Nice review HH, like always. I was just expecting you do HW/off benchmark to compare. Its really taxing the AMD hardware. One of the articles said 290 is 77% faster with HW/off.

I'll have a look tomorrow, Hairworks is a NV feature and was disabled for fair comparison. This article is work in progress and I wanted the initials number out. Basically I was waiting (hoping) on AMD for a driver ... but well.

Nice review HH, like always. I was just expecting you do HW/off benchmark to compare. Its really taxing the AMD hardware. One of the articles said 290 is 77% faster with HW/off.

HW is off for the test.

Everything is set to Ultra, AA is enabled, Nvidia Hairworks is disabled and we use SSAO. The graphics cards used in this specific article are:

Oh, are you just asking for a comparison? 980 is 40% faster than a 780Ti in the QHD test compared to the 10-15% when the 980 launched. Nvidia posted on their forums that they are looking into it. I wonder if a 6GB 780Ti would perform better in the test? Also seems weird that the 780Ti only performs ~8% better than a 780. I wonder if it's a driver bug or something with the 780Ti

Aaaand yet another benchmark that shows what all other benchmarks has shown - kepler having been seriously downgraded (intentional or accidental - im leaning towards intentional). Some people claim its simply because Maxwell has been optimized, but if that was the case, then a gtx 980 should be MILES ahead of a 290x, and a 780 ti should still be on par with the 290x, but it isnt. The gtx 980 is a tad ahead of the 290x as usual, while the 780 ti is waaaaay behind the 290x - the same can be seen with all the other kepler Cards, with the 780 almost performing the same as a freaking 960...

Idk, honestly I'd like to see a 980's 3DMark scores over the last several driver releases. I mean basically it seems like Maxwell on average is performing 20% faster than it was at launch. At launch a 980 was on average about 10-15% faster then a 780Ti. Now it seems like it's about 30-40% faster then a 780Ti in most recently released games. A 960 was also about 20% slower then a 780. Now it seems like a 960 is about 2-5% slower. The only real way to see if Nvidia is "downgrading" performance is to either A. Show me a driver where a Kepler series card performs 20% slower. Or B. Show me the 3DMark/Any Benchmark between release drivers of Maxwell and current drivers of Maxwell. I think more than likely Maxwell is just getting the better end of the deal with drivers. Driver improvements slowed with Kepler by the time it was finished anyway. The 680 showed a 25% increase in performance from drivers over it' lifetime. 780 only showed a ~8-10% increase during it's lifetime. It's pretty clear that Kepler performance was maxed. Regardless, ManuelG posted on the Nvidia forums that the driver teams are looking into it. So if there is an issue I'm sure it will be corrected.

You seem to miss my point though. I get that Maxwell will recieve more achitectural improvements, that is to be expected, but that would mean that Maxwell should increase in performance in comparison to amd, and that kepler should stay where it is, so to speak. But Maxwell performance in comparison to amd is the same as at launch, while kepler performance has taken a huge dive in comparison to amd. Seems "odd", no?

Idk, you would also need to examine AMD's performance increases as well. You have to remember that AMD is actually better suited at driver increases than Nvidia due to the nature of their architecture. Don't get me wrong, I definitely think that something could be going on. Could be a driver issue/bug, could be Nvidia is intentionally crippling performance, I'm not sure. It just bothers me that people are throwing out all notions of actual testing and just bandwagoning based on circumstantial evidence at best. No one is actually doing any tests or looking at all possible reasons as to why this could be happening. I guess when I get home I'll do some A/B tests between launch drivers and current drivers on various games to see the % impact in performance on the 980. Someone would probably also do the same with a 290x, between the AMD drivers that were used on the 980 launch day, compared to current. Edit: So launch 980 drivers were 14.3 for the 290x. Most of the comparisons used that. The Catalyst Omega Driver (14.12? I think) came out in Dec, after the launch of the 980 and increased 290x performance by up to 19% in some games. Again, this isn't anything definitive but it definitely goes to show you that just comparing things to the past isn't as simple as you would think. Both sides are making significant strides in performance via drivers.

Added: GTX 770 image quality settings scaling on last page.

Kepler beating Maxwell on Nvidia, R9 285 getting better FPS over the 7970ghz edition/R9 280x on AMD - what the heck is going on?.

Look again, Kepler isnt even close to Maxwell. When we talk AMD, GCN 1.2 cards are getting better driver support/optimizations. Tahiti cards still good but getting old...

I mean basically it seems like Maxwell on average is performing 20% faster than it was at launch.

Drivers getting better? I gained 9 FPS on Metro LL and 28 on Thief, when i compare 344.11 and 352.86.

Kepler not doing as well as it should be on Nvidia, an R9 285 getting better FPS over the 7970ghz edition/R9 280x on AMD - what the heck is going on?.

Absolute joke. Those R295x2 results are disgraceful 😛uke2:

I did fall over the same thing, that i wasnt able to alter tesselation settings. Nvidia claims in their witcher 3 guide, that Maxwell is 3 times faster at doing tesselation than kepler... so if all benchmarks are done at ultra (Thus 64x tesselation), that might contribute (if not all together explain) to why kepler is struggeling.

But Kepler itself is several times faster at tessellation than the 290x. Which was the whole argument behind the Crysis tesselation-gate thing. So if tessellation is really a factor here, why is the 290x also not struggling? Did AMD unlock tessellation performance in a recent driver? Is it more optimized towards their method of tessellation? I'm not being facetious, I'm just trying to understand why there is such a performance difference with Kepler here.

Tessellation is improved alot on AMD. They are in pair since HD 5000/6000 series. I dont think thats the issue here.

Witcher 3 is one demanding game! Thanks for the review, and I'm looking forward to seeing your upcoming review on how mainstream cards perform at different quality settings - this is really what I'm interested in, gives me an idea how my card would perform!

Tessellation is improved alot on AMD. They are in pair since HD 5000/6000 series. I dont think thats the issue here.

Nope Kepler has still faster tessellation. Although I agree it improved a lot since GCN series 7000 and now 290.. http://images.anandtech.com/graphs/graph8526/67750.png http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/19 http://www.computerbase.de/2013-10/amd-radeon-r9-290x-test/13/#diagramm-stone-giant-2560-1600 Compute ain't either 980 vs 780ti, especially once you start to OC Kepler quick percentage by gaming and compute http://www.anandtech.com/show/8549/short-bytes-nvidia-geforce-gtx-980-in-1000-words

Perhaps more importantly, the GTX 980 is also on average 8% faster than the GTX 780 Ti and 13.5% faster than AMD's Radeon R9 290X (in Uber mode, as that's what most shipping cards use). Compute performance sees some even larger gains over previous NVIDIA GPUs, with the 980 besting the 680 by 132%; it's also 16% faster than the 780 Ti but "only" 1.5% faster than the 290X – though the 290X still beats the GTX 980 in Sony Vegas Pro 12 and SystemCompute.

To me it looks like they made the game in favor of Maxwell and crippled on the rest, its bundled with 960+ afterall lol, fail marketing.

From WCCFTech: However a quick look at the R9 285’s tessellation performance reveals that the situation isn’t exactly as Burke made out it to be. Based on tessellation performance alone, an R9 285 should see a comparable performance hit to the GTX 960 when enabling HairWorks. However that’s not the case, not even remotely. The R9 285 actually sees a performance penalty that’s twice as bad as that seen on the GTX 960. And because CD Projekt Red and AMD are not allowed to work towards identifying and addressing the real culprit behind this poor performance in the game code we’re back to square one.

But how does that explain a 780Ti performing worse than a 290x? I mean, I think we can all agree that gameworks is just ****ty, no one is really arguing that. The performance in the test here though, just doesn't make sense in regards to Kepler and Hairworks is off for this test.

Now lets see how things go with other upcoming titles. Question probably still remains in many peoples minds, Nvidia 'neglecting' kepler or has that arch seen its limits with newer games? The problem is that this is occurring on a gameworks title. If "neglect" (which I doubt), I will probably not buy an Nvidia card again.

Now lets see how things go with other upcoming titles. Question probably still remains in many peoples minds, Nvidia 'neglecting' kepler or has that arch seen its limits with newer games? The problem is that this is occurring on a gameworks title. If "neglect" (which I doubt), I will probably not buy an Nvidia card again.

Indeed. It would interesting to see how it performs in next couple of games. Even so, this started with FC4 and up till now, TW3 is only the worst one yet. Idk, seem so it can only go downhill from this...

IMO, it's suspicious that Kepler started to fall behind in new games just after Maxwell was released.

I'd like to know what makes Maxwell perform so much better than Kepler in this. I don't believe in conspiracy theories.

The performance hit from hairworks is rather gnarly :S.

Well since poor AMD hasn't even had a chance to release a performance driver for the W3 yet and not even CF support then i can't really take these benches seriously. What we can see though is 100% that nVidia has left Kepler to the wolves. They seem to only be improving Maxwell performance. If i was a 780ti/sli owner right now then i would be very angry with nVidia because it's not too long ago that these cards were still on sale at high prices like nearly 400 squid. Man that's hard to take and unless they fix it i'm sure some old skool NV fan bois will make the switch from Green to Red.