Has Nvidia Forgotten Kepler? The GTX 780 Ti vs. the 290X Revisited

[DuckDodgers]

 

Who aged better -- Kepler or Hawaii? The two former top-end architectures from 2013 have received different treatment over the their lifespan. While Nvidia simply moved along with a new GPU architecture (Maxwell), AMD's GCN have seen much less revolutionary changes, with only iterative architectural upgrades from Tahiti to the latest HBM-powered Fiji GPUs. AMD were definitely forced to spend time with the driver optimizations instead, with some notable effects. 

 

Quote

The GTX 780 Ti was released in 2013, not long after AMD’s launch of the R9 290X at $550 which beat the $1,000 Titan.  Nvidia wasted no time in releasing the 780 Ti as their flagship card at $699 which soundly beat the Titan and also beat the 290X.  However, in the 2-1/2 years since then, Nvidia has refreshed their architecture completely with Maxwell, replacing the Kepler GTX 780 Ti with the faster and cheaper GTX 980, followed by the Titan X, and then by their even faster GTX 980 Ti gaming flagship.  However, although AMD introduced new Fiji architecture with the Fury lineup, the Hawaii 290X was not refreshed, but rather rebranded into a Grenada 390X by increasing the reference vRAM capacity from 4GB to faster 8GB GDDR5 with tighter timings.

 

Although the Maxwell GTX 980 still trades blows with the Fury X at 1920×1080 and the GTX 980 Tis is faster than either card for most games and resolutions, the Kepler GTX 780 Ti appears to have gotten slower compared with the 290X in newer games.  We want to see if the 3GB limitation of the vRAM is an issue by comparing it with the original Titan 6GB and with the 290X 4GB.  Has Nvidia neglected Kepler in favor of the newer Maxwell architecture, or has AMD simply continued optimizing for Hawaii/Grenada as their Hawaii/Grenada architecture has not changed for more than 2 years?

 

When it launched in 2013, the GTX 780 Ti was solidly faster than the original Titan, and it also beat the 290X.  However, that was 2-1/2 years ago, and PC games have become far more demanding. The newest generation of consoles and console ports to PC have placed new demands on lower-capacity vRAM-equipped video cards.

 

-- TL;DR --

Quote

This was an interesting evaluation for us in revisiting a 2013 evaluation.  We have seen that the 290X is still being optimized by AMD as its architecture is unchanged with the current 390X. Partly by being 2-1/2 year old architecture, and partly by having more vRAM, the 290X has turned out to be generally more “future proof” for playing today’s latest modern PC games.  However, we can clearly see that the Kepler GTX 780 Ti still is in the same class as the 290X and it runs 2013 games just as fast or faster in 2016, although it has fallen further behind the Maxwell 980 series with the very latest games.  It also appears that 3GB of vRAM is sufficient for its target resolutions of 1920×1080 and 2560×1600 or 2560×1440.  None of today’s single GPUs are well-suited for 4K resolutions with maximum details, and we await Nvidia’s new Pascal architecture and AMD’s Polaris.

 

Source: http://www.babeltechreviews.com/nvidia-forgotten-kepler-gtx-780-ti-vs-290x-revisited/view-all/

[MEC-777]

Good ol' Hawaii. Still king of performance per dollar.

[Kinda Bottlenecked]

Has Nvidia forgotten Kepler? Nah, I think its just that the amount potential in the 290/x took some people by surprise. I wouldn't even expect a 290x to beat a 780 Ti back in the day when the 780 Ti costed much more than the 290x. 

[Morgan MLGman]

Just now, Pohernori said:

Has Nvidia forgotten Kepler? Nah, I think its just that the amount potential in the 290/x took some people by surprise. I wouldn't even expect a 290x to beat a 780 Ti back in the day when the 780 Ti costed much more than the 290x. 

Well, but you have to admit that it's weird how is it possible that in some games, the 780Ti is actually slower on current drivers than it was on the drivers when it first released? For instance, in the DIRT Showdown game, 780Ti lost 2-3 FPS on current drivers compared to the old ones, while 290X gained 10-15 FPS

[Shahnewaz]

I still believe its due to the architectural differences. There is still so much beef in GCN cards to be exploited yet.

Hopefully DX12 will fully open up these beasts.

[Mr.Meerkat]

780ti back in the dayz wrecked the 290x for a bit more monies, now the 290x is trading blows with the 780ti and winning a few as well. The monies I spent in 2014 for my dual 290x (same price as a reference 780ti) was a better choice it seems...(£250 each for my 290x while it was £500~ for a 780ti...okay that's a lie...It was £258.25 for a 290x so not exactly half of one...)

[don_svetlio]

290X being better, nothing new here

[Kinda Bottlenecked]

8 minutes ago, Morgan MLGman said:

Well, but you have to admit that it's weird how is it possible that in some games, the 780Ti is actually slower on current drivers than it was on the drivers when it first released? For instance, in the DIRT Showdown game, 780Ti lost 2-3 FPS on current drivers compared to the old ones, while 290X gained 10-15 FPS

 

It happens to the 290x as well. Sometime there are some driver benchmarks where a there would be a 1-2 fps lost in some games. If the 780 Ti lost more than 5fps in many games however, then I'd be worried. 

[Citadelen]

If AMD can improve the overhead in their drivers the 290X will skyrocket in performance. I'm hoping for something like that at the end of the year, when they release a major overhaul of Radeon Settings.

[SkepticKrab]

Buy nVidia if you want performance today (Dragster),

buy AMD if you want performance that lasts (Le Mans 24)

[xAcid9]

I want to see the battle between first 28nm flagship, the 7970 vs 680 with their release driver vs current one.  D:

[don_svetlio]

7 minutes ago, xAcid9 said:

I want to see the battle between first 28nm flagship, the 7970 vs 680 with their release driver vs current one.  D:

7970 GHz*

[xAcid9]

5 minutes ago, don_svetlio said:

7970 GHz*

I said first.

[don_svetlio]

5 minutes ago, xAcid9 said:

I said first.

That one was quite a bit lower clocked tho

[shdowhunt60]

Planned obsolescence, thy name is Kepler. Nevermind what Fermi benchmarks must look like, the 7970 still holds up to this day.

 

I'd expect something similar to happen with Fiji though. Fiji has been beating out the 980 ti lately, but I don't see that lasting for long with it's 4gb.

[Mew]

How do I do quotes on mobile? 

PowerColor R9 290X PCS+ 4GB; original reference version factory clocked at 1030Mhz Uber clocks, and with the fan allowed to spin up to 100% to prevent throttling

 

They're testing an overclocked card that sounds like a leaf blower resulting in a biased result. 

 

GALAX GTX 970 EXOC 4GB, GALAX factory clocks (1164MHz/7010MHz with 1430MHz boost on this sample), supplied by GALAX

 

Why are they testing an overclocked 970?

 

There are so many things wrong with their testing methodology: they wanted to show the difference between launch drivers in 2013 but neglected to test the titles on the same versions. They should be tested onaunxh versions of the game if they were lazy or current versions and old vs new drivers. They should be testing with a faster cpu to minimize bottlenecking, heck they even said that the difference in cpu used would impact the scores. 

.

Testing Differences between 2013 and now

Although we tested the same games at the same settings, our testing platform has changed fromWindows 7 to Windows 10 which should make for only potentially slight differences in performance which can fall under “benching noise”, or margin of error; and we are using a slightly faster CPU now as we have upgraded from Ivy Bridge’s i7-3770K at 4.5GHz to the Devil’s Canyon i7-4790K at 4.4GHz which will only slightly affect CPU-dependent benchmarks. And of course, game patches and updates will also make for larger potential performances differences which reflect playing the game then versus now.  However, the relative performance of our 3 featured video cards will be similar, and our results should be very close.

 

Similar is not good enough. It's like testing the AMD system with 16gb of ram and the other with 8gb of ram. The results shouldn't be affected by the lack of ram so "the results should he very close". This is not good enough, especially in a test where every little change is important.  

 

I don't like seeing flawed testing but I guess I should be getting used to this because some of the more well known publications also have flawed testing. 

[Prysin]

1 hour ago, Morgan MLGman said:

Well, but you have to admit that it's weird how is it possible that in some games, the 780Ti is actually slower on current drivers than it was on the drivers when it first released? For instance, in the DIRT Showdown game, 780Ti lost 2-3 FPS on current drivers compared to the old ones, while 290X gained 10-15 FPS

The GTX 780TI has most likely NOT gotten any changes. What has changed is the games themselves. I dont think reviewers are that careful sticking with the same old version of their games. Game updates changes things, adds things, removes things. All of this can have a small impact. This is why the GTX 780Ti is mostly the same with a few boosts here and there. Rather then a clear boost all over.

 

Nvidia is blatantly NOT optimizing for it. They are simply leaving it in a limbo where it performs well enough to not make the masses sharpen their pitchforks.

[Shahnewaz]

5 hours ago, xAcid9 said:

I want to see the battle between first 28nm flagship, the 7970 vs 680 with their release driver vs current one.  D:

@Slick@LinusTech PLEASE please please please please???

[MMKing]

Is anyone really surprised? Take a look at how Gameworks butchered fallout 4 performance on the 600/700 cards, while only causing mild reductions on the 900 series.

[ShadowCaptain]

my 780ti still kicking ass at 1440p

just been playing the latest tomb radier on high at 1440p 60 and it manages to just run everything fine

[Mushious]

Sitting here with SLI 780Ti, waiting for nvidia to gimp them to hell with a "buy a new videocard!" driver update.

[Lynx]

Forgotten, and AMD are more optimized?

 

IMO, NOPE. It's just that AMD Architecture are better than nvidia for newer games (hawaii vs kepler)

[don_svetlio]

14 hours ago, Prysin said:

The GTX 780TI has most likely NOT gotten any changes. What has changed is the games themselves. I dont think reviewers are that careful sticking with the same old version of their games. Game updates changes things, adds things, removes things. All of this can have a small impact. This is why the GTX 780Ti is mostly the same with a few boosts here and there. Rather then a clear boost all over.

 

Nvidia is blatantly NOT optimizing for it. They are simply leaving it in a limbo where it performs well enough to not make the masses sharpen their pitchforks.

Burn the heretics?

[Mushious]

Just now, don_svetlio said:

Burn the heretics?

Put the Fermi away and calm down a bit. The cards are still good, just not as good as they should be.

[don_svetlio]

Just now, Mushious said:

Put the Fermi away and calm down a bit. The cards are still good, just not as good as they should be.

But I'm hungry :?