 2013 would come close to giving Metro Exodus a fair shake. The game might (thats a big might) run, but it wouldnt be stable and many in game features wouldnt work due to being unsupported at the time. The only way such a test could be run is if you compare a game that was released at the time of the 2013 drivers. Another issues you run into with older drivers is that you would also likely need a CPU/mobo from around that time frame too since there is a good chance that 2013 nvidia drivers wouldnt be compatible with Skylake era CPUs
 
Dude, there is no way that drivers from 2013 would come close to giving Metro Exodus a fair shake. The game might (thats a big might) run, but it wouldnt be stable and many in game features wouldnt work due to being unsupported at the time. The only way such a test could be run is if you compare a game that was released at the time of the 2013 drivers. Another issues you run into with older drivers is that you would also likely need a CPU/mobo from around that time frame too since there is a good chance that 2013 nvidia drivers wouldnt be compatible with Skylake era CPUs

There are custom drivers built by enthusiasits that might fit the testing bill, giving all cards a fair shake of the sauce bottle.
 

pawel86ck

Banned
780 Ti isn't even on there. Even the $1200 RTX 2080 Ti is more widely used in those stats than 780 and 780 Ti.

760(highest Kepler card in the list) has less than 0.9% in those stats.
https://store.steampowered.com/hwsurvey/videocard/

I can see multiple kepler cards on this list, and not to mention there's still many maxwell cards owners, so like I have said. If kepler generation is forgotten by Nv today so people can expect Nv will do the same thing next with Maxwell cards. Then 970/980/980ti owners will be soon forced to buy new GPU. In fact there are already examples where maxwell card like 980ti is beaten by 1060, and 1060 theoretically should never match 980ti because 980ti always was much faster card. Nv still optimize games for pascal correctly but who knows what will happen in the near future. But hey all Nvidia lovers, lets assume it should be expected, because pascal GPUs are gettings older and your logic says Nv should not care their old cards. I will be not surprised if next year RTX 2080 will crush my 1080ti even in games without raytracing.
 
Last edited:

JohnnyFootball

GerAlt-Right. Ciriously.
To address the original point of the OP:
Original Kepler came out in mid-2012 and is now approaching 6 years old. That’s ancient in GPU years and as much as despise what Nvidia is doing these days I don’t feel that they are obligated to perfectly optimize Kepler anymore.

If you still own a Kepler and are unhappy with the performance it’s now time to look for an upgrade and the good news is that significantly better GPUs are available for under $250 like the Radeon 580.
 

dirthead

Banned
The latest drivers were completely trucked for me. G-Sync didn't even work right with them. I rolled back and it's fine. Really disappointed in Nvidia. Pretty soon they and not AMD is going to be known for shitty, unreliable drivers.
 

LordOfChaos

Member
Someone should do a really good long term investigation of this. AMD FineWine and Nvidia not giving as much care to older parts are such memes but these individual data points aren't enough proof.
 

Ramzy

Member
I'm a bit confused -- what business sense would it make for Nvidia to continue tuning their drivers for a 5 year old card that isn't sold anymore? A card that doesn't even meet the minimum spec requirements of the game that you're attempting to play?

Nvidia provided years of updates and support for your card, how many more years were you expecting? 10? 15?
 
Top Bottom