[Hardwarecanucks] GTX 780 Ti vs R9 290X; The Rematch

[Good_fella]

http://www.hardwarecanucks.com/foru...iews/70125-gtx-780-ti-vs-r9-290x-rematch.html

It's forth to copy whole introduction and conclusion pages but just go to site and read it. Some rumors were completely destroyed.

 

[Silverforce11]

Depends on the game suite. Check out TPU's numbers.

780Ti on release. 7-8% faster than R290X.

Latest GPU review, latest drivers. R290X caught up, in the lead slightly.

Compared with 970 & 980 on their release, in relation to 780Ti:

 

[5150Joker]

http://www.hardwarecanucks.com/foru...iews/70125-gtx-780-ti-vs-r9-290x-rematch.html

It's forth to copy whole introduction and conclusion pages but just go to site and read it. Some rumors were completely destroyed.

I already knew those claims were b.s. long ago and made a long post here with review graphs + links disproving the "kepler is neglected" claim. It's just certain parties have an agenda to spread misinformation.

 

[Silverforce11]

Sweclockers Performance Index.

780Ti launch

Now

According to some, this must be misinformation!

 

[3DVagabond]

HWC always gets stuck with the slowest AMD hardware, for some reason. Bad luck, I guess.

 

[raghu78]

its a well known fact that SKYMTL is Nvidia biased. I have not seen him write a single negative article about Nvidia in the recent past. Even for the GTX 970 he was more like defending that the performance was the same irrespective of the fact that Nvidia lied about specs. On the contrary he loves to pick on each and every AMD mistake and in fact loves to make sarcastic comments on AMD. He is the text book definition of somebody who roots for the market leader.

Here is computerbase testing at Maxwell launch and now

http://www.computerbase.de/2014-09/geforce-gtx-980-970-test-sli-nvidia/6/

GTX 980 is 7% faster than GTX 780 Ti at 2560x1440 4xAA/16AF . 780 Ti is 8% faster than throttling ref R9 290X and 6% faster than GTX 970.

http://www.computerbase.de/2015-08/...h/2/#diagramm-rating-2560-1440-hohe-qualitaet

GTX 980 is a whopping 20% faster than GTX 780 Ti. GTX 970 is actually 3% faster than 780 Ti. 780 Ti is 4% faster than throttling ref R9 290X . R9 290X OC at 1030 Mhz is 10% faster than 780 Ti.

We know from the reviews in the past that 780 Ti was around 7-9% faster than custom R9 290X(1 Ghz). custom 780 Ti OC widened that gap against R9 custom 290X OC because the 780Ti had much higher OC headroom and Nvidia partners used that well.

http://www.computerbase.de/2013-11/geforce-gtx-780-ti-gigabyte-inno3d-test/2/

Custom 780 Ti cards were 10-15% faster than ref 780 Ti and. R9 290X OC cards had slight overclocks of 30-50 Mhz and could not match even ref 780 Ti.

https://www.techpowerup.com/reviews/MSI/GTX_780_Ti_Gaming/24.html
https://www.techpowerup.com/reviews/Sapphire/R9_290X_Tri-X_OC/24.html

So whatever SKYMTL is trying to tell is just plain rubbish. He is just flat out trying to paint Nvidia in a good light. But everyone here knows the truth. Nvidia Kepler cards started losing in more games to their GCN counterparts as 2014 and 2015 games launched and thats why today on average across a wide range of games we see a custom R9 290X OC on par with the best custom 780 Ti OC cards.

 

[Silverforce11]

@raghu78
Better becareful with all those sources & data, some people here thinks that is misinformation!

 

[raghu78]

@raghu78
Better becareful with all those sources & data, some people here thinks that is misinformation!

Only a completely biased person can deny that Kepler has lost ground to GCN and Maxwell in the last 12 months. I don't care if that person is a reviewer himself

 

[Zodiark1593]

Only a completely biased person can deny that Kepler has lost ground to GCN and Maxwell in the last 12 months. I don't care if that person is a reviewer himself

Unless you ran said hardware yourself and found the results to be accurate, then when you post your findings, are subsequently called out as biased anyway. But I digress..

I've seen that in terms of compute, Kepler seems to be at a disadvantage compared to Maxwell, GCN, and even Fermi. There are probably some deficiencies in how the shader cores are utilized that was fixed in hardware via Maxwell.

 

[Spjut]

Optimally the tests should include some older drivers as well, to show if the performance increased or not. Far Cry 4 launch drivers vs today, Dying Light launch drivers vs today etc.

Hopefully the controversy about Kepler being neglected will ensure Nvidia isn't neglecting it

I personally do believe that the work on Kepler's DX12 drivers does/did divert resources from optimizations for its DX11 drivers and can explain why it seems it was neglected

 

[Face2Face]

I can confirm, Kepler is not dead...yet.

The recent driver updates have made games like GTA V and TW3 (which kepler initially had a tough time with) perform much better and run smoother. The frame-rate increase in TW3 is minimal, but the smoothness is A night and day difference. With GTA V, it's pretty amazing how much performance was added per the newest round of drivers. The game plays amazing well @ 1440p, though my CPU seems to be the real bottleneck now. I wish I had the time and a R9 290X, so I could do some comparisons with multiple games and driver versions.

 

[tviceman]

AMD has done a great job of further optimizing their already mature GCN drivers. Kepler was discontinued a year ago and thus driver optimization focus has shifted to their current products. It does seem terms like "tanking" are a massive exaggeration, though. It's unfortunate, but it's also life in the computer-hardware-cycle world.

 

[Hi-Fi Man]

It's funny that people are posting graphs showing how 290x has now caught up as proof that Kepler is being neglected. Did you ever think that maybe AMD had more driver overhead at launch and that this driver overhead was reduced? What you guys are essentially saying is that since GCN has improved it's results (but still not ahead like some claim) that Kepler must be neglected...

You guys are grasping at straws and it's pathetic. What do you get out of spreading more FUD about Kepler? Some personal satisfaction? Maybe if you actually owned one of these cards you would realize that these claims are bogus.

 

[Silverforce11]

@Hi-Fi Man
It's not just in comparison to GCN, but in relation to the 970 & 980, on release, the 780Ti was within 5% to a 980, much faster than 970, but its quite common now to see 970 = 780Ti or faster and the 980 really blows past it.

I had a look at hardware.fr reviews as well since they do a summary performance chart, the trend holds, with 980 leading 780Ti these days by 20%, which is ridiculous since they were basically within the same perf level a short while ago.

It could simply be a case of older hardware, so newer stuff get the focus of optimizations. But to say it doesn't exist, is to ignore evidence from many review sites in recent times.

And no, I am not spreading fud or misinformation, the data is available on these review sites for anyone who gives a damn about FACTs to go and look for themselves.

Edit: Had to double check that HWC article again because the level of BS is strong, so strong.

Compare FC4 numbers, R290X vs 780Ti:

SoM (A Reference R290X faster than 980 says a lot, when HWC finds trash results):

Also they didn't even try to fudge results well because an R290X being slower than a 780Ti IN HITMAN? What a joke.

Everyone and their dog knows AMD has an advantage in Hitman and the R290X trumps the 780Ti in that game... HWC, pure NV PR.

 

[Pariah]

@Hi-Fi Man
It's not just in comparison to GCN, but in relation to the 970 & 980, on release, the 780Ti was within 5% to a 980, much faster than 970, but its quite common now to see 970 = 780Ti or faster and the 980 really blows past it.

I had a look at hardware.fr reviews as well since they do a summary performance chart, the trend holds, with 980 leading 780Ti these days by 20%, which is ridiculous since they were basically within the same perf level a short while ago.

It could simply be a case of older hardware, so newer stuff get the focus of optimizations. But to say it doesn't exist, is to ignore evidence from many review sites in recent times.

And no, I am not spreading fud or misinformation, the data is available on these review sites for anyone who gives a damn about FACTs to go and look for themselves.

980 was about 10% faster than a 780Ti at release. It was a new architecture at that point. Obviously it should be faster now. Driver improvements are ALWAYS better in year one than year 2. That is common sense, not some conspiracy. It's asinine for people to think that the gap between old and new architectures won't increase after the initial launch of the new tech.

 

[crisium]

Although their results aren't as bad as other sites, they still show that Kepler has identical performance loss to both Maxwell and GCN - as far as single digit numbers without decimals allow.

I don't think it's anything shocking what they posted (besides oddities like Nvidia performing amazingly in SoM and Hitman). Most people know that Kepler performance is game-to-game dependent. Depending on the test suite, results will vary. We will simply add this to the many other tests, as seen in this thread. And regardless of his Baghdad Bob assurances that nothing is wrong, even he shows that Kepler is indeed losing ground.

 

[Hi-Fi Man]

@Pariah You beat me to the punch, this is exactly my point. It's funny because people were complaining about how the 980 was underwhelming because it was only 10% faster at launch than the 780 Ti and now they are complaining even more because nVIDIA did what is expected and optimize their drivers for Maxwell aswell. It's insane.

 

[Silverforce11]

980 was about 10% faster than a 780Ti at release. It was a new architecture at that point. Obviously it should be faster now. Driver improvements are ALWAYS better in year one than year 2. That is common sense, not some conspiracy. It's asinine for people to think that the gap between old and new architectures won't increase after the initial launch of the new tech.

Exactly, I dont really care that Kepler drops in performance, I don't own it anymore (used to have a 670). But the issue here in fact isn't about that at all anymore. It's about HWC being completely NV PR and faking numbers. Look at their results and you go and compare it to any other reputable review sites. Completely different.

 

[Hi-Fi Man]

@Silverforce wow so this is what you just posted:
1. 980 performance with old drivers that show it tied with a 290x in one game, what are you even trying to prove here? I can use the same argument for using canucks' hitman numbers.
EDIT: Actually most of your cherry picked graphs show 980 tying 290x. What does that tell us about your agenda? Because it's well known 980 is indeed faster.

2. Kepler performance hasn't dropped, at all. GCN has gotten better since launch, like ALL cards should...

3. Just because you don't agree with canucks' numbers they are automatically NV PR? What kind of flawed logic is that?

 

[Pariah]

Exactly, I dont really care that Kepler drops in performance, I don't own it anymore (used to have a 670). But the issue here in fact isn't about that at all anymore. It's about HWC being completely NV PR and faking numbers. Look at their results and you go and compare it to any other reputable review sites. Completely different.

Kepler is NOT dropping in performance. It's a replaced architecture so it's driver improvements are not keeping pace with its replacement. Again, this is the way it always works. All this article is showing is that the difference between Maxwell and Kepler are within the expected range and there is no evidence of foul play on NVidia's part to intentionally handicap Kepler. Anything more than that and you are missing the point and trying to create controversy where there isn't any.

 

[Silverforce11]

Again, let the data speak for itself, as to why the HWC results are trash.

SoM (HWC Thinks 780Ti is faster!)

Hitman (HWC thinks 780Ti is faster!)

ps. I am not questioning whether NV prematurely obsolete Kepler, don't care. Point is HWC's results do not match that of every other reputable site I looked at, you can put TPU, Guru3d on there too, along with Hardware.fr, Computerbase.de etc, the ones I linked are all recent reviews so there's no cherry picking, cos you can find those cherries across multiple sites saying basically the same thing.

HWC is basically saying 780Ti is matching or faster than R290X in ACU, FC4, Hitman, SoM etc when other sites have it the reversed! So either HWC is making up fud, or every other site has it wrong.

 

[DustinBrowder]

What a fake test. They don't even post the machine they are using or OS.

For all we know they might be testing on windows XP and some weird combination of hardware, plus if you read my thread about reviews, you'd know ALL reviews are inherently flawed! Anyone can come up with any numbers and technically it could be true, since they can test only a very specific part on a 30 second running in a straight line!

 

[Hi-Fi Man]

You just reposted the same cherry picked graphs and in your graphs the 780 ti, 980 and 290x are essentially tied with the results being within the margin of error. A 980 is obviously faster than both of those cards which means your graphs are worthless for comparison.

 

[Hi-Fi Man]

What a fake test. They don't even post the machine they are using or OS.

For all we know they might be testing on windows XP and some weird combination of hardware, plus if you read my thread about reviews, you'd know ALL reviews are inherently flawed! Anyone can come up with any numbers and technically it could be true, since they can test only a very specific part on a 30 second running in a straight line!

You know Maxwell doesn't even have XP drivers right?

 

[Termie]

Having published my own benches of my 980, 290, and 780 ti, cards I bought myself and wanted to perform well, I can tell you that while skymtl's benches are not wrong, his conclusion is.

Kepler didn't suddenly get worse in old games, which make up the bulk of his test suite. Kepler underperforms in new games, like witcher 3, far cry 4, and a few others. In averaging the entire suite, skymtl papers over the losses evident in his own benchmarks in new games.

My benchmarks can be found in the article discussed in this thread:

http://forums.anandtech.com/showthread.php?goto=newpost&t=2440066