[wccftech] rumor: amd hawaii benchmarked in 3dmark11 firestrike AMD attacks titan

[boxleitnerb]

...

Maybe I should have been more clear. Within a generation from a single IHV. 3DMark predicts AMD vs AMD and NV vs NV well.

 

[Keysplayr]

Maybe I should have been more clear. Within a generation from a single IHV. 3DMark predicts AMD vs AMD and NV vs NV well.

That might be true if we lived in a world where 3DMark scores weren't sometimes falsified to generate interest and hype. Unfortunately, we don't.

 

[boxleitnerb]

That might be true if we lived in a world where 3DMark scores weren't sometimes falsified to generate interest and hype. Unfortunately, we don't.

Any (recent) proof? Afaik there is no cheating going on and hasn't for years.

 

[Keysplayr]

Any (recent) proof? Afaik there is no cheating going on and hasn't for years.

Proof? Can you prove the 3DMark score is legit or not? I can't.

 

[Lonyo]

Any (recent) proof? Afaik there is no cheating going on and hasn't for years.

I think he means faked by the people leaking the information, not AMD/NV cheating to inflate their 3D Mark scores.
There is nothing to confirm that the 3D Mark score in the OP is a legit score/from a new card, so it could be faked (as people already know).

 

[raghu78]

Proof? Can you prove the 3DMark score is legit or not? I can't.

for that you have to wait roughly another 2 months for launch reviews. :biggrin: common mr.keys we know that could be a fake screenshot. but the question is does 3D Mark firestrike / firestrike extreme score have any correlation with game performance. I would say yes. the perf diff on avg between GTX Titan and HD 7970 Ghz in games and in 3d mark 2013 is roughly the same. 20 - 22% for GTX 780 and 33 - 35% for Titan.

 

[RussianSensation]

Still, if we know the general performance of say the 7970, 770, 780 in relationship to each other and how they compare in Firestrike, we can draw some reasonable conclusions to where Hawaii will lie, I think. That's assuming the bench is legit in the first place.

That's the problem I have with Firestrike. 3D11 favoured Kepler, but now Firestrike favours GCN for no reason.

I have major issues with 925mhz 7970 beating GTX680 and 770 losing to 7970GE. In the real world 770 is actually slightly faster and 680 beats 7970.

Looking at Asus DCUII 780. In FireStrike it is 21% faster and in FS Extreme it is 25% faster than 7970GE.

In real world games, 780 DCUII is 28% faster at 1200p and 33% faster at 1600p. Latest 3DMark is low-balling Kepler performance / or inflating GCN performance. Same BS that 3D11 did for Kepler vs. GCN but now AMD gets the unfair benefit.

--

Ignoring all that, let's look at 7970GE (3458) vs. leaked 9970 score (4816). If we for a second presume that 3DMark scaling --> gaming scaling, then 9970 ends up 39% faster than 7970GE. In that case the 9970 would beat the Titan and tie most after-market 780s in stock form. Remember the reference 780 is just 17-22% faster than 7970GE, so if a reference 9970 is 39% faster, it will blow the doors off reference 780s and after-market 9970s would be even faster than 39%. Why does that sound too good to be true at $550?

Then you have a random score posted with no evidence that it even came from 9970. Why should we even believe that leak in the first place? Remember the Titan was supposed to score X7000!

 

[SirPauly]

To be completely honest I stopped looking at 3dMARK scores many years ago.

It may not be as important as many real world gaming examples but do desire to see how the competitors compare on this synthetic test that pushes the technology boundaries, imho!

Place more merit with 3dmark than Crysis 2 with tessellation or Dirt Showdown with advanced lighting.

 

[Fanatical Meat]

All I can say is finally we're seeing some good value in the $150-200 range. Hopefully this continues that trend.

 

[raghu78]

That's the problem I have with Firestrike. 3D11 favoured Kepler, but now Firestrike favours GCN for no reason.

3DMark 2013 is compute shader intensive and the fact that 3DMark felt it pertinent to stress the graphics cards compute perf illustrates the direction in which games are going.

Secondly with GCN being the foundation of PS4 / Xbox One and both of them being well endowed with compute power it stands to reason that going forward console games and so PC ports would be compute heavy.

http://www.brightsideofnews.com/new...new-3dmark-fire-strike-extreme-benchmark.aspx

"Fire Strike is made up of four different tests, two graphics tests, a physics test, and a combined test that unifies both graphics and physics into one incredibly strenuous test. The first graphical test is designed to test geometry and illumination, this is done using hundreds of spot lights that both cast and don't cast shadows. The second graphical test is designed to test particles and GPU simulations by using lots of smoke and spot lights to create detailed scenes that also heavily stress the GPU. The third test is the physics test, and it is designed to put the least amount of stress on the GPU as possible and to focus more on the CPU utilizing the Bullet Open Source Physics Engine Library. The final test, the combined test, in fact combines elements of all of the previous three tests. In addition to that, it does these things using tesselation, volumetric illumination, fluid simulation, particule simulation, FFT based bloom and depth of field."

I have major issues with 925mhz 7970 beating GTX680 and 770 losing to 7970GE. In the real world 770 is actually slightly faster and 680 beats 7970.

many reviews have shown HD 7970 Ghz on par with GTX 770. so I don't understand your disagreement.

http://www.hardwarecanucks.com/foru...369-nvidia-geforce-gtx-770-2gb-review-13.html
http://techreport.com/review/24996/nvidia-geforce-gtx-760-graphics-card-reviewed/10
http://www.tomshardware.com/reviews/geforce-gtx-770-gk104-review,3519-31.html
http://www.computerbase.de/artikel/grafikkarten/2013/nvidia-geforce-gtx-770-im-test/4/
http://www.hardware.fr/articles/896-22/recapitulatif-performances.html

Ignoring all that, let's look at 7970GE (3458) vs. leaked 9970 score (4816). If we for a second presume that 3DMark scaling --> gaming scaling, then 9970 ends up 39% faster than 7970GE. In that case the 9970 would beat the Titan and tie most after-market 780s in stock form. Remember the reference 780 is just 17-22% faster than 7970GE, so if a reference 9970 is 39% faster, it will blow the doors off reference 780s and after-market 9970s would be even faster than 39%. Why does that sound too good to be true at $550?

if it sounds too good to be true don't bother. everything will be known in 2 months time. :biggrin:

 

[SirPauly]

What I enjoy are the white papers and potentially allows a bit more of an understanding of what is going on:


White paper DirectX 11:

http://www.3dmark.com/wp-content/uploads/2010/12/3DMark11_Whitepaper.pdf

I do agree with this statement to some degree and look at 3dmark as a future gauge and why this synthetic test holds some merit or value:

The Extreme preset extends the lifetime of the benchmark by representing the likely loads used by high end games in years to come

3DMARK Technical Guide:

http://s3.amazonaws.com/futuremark-static/downloads/3DMark_Technical_Guide.pdf

 

[RussianSensation]

Place more merit with 3dmark than Crysis 2 with tessellation or Dirt Showdown with advanced lighting.

Only if you never intend to play Crysis 2 or Dirt Showdown. No one tries to use these outlier games to gauge average GPU performance. However, if you play those 2 games, you do want to see how your card performs, regardless if it favours NV or AMD. That's the thing with games - we brand agnostic PC gamers could care less if Game 1 runs faster on GPU red or if Game 2 runs faster on GPU green. What I care about is that I can play the game with good enough quality settings at decent frame rates with my card; if not, what are my upgrade choices?

There is nothing wrong with buying a GPU that runs your favourite games better. If all you do is play Dirt/GRID games or Crysis games, well you can make a more informed choice about your preferred GPU vendor for those games. On the contrary, 3DMark scores tell us absolutely nothing about how my card will play Tomb Raider, Metro LL, Crysis 3, Watch Dogs, Witcher 2&3, handle SSAA, etc.

If Futuremark ever desires to become relevant again to PC gamers, they need to license top 5 game engines and make 5 tests using each of those. Then 3DMark will represent a index based performance of the most modern/forward looking PC game engines. Then I'd take it seriously. Right now it's utter junk. Throwing X% tessellation, Y% global illumination and Z% Depth of Field does not make it representative of future games unless you know 90% of games in the next 2 years will use X/Y/Z% of those next gen effects in their engines. Unigine is flawed for the same reason -- it uses a disproportionate amount of tessellation that no game will use because it would destroy performance. That benchmark itself is only useful if you want to do stability testing or test tessellation performance of a particular GPU architecture. It also has no relation to real world games.

I do agree with this statement to some degree and look at 3dmark as a future gauge and why this synthetic test holds some merit or value:

They publish the same garbage every time a new 3DM test is released and you know if you look since 2006, every time they are wrong. Please tell me you believe GTX680 is 67% faster than GTX580 for modern games as 3DM11 shows? Another major flaw of 3DMark is by the time next generation games come out, we have next generation GPUs that we can actually test in those very games. Right now I am testing my 7970GE in 3DMark but by the time Next Gen PC games are out, I am going to be on 20nm GPUs. What do I need 3DM for, to tell me that my 7970GE will be too slow for 2015 games? What a revelation. Will Witcher 3 run on 3DM game engine? Nope. It's a worthless marketing synthetic test that needs to die or adopt licensed PC game engines to be truly applicable.

If 3DM software engineers can predict the likely loads of next gen games, why for the last 3-4 years none of their benchmarks used DirectCompute extensively as we have seen in recent games? Cuz they are full of it. The reason 3dM sells is the perpetual swinging e-Peen on PC forums. People just want to see higher scores with their new card, regardless of understanding what that score actually means. But eh my card scored XXXXX 3DM. Woo-hoo. BTW, how did you beat the final boss? I can't seem to figure it out. :hmm:

 

[Keysplayr]

for that you have to wait roughly another 2 months for launch reviews. :biggrin: common mr.keys we know that could be a fake screenshot. but the question is does 3D Mark firestrike / firestrike extreme score have any correlation with game performance. I would say yes. the perf diff on avg between GTX Titan and HD 7970 Ghz in games and in 3d mark 2013 is roughly the same. 20 - 22% for GTX 780 and 33 - 35% for Titan.

It doesn't appear that you know that at all. Sorry. :: shrugs ::

 

[raghu78]

On the contrary, 3DMark scores tell us absolutely nothing about how my card will play Tomb Raider, Metro LL, Crysis 3, Watch Dogs, Witcher 2&3, handle SSAA, etc.

A specific game might run slightly better on a certain vendor's GPU. eg: AC3 runs better on nVIDIA GPUs and Tombraider on AMD GPUs. but thats why reviews include all types of titles - both TWIMTBP and AMD GE titles. on average the GTX 780/ Titan is 20% /33% faster than HD 7970 Ghz . that perf diff agrees with 3DMark Firestrike scores of the above cards. that in itself is commendable.

If Futuremark ever desires to become relevant again to PC gamers, they need to license top 5 game engines and make 5 tests using each of those. Then 3DMark will represent an index based performance of the most modern/forward looking PC game engines.

No way. you write a benchmark which will test GPU performance by incorporating a mix of technologies. Volumetric illumination, fluid simulation, particle simulation, FFT based bloom and DOF are exactly these. They test shader performance, tesselation performance, compute shader performance, raster performance . these technologies need to be forward looking and represent current and future trends in the industry. sorry RS but I disagree with your view.

 

[raghu78]

It doesn't appear that you know that at all. Sorry. :: shrugs ::

how would I know until launch ? you are talking as if you know for a fact that its a fake screenshot. :biggrin:

 

[SirPauly]

There is nothing wrong with buying a GPU that runs your favourite games better. If all you do is play Dirt/GRID games or Crysis games, well you can make a more informed choice about your preferred GPU vendor for those games. On the contrary, 3DMark scores tell us absolutely nothing about how my card will play Tomb Raider, Metro LL, Crysis 3, Watch Dogs, Witcher 2&3, handle SSAA, etc.

It's information and data on future forward features and technology; it's a valuable tool to tweak one's system; its a valuable tool to compare one's system with others; it's very popular and could make the point, most popular!

What one desires to do with the information is up to each individual!

You believe it is irrelevant and utter junk! Too extreme!

 

[Zanovar]

It doesn't appear that you know that at all. Sorry. :: shrugs ::

Well raghu knows everything,i demand you submit to his knowledge. just to add ill piss myself if these amd next cards are 20nm,although unlikley.

 

[Keysplayr]

how would I know until launch ? you are talking as if you know for a fact that its a fake screenshot. :biggrin:

Actually, I'll stick with saying it is a void screenshot without any proof of origin and should be dismissed cleanly.

 

[zebrax2]

While 3DMark might not be entirely accurate in showing how different cards will perform nothing else really is eg. if a gpu A performs faster than gpu B in a certain does that mean that the results would be similar all the time in other games? or if gpu A performs faster than gpu B in a set of 20 games does that mean that the card would always be faster in every games out there and every game that will be made in the future? no. As boxleitnerb has said 3dmark is pretty good for predicting how the cards from the same generations specially from the same company roughly compare to each other.

 

[raghu78]

Well raghu knows everything,i demand you submit to his knowledge. just to add ill piss myself if these amd next cards are 20nm,although unlikley.

this thread started off with a so called benchmark screenshot. might be fake. who knows. Only AMD and their partners know what Hawaii is capable of. the rest are guessing and/or speculating.

 

[Zanovar]

this thread started off with a so called benchmark screenshot. might be fake. who knows. Only AMD and their partners know what Hawaii is capable of. the rest are guessing and/or speculating.

Well,the rest are speculating and guessing.are you?*coughs*guessing use it more often*winks*just a tip

 

[RussianSensation]

As boxleitnerb has said 3dmark is pretty good for predicting how the cards from the same generations specially from the same company roughly compare to each other.

That's self explanatory. It's obvious that 680 > 670 > 660Ti > 660. I don't need 3DM to tell me that. The whole point of 3DM is to try to predict performance in future games. How do you do that without using next generation game engines that will be the dominant force for PC gaming? You need to make benchmarks incorporating tessellation, global illumination, particle effects using UE4, Frostbyte 3.0, Crytek 3.4, etc. What 3DM does it take their proprietary engine and throws these effects in. That tells us absolutely nothing because no future game will use 3DM game engine. Just because they threw in tessellation and depth of field doesn't tell me anything about how GTX780 will cope with UE4 games vs. 7970GE.

3DM's primary purpose -- to try and predict performance in future games -- is undermined by the fact that the benchmark incorporates NONE of the next generation PC game engines on which games will be based. This is exactly why 3DM is a waste of time and has been for a long time.

And if we only wanted to compare how GPUs stack up, we look to reviews of real world games at ComputerBase, TPU, TechReport, AnandTech, Xbitlabs, TechSpot, HWC, etc.

For example, if 90% of EA's games will be based on FrostByte 3.0 in the next 3 years, the best predictor of performance in those games is testing existing GPUs in a game that uses FrostByte 3.0 engine, not 3DM. The same is true for UE4, Crytek, etc.

Volumetric illumination, fluid simulation, particle simulation, FFT based bloom and DOF are exactly these. They test shader performance, tesselation performance, compute shader performance, raster performance . these technologies need to be forward looking and represent current and future trends in the industry. sorry RS but I disagree with your view.

Those effects do not exist in a vacuum unless you are talking about a synthetic benchmark. You can take a game like Hitman Absolution vs. Dirt Showdown and see that global illumination affects performance completely differently. You can take a game like CivV, Batman Games, Crysis 2 vs. Unigine and see that tessellation has a completely different impact in all of them. In the end, different game engines run different effects with different efficiency. You cannot assume that FrostByte 3.0 has the same ability to generate millions of particles as UE4 does. What you want is to test millions of particles in various game engines that will underline next gen games. Testing performance in next gen game engines is the closest true predictor of future gaming performance. This is even more true since NV/AMD will optimize drivers for those game engines.

The counter-point -- if you look at all previous versions of 3DM, none of them ever predicted performance of GPUs in next gen games on their own. Imagine 1.5 years ago you ran 3DM11, what conclusions could have you have made about 680 vs. 7970GE in Crysis 3, TR, Sleeping Dogs, Dirt Showdown, Shogun 2? No conclusions. Just like now testing 3DM13 tells us 0 about how our cards will run Watch Dogs, Witcher 3, BF4, etc.

Most importantly, no one here addressed the benchmark I linked where 680 is more than 60% faster than 580 in 3dM11 but in real world games it's more like 35-36%. That proves 3dM is unreliable. Why should we rely on comparing 7970GE vs. 9970 vs. 780 in another iteration of such bench from the same firm?

 

[DiogoDX]

Maybe I should have been more clear. Within a generation from a single IHV. 3DMark predicts AMD vs AMD and NV vs NV well.

Agree. 3DMark is good to compare only the same brand.

A little example: http://www.legitreviews.com/msi-n760-hawk-video-card-review_2253/9

Titan gpu score in firestrike extreme is 36% faster than 7970GHZ while in the five games tested the average is ~30%.

Radeon 7970GHZ gpu score is 24% faster than 7950be and in the games the average is ~25%.

 

[raghu78]

How do you do that without using next generation game engines that will be the dominant force for PC gaming? You need to make benchmarks incorporating tessellation, global illumination, particle effects using UE4, Frostbyte 3.0, Crytek 3.4, etc. What 3DM does it take their proprietary engine and throws these effects in. That tells us absolutely nothing because no future game will use 3DM game engine. Just because they threw in tessellation and depth of field doesn't tell me anything about how GTX780 will cope with UE4 games vs. 7970GE.

no game needs to use 3DM engine. but 3DM gives a good idea of capable game performance. very simple. between competing products a GPU vendor might have a slight edge in a particular engine / game. But thats rarely the case across a wide range of games/ game engines.

you still haven't replied to my statement

"on average the GTX 780/ Titan is 20% /33% faster than HD 7970 Ghz when compared across a wide range of games. that perf diff agrees with 3DMark Firestrike perf diff of the above cards. that in itself is commendable."