AMD Radeon HD 6970 already benchmarked? Enough to beat GTX480 in Tesselation?

[Scali]

Oh My goodness, a $240 part doesn't do something as well as a $500 part?

No, a $240 part doesn't do something as well as a $200 part.

 

[Nemesis 1]

Actually AMD has no excuse to botch up their tessellation engine:
http://en.wikipedia.org/wiki/TruForm
They have had hardware tesselllation since the Radeon 8500.
They should have wiped the floor with NVIDIA's performance in tessellation, but they made it to be the other way around.


I suspect this is another area where AMD's small die philosophy once again hit performance/features in a bad way.

I don't really understand why you think ATI/AMD should have included tessellation as a featureded performance part. ATI has had it for years and such . AMD never pushed anyone to make a benchmark for tess. Never viewed it as benchmark did they . SO NV in their first tess. Included way more than is required for present games. I personnally don't want to see the mother of all tesselattion on the 6950 6970 6990 its a waste of space.
Its the only thing the NV fans have to blow horn about in gaming GPUs .

I can relate tessellation in the way I did Hammer AMD 64. Big deal than . NOT! But hyped to heaven / But now its great.
NV cuda . Again NV fanbois blabbing about that which has nothing to do with gaming .
Still isn't all that useful. Few programms.

Intel AVX . How long befor it takes hold ? Out of the gate or more like cuda but easier.
Open CL . When will it be really useful.

Intel if your smart on the next bridge release @22nm. Include the mother of all tessellation engines and you will be declared the winner in the race to find the Holy Grail known as tessellation . I going to bookmark this one the ducks are all lined up . Easy targets they made of themselves they did.

 

[Creig]

'Overkill' depends entirely on what it is you're tessellating, and how.

So then do you keep repeating over and over and over that AMD's tessellator is lacking if it's going to come down to a game by game evaluation to determine the optimal tessellation setting for each particular future title? Why not simply adopt a "wait and see" attitude like most everybody else here? Only time will ultimately determine whether AMD's 6800 series has enough/not enough tessellation power for the games that will be released during its useful lifespan.

 

[Scali]

So you can hardly make the claim that AMD's tessellator is lacking.

Yes I can. Compared it to nVidia's tessellator, it looks bleak, VERY bleak.
You may argue that the tessellator is not that important in the overall picture, but NOT that the tessellator itself is not lacking, when the performance difference is so incredibly large:
http://www.geeks3d.com/20100826/tessmark-opengl-4-gpu-tessellation-benchmark-comparative-table/
Look at the extreme and insane results: Even nVidia's GTX460 is about 3 times as fast as AMD's 5870. And the 470 is twice as fast as that. That is a factor 6 performance difference between AMD's best and nVidia's best.
This is nothing short of a massacre. Not lacking you say? Denial I say.

 

[Aristotelian]

Have you seen the tesselation and GPGPU benchmarks?

Can you tell me how that translates into gaming performance now?

 

[Nemesis 1]

Yes I can. Compared it to nVidia's tessellator, it looks bleak, VERY bleak.
You may argue that the tessellator is not that important in the overall picture, but NOT that the tessellator itself is not lacking, when the performance difference is so incredibly large:
http://www.geeks3d.com/20100826/tessmark-opengl-4-gpu-tessellation-benchmark-comparative-table/
Look at the extreme and insane results: Even nVidia's GTX460 is about 3 times as fast as AMD's 5870. And the 470 is twice as fast as that. That is a factor 6 performance difference between AMD's best and nVidia's best.
This is nothing short of a massacre. Not lacking you say? Denial I say.

Actually NV is 1 generatiom ahead of AMD/ATI . That first r500 really put ATI behind the 8 ball . But is often the case , What goes around comes around . And ATI/AMD have caught back up with the 6,000 seies. If the 6950 beats the 480 which is likely. Than ATI has the game in hand on = generations and timing. I will call it right now when we see the 6950/6970 what card do you think AMD/ATI are gunning for with the 6950. Certainly not the 5870 because the next AT review will be on the 6950/6970 . This is when well see the drivers that maybe AMD should have put out for the 6800s but maybe AMD did it correct this way . I can't wait for the driver release and the improvement we see in the 6850/6870 series. The 6970 is going to be a great day of celabration for most all gpu fanbois who care about performance and nothing else.

 

[lavaheadache]

Yes I can. Compared it to nVidia's tessellator, it looks bleak, VERY bleak.
You may argue that the tessellator is not that important in the overall picture, but NOT that the tessellator itself is not lacking, when the performance difference is so incredibly large:
http://www.geeks3d.com/20100826/tessmark-opengl-4-gpu-tessellation-benchmark-comparative-table/
Look at the extreme and insane results: Even nVidia's GTX460 is about 3 times as fast as AMD's 5870. And the 470 is twice as fast as that. That is a factor 6 performance difference between AMD's best and nVidia's best.
This is nothing short of a massacre. Not lacking you say? Denial I say.

just downloaded and ran tessmark with my 5870's. neat little program. Anyhow, when you look at how many wire frames going on in there it is like WOWZERS.

Needless to say, when selesting anything beyond medium tesselation I saw dimishing returns but with huge performance loss. To me this sounds like AA. When you start getting past 4xAA the penalty grows exponentially and the visually quality is only really detectable under zoomed in still screen shots. Hardly something that I would even concern myself with unless it were a life or death situatiuon. If the time comes around that this kind of tesselation becomes availabe to gamers and offers a "tangible" visual gain I'll reconsider my view. Until then, I hardly see insane amounts of tesselation relevant to today's gamer, in todays games.

 

[Kenmitch]

Yes I can. Compared it to nVidia's tessellator, it looks bleak, VERY bleak.
You may argue that the tessellator is not that important in the overall picture, but NOT that the tessellator itself is not lacking, when the performance difference is so incredibly large:
http://www.geeks3d.com/20100826/tessmark-opengl-4-gpu-tessellation-benchmark-comparative-table/
Look at the extreme and insane results: Even nVidia's GTX460 is about 3 times as fast as AMD's 5870. And the 470 is twice as fast as that. That is a factor 6 performance difference between AMD's best and nVidia's best.
This is nothing short of a massacre. Not lacking you say? Denial I say.

What people are more concerned with is how it will effect their gaming experience not how big their synthetic e-peen will be.

What if Cayman does out perform Fermi in tessellation? Doesn't matter in the end if during gaming it doesn't make a noticable difference anyways.

 

[Scali]

just downloaded and ran tessmark with my 5870's. neat little program. Anyhow, when you look at how many wire frames going on in there it is like WOWZERS.

It's meant to be a synthetic test, to isolate the performance of the tessellator, hence the name TessMark.
It neither tries to render pretty pictures nor tries to use the most optimal distribution of tessellation factors.

It simply demonstrates that in areas where you were to hit a tessellation factor of 16, a GeForce will be twice as fast as its direct competitor, and when you push further to 32 or 64, it turns into three times as fast or more.
Needless to say that even if you were to use tessellation factors of 16 or higher only on small parts of the screen, the Radeons will already get a noticeable performance hit (which is what you see in Unigine Heaven for example).

It also demonstrates that there still is quite a visible difference between 8x and 16x tessellation in this case, and even some when you go to 32x. So 16x is certainly not 'overkill' yet in terms of detail. It's perfectly valid for a game to use 16x tessellation factors in some areas to avoid geometry aliasing.

 

[3DVagabond]

@Scali

Is tessellation any different than subdivision while modeling? I know it's not exactly the same thing. What I mean is does it basically increase model density the same as subD does?

 

[Scali]

What people are more concerned with is how it will effect their gaming experience not how big their synthetic e-peen will be.

I was asked to defend my claim that the tessellator is lacking, which I did.
I don't need to hear your excuses really.

 

[Scali]

@Scali

Is tessellation any different than subdivision while modeling? I know it's not exactly the same thing. What I mean is does it basically increase model density the same as subD does?

Yes, pretty much. A very common form of subdivision in modeling software is with NURBS (Non-Uniform Rational B-Splines). Maya is pretty much entirely built around the concept of NURBS. With DX11 tessellation, you can program the hardware to subdivide NURBS patches in realtime.

But you can also do other things, such as displacement mapping... or a combination of techniques.

 

[Janooo]

just downloaded and ran tessmark with my 5870's. neat little program. Anyhow, when you look at how many wire frames going on in there it is like WOWZERS.

Needless to say, when selesting anything beyond medium tessalation I saw dimishing returns but with huge performance loss. To me this sounds like AA. When you start getting past 4xAA the penalty grows expontially and the visually quality is only really detectable under zoomed in still screen shots. Hardly something that I would even concern myself with unless it were a life or death situatiuon. If the time comes around that this kind of tessalation becomes availabe to gamers and offers a "tangible" visual gain I'll reconsider my view. Until then, I hardly see insane amounts of tessalation relevant to today's gamer, in todays games.

That's the most important point. Some people just don't get it.

 

[tviceman]

What if Cayman does out perform Fermi in tessellation?

Wouldn't it be ironic given that AMD has been preaching about tessellation done "the right way"? Leading up to the launch of the 6800 series Huddy and the rest of AMD have been preaching that Nvidia's approach is overkill and is actually reducing image quality in the end.

It woudl pretty much be like AMD saying "Nvidia's idea of tessellation is too much and it actually makes things look worse, but we went ahead and wasted resources on making sure we can still out tessellate them anyways."

 

[Dark Shroud]

With the 6000 series AMD

• Brought Crossfire performance upto SLI's levels if not better.
• Added 3D support for Blu-ray & games. You just have to buy the 3D kits from 3rd party partners.
• Added OpenCL to their drivers for GPGPU.

So what's left to complain about & rip on AMD for tessellation. A feature Nvidia has never had before and actively blocked from being added to DX10. Only a handful of games in the market even use this feature, one of which just had it bolted on.

Some of you guys should at least wait for the 6900 series cards to be released before just tearing into AMD. And by the time tessellation is actively used in common games the people whom that matters to will already have upgraded to newer cards with better tessellation performance.

 

[3DVagabond]

Yes, pretty much. A very common form of subdivision in modeling software is with NURBS (Non-Uniform Rational B-Splines). Maya is pretty much entirely built around the concept of NURBS. With DX11 tessellation, you can program the hardware to subdivide NURBS patches in realtime.

But you can also do other things, such as displacement mapping... or a combination of techniques.

Well, the reason I'm asking is I have a model I'm making using SubD (hypernurbs in C4D). The basic model, without SubD, is 728 poly (quads). It's about as smooth as Legos. Way too lowres for game use in anything but very distant shots. If I SubD w/ a factor of only 6 the poly count jumps to +2.9 million. At that poly count the model appears virtually solid. Way too dense for any game use. A SubD factor of 4 gives it 187K. Hires enough to appear perfectly smooth for any rendering short of Hollywood. This is a model with a lot of curves, so it's not because it's boxy. Unless I'm not understanding something, I can't see any use for a Tess factor above 4 in games.

 

[Scali]

A feature Nvidia has never had before and actively blocked from being added to DX10.

Partly incorrect: GeForce3 supported RT-patches in DirectX 8.
The rest, that's a pretty strong accusation, have any proof to back it up?

 

[lavaheadache]

For anybody looking to play around with Tessmark. In order to utilize crossfire, the exe must be renamed to something else that supports opengl. I used ETQW. blamo, crossfire works. I found nothing on the web about this just a little trial and error.

 

[Scali]

Well, the reason I'm asking is I have a model I'm making using SubD (hypernurbs in C4D). The basic model, without SubD, is 728 poly (quads). It's about as smooth as Legos. Way too lowres for game use in anything but very distant shots. If I SubD w/ a factor of only 6 the poly count jumps to +2.9 million. At that poly count the model appears virtually solid. Way too dense for any game use. A SubD factor of 4 gives it 187K. Hires enough to appear perfectly smooth for any rendering short of Hollywood. This is a model with a lot of curves, so it's not because it's boxy. Unless I'm not understanding something, I can't see any use for a Tess factor above 4 in games.

The subdivision factor is probably not the same as the tessellation factors in DX11. Sounds like it adds polys far faster than what DX11 does (could be a recursive algo, each factor subdividing each quad into 4? Then you'd get 2.9M at 6 subdivisions).
The tessellation factor for a quad is just the number of points to generate on an edge.
So if you use a tessellation factor of 6, you'd get 6x6 = 36 quads.
So 728 polys would turn into 728*36 = 26k polys, not quite in the millions.

 

[Janooo]

Partly incorrect: GeForce3 supported RT-patches in DirectX 8.
The rest, that's a pretty strong accusation, have any proof to back it up?

http://www.tomshardware.com/reviews/opengl-directx,2019-7.html

Initially planned for Direct3D 10 (which explains its presence in the Radeon HD series), it seems that Microsoft, ATI, and Nvidia weren’t able to reach an agreement in time, and so it disappeared from the specifications, only to return with a vengeance with Direct3D 11.

 

[Nemesis 1]

Partly incorrect: GeForce3 supported RT-patches in DirectX 8.
The rest, that's a pretty strong accusation, have any proof to back it up?

I will help him out. Go get the specs of Vista befor MS changed the specs because NV was cring to MS.

 

[Chiropteran]

Speed is proportional to the price of the hardware. You expect each card to perform according to its price.
And that's the thing with tessellation... it shifts this metric in nVidia's favour currently. Not only on the fastest cards, but on all cards.

Nvidia is better at current artificial tessellation benchmarks, I agree with that. I don't see how that shifts the entire price/performance in favor of nvidia, though. I mean look at the big picture. In games, AMD is better AND cheaper. In artificial tessellation benchmarks, nVidia is better, not really cheaper though. And then you must consider the consumer's usage pattern. Is it 50% games and 50% artificial tessellation benchmarks? I don't think so, I think it's more like 98% games and possible 2% tessellation benchmarks, and that is just being generous to nvidia. Other than a few geeks and programmers the vast majority of the consumers buy cards to play games. Nvidia's slight edge on some artificial benchmark is just that- a slight edge. It isn't nearly enough to "shift the metric in nvidia's favor".

 

[Scali]

http://www.tomshardware.com/reviews/opengl-directx,2019-7.html

Okay, so since Microsoft, ATi and nVidia couldn't reach an agreement, it's all nVidia's fault?
As for the presence in ATi's DX10 hardware? Nonsense. AMD has been adding tessellators to their hardware since the Radeon 8500 (as they themselves showed on the slides when they were explaining their improved tessellators for the 6800 and 6900 series).
In that light it's not so 'remarkable' that their DX10 hardware had a tessellator as well.
The chip in the XBox 360 also has one, and that is also DX9-class hardware.

 

[Scali]

I will help him out. Go get the specs of Vista befor MS changed the specs because NV was cring to MS.

You mean Intel, not nVidia.

 

[Scali]

I don't see how that shifts the entire price/performance in favor of nvidia, though. I mean look at the big picture.

This is going round in circles.