[HARDOCP] GTA V - The death of Kepler and GCN 1.0

[escrow4]

http://www.hardocp.com/article/2015/04/27/grand_theft_auto_v_single_gpu_performance_review_part_1/7

"The AMD Radeon R9 280X was the worst offender, not only having problems rendering higher levels of grass, but just in overall performance. On paper, the specifications of the R9 280X are better than the R9 285, especially in memory bandwidth and capacity. However, these things did not help the AMD Radeon R9 280X at all."

"The same is also true for the GeForce GTX 780, a once $649 high-end GPU, now reduced to the performance range of a $199 GeForce GTX 960 in GTA V. This was a high-end Kepler GPU, but you won't find it providing anywhere near an enjoyable 1440p gameplay experience. You will have to settle for 1080p and lower settings. The performance is a bit faster than the GTX 960, but not anywhere near what NVIDIA's current generation Maxwell architecture can provide."

"It seems the newer the generation, the better the gameplay experience in GTA V on both AMD and NVIDIA GPUs. This is the first time past generation architectures have suffered so greatly in a new game that we can remember. This is the first game that may persuade a lot of gamers to upgrade to the latest architectures."

"Again, not shown in this evaluation, but we did dabble with the GTX 750 Ti. It is a joke for GTA V, just like the AMD 260X. We'd stay away from those low-end cards for this game."

Stop tempting me Titan X . . . . . :whiste:

 

[hawtdawg]

Their own benchmarks have the 780 dead even with the 970 in 1080p, and they don't even have 780 1440p numbers.

We've seen Kepler slipping for a while, I attribute this more to Nvidia not improving drivers for them so they can make their newer cards look more impressive. It's not as though Kepler is fundamentally different from Maxwell.

 

[Magic Carpet]

Not bad, but instead of explaining what exactly wrong with the older GPUs, they went to label them, just as "old". Not really professional.

 

[Silverforce11]

It's interesting the R285 is so far ahead of R280X, other sites did not find this discrepancy at all. Certainly the 960 is not playing on highest settings. :/

 

[Magic Carpet]

It's interesting the R285 is so far ahead of R280X, other sites did not find this discrepancy at all.

They cherry picked settings that suited Tonga better than Tahiti. At standard settings Tahiti is significantly faster as GameGPU showed. Don't know about you, but I prefer higher minimums over eye-candy. Both of these GPUs are too slow anyway for Ultra settings. They rushed with the review... for some reason. Should have included Titan X in the first place. Cheap.

 

[Silverforce11]

They cherry picked settings that suited Tonga better than Hawaii. At standard settings Tahiti is significantly faster as GameGPU showed.

Yes but if its real, regardless of cherry picking settings or location of testing (they did custom bench, not built in bench).. it is very interesting that R280X is THAT much slower than crippled Tonga.

It's very reminiscent of Godray Tessellation in FC4. I suspect the scenes they tested, mostly outdoors, wilderness driving where all cards suffer most fps drops, tessellation is used heavily for scenery or grass on highest details. That would explain GCN 1.0 tanking hard as its tessellation is a lot weaker than in Tonga.

 

[raghu78]

They cherry picked settings that suited Tonga better than Hawaii. At standard settings Tahiti is significantly faster as GameGPU showed. Don't know about you, but I prefer higher minimums over eye-candy. Both of these GPUs are too slow anyway on Ultra settings.

pcgameshardware ran actual gameplay with fraps and the R9 285 was on par with R9 280X

http://www.pcgameshardware.de/GTA-5...219948/Tests/GTA-5-PC-Test-Benchmark-1156215/

 

[Magic Carpet]

Silverforce11,

Tessellation kills the performance of the 280x, but that is to be expected. If you lower it, 280x performs much better. The point is, both cards are too slow for >=60 FPS, so no point trying to max it out. More use, if they had included the exact gfx settings for each card to maintain fluid frame rate, so people could save time, instead wasting time trying to find out XYZ settings for comfortable FPS levels.

 

[Skurge]

This 680 runs GTAV fine at 1080p at a high settings with no AA. Maybe it will get slower as I play further into the game, but so far, its pegged at 60FPS.

 

[Magic Carpet]

This 680 runs GTAV fine at 1080p at a high settings with no AA. Maybe it will get slower as I play further into the game, but so far, its pegged at 60FPS.

According to GameGPU, your card performs quite well. And what's more important, your own experience is in line with those numbers. So, we can build some trust. I am going to get the game soon and see how it does on my ancient 670.

 

[AtenRa]

This is with normal grass and no advance options, Tonga demolishes Tahiti here.
At those settings all cards are able to get 60fps avg, exept 280x.
Too bad they didnt include 280x/285 and 960 at the same settings in 1440p.

 

[Silverforce11]

This is with normal grass and no advance options, Tonga demolishes Tahiti here.
At those settings all cards are able to get 60fps avg, exept 280x.
Too bad they didnt include 280x/285 and 960 at the same settings in 1440p.

It's their test scene, outdoor wilderness driving, not in cities.

Something there is crippling R280X performance which doesn't occur elsewhere (as other sites show).

Interesting if its a driver issue, or just lots of tessellation is used in those scenes.

 

[XiandreX]

As nice as it would be to see 1440p tests, these cards are not intended for 1440p and therefore holds little bearing for most users.

 

[Deders]

I'm running GTAV perfectly well at max settings but with grass turned down a notch from ultra and 2xAA @2560x1080 on my 780. I'm even using more than 3600MB's worth of vram apparently (I disable themes so aero doesn't take up precious memory space when playing this game).

 

[XiandreX]

I'm running GTAV perfectly well at max settings but with grass turned down a notch from ultra and 2xAA @2560x1080 on my 780. I'm even using more than 3600MB's worth of vram apparently (I disable themes so aero doesn't take up precious memory space when playing this game).

Perfectly well is subjective. What one person regards as playable, another regards as unplayable.
Roaming the city is no problem, however if you are spending a lot of time in the hills, frame rates tend to be iffy.

 

[SPBHM]

"Again, not shown in this evaluation, but we did dabble with the GTX 750 Ti. It is a joke for GTA V, just like the AMD 260X. We'd stay away from those low-end cards for this game."

really? going by Eurogamer these cards can handle this game pretty well without losing to the target platforms the game was made for
https://www.youtube.com/watch?v=9OE2iI7OLh8

even the 1GB 260X is not doing to badly
"Meanwhile, even a 1GB card like the R7 260X holds strong, only falling a little shy of 30fps during the same shoot-out sequence. Switching off the ultra presets smooths out performance for a capped experience, despite flying in the face of the game's suggested RAM limits."

http://www.eurogamer.net/articles/digitalfoundry-2015-grand-theft-auto-5-pc-face-off

 

[Face2Face]

"The same is also true for the GeF... https://www.youtube.com/watch?v=FmjbC7LehdU

 

[Silverforce11]

Wow...just wow.

GTA V is completely unplayable at 1440p. I mean, I have to use low settings just to ekk out 30 FPS.

https://www.youtube.com/watch?v=FmjbC7LehdU

You have to find where [H] bench their scene to get such terrible results. heh

 

[master_shake_]

oh noes the death of a 4 year old piece of hardware lol

 

[Head1985]

Nothing new here.Kepler in new games failing.

Face2face: they tested it at max settings with ultra grass and advanced graphics all max.

 

[Face2Face]

You have to find where [H] bench their scene to get such terrible results. heh

Agreed. There's not doubt it would choke and spit at max settings @ 1440p out in the county where they tested, but so does everything else from what I can see.

I was just responding to the bit where they said a GTX 780 will not provide an enjoyable gaming experience at 1440p, and you'll have to play at 1080p - Low settings. I know I run my card overclocked, but even with no overclock, the game is still very playable @ 1440p H-VH settings. Admittedly with Ultra grass, the game does have slowdowns in the country side like they stated, but turning it down to high helps out a lot, and the difference in quality is not that drastic, nor does it warrant an upgrade.

Now if it chokes on The Witcher 3... I'll revisit.

 

[5150Joker]

Nothing new here.Kepler in new games failing.

Face2face: they tested it at max settings with ultra grass and advanced graphics all max.

Ultra grass even destroys the fps on both my Titan X's in SLI. So making some stupid statement about the death of Kepler/GCN based on that kind of testing environment is beyond ridiculous. Face2Face's test has a lot more credibility IMO.

 

[Spjut]

Not bad, but instead of explaining what exactly wrong with the older GPUs, they went to label them, just as "old". Not really professional.

Yeah, the 780 was one of Nvidia's best cards before the 900 series was released back in September 2014. It hasn't even been once year since then, not exactly what I'd call old.



I hope Nvidia hasn't stopped optimizing the drivers for Kepler.
I'd guess that Nvidia's driver team could have too much work for DX11 optimizations for Fermi/Kepler right now, instead working on the DX12 drivers.

 

[Deders]

The Ultra grass setting just extends the shadow distance for grass which I'm sure you can imagine for that much grass can be a big hit.

 

[Face2Face]

Ultra grass even destroys the fps on both my Titan X's in SLI. So making some stupid statement about the death of Kepler/GCN based on that kind of testing environment is beyond ridiculous. Face2Face's test has a lot more credibility IMO.

That's nuts on such a beast PC.

It does seem that Nvidia hasn't been giving the driver love to Kepler these days, but I can't beleive that [H] would say a GTX 780 = GTX 960 in this game

The Ultra grass setting just extends the shadow distance for grass which I'm sure you can imagine for that much grass can be a big hit.

You're right, and the visual difference isn't huge IMO. Certainly not worth the mega performance hit.

I have my grass set to High - see the comparison

http://international.download.nvidi...y-interactive-comparison-4-ultra-vs-high.html