Tessellation

[eastmen]

To avoid discussions about whose tessellator is bigger cause that's where a lot of people seem to end up and in doing so completely miss the point. Even if there was no Fermi we could still make the argument that tessellation as we've gotten in games and hardware is rather unimpressive. I see folks saying that AMD's tessellation hardware is handling the weak implementations in current games just fine. That's not exactly a redeeming argument.

The question isn't if they are weak or not. Your the one saying lets not compare to other hardware. So the question should be , is the tessellation amd's hardware capable of , worth while. I have a 5850. On heaven benchmark there are diffrent levels of Tessellation disabled , moderate , normal , extreme.
My card runs the benchmark at high frames with normal tessellation at 1920x1080 and can run it quite well at 5760x1080 with moderate tesselation. Both of these modes offer a leap in image quality . Why are we dismising what these modes can do ?

AMD has been increasing tessellation performance. Barts offers better tessellation performance than cypress at lower costs and a smaller chip. Cayman may again offer better performance than bart. So its not like developers can't start using tesselation.

 

[trinibwoy]

Why are we dismising what these modes can do ?

Heaven wasn't dismissed and is widely used as a synthetic tessellation benchmark. I still have an open question though. Given all the hype around 7 generations of tessellation hardware where are AMD's demos to back that up? It's really surprising how little DX11 content they've actually produced that highlights tessellation after pushing it for so long. If they did that we would already know whether Evergreen tessellation was good enough.

 

[AlphaWolf]

Heaven wasn't dismissed and is widely used as a synthetic tessellation benchmark. I still have an open question though. Given all the hype around 7 generations of tessellation hardware where are AMD's demos to back that up? It's really surprising how little DX11 content they've actually produced that highlights tessellation after pushing it for so long. If they did that we would already know whether Evergreen tessellation was good enough.

AMD isn't a content provider.

 

[eastmen]

Heaven wasn't dismissed and is widely used as a synthetic tessellation benchmark. I still have an open question though. Given all the hype around 7 generations of tessellation hardware where are AMD's demos to back that up? It's really surprising how little DX11 content they've actually produced that highlights tessellation after pushing it for so long. If they did that we would already know whether Evergreen tessellation was good enough.

There are tessellation demos from amd.


Also I disagree that we woukd know what evergreen's tessellation was worth. It takes years for games to come out that use the full feature set of a new dx or to implement it into the core of the game.

remember Evergreen was a lower end gpu priced at under $200 in late 2009. It had diffrent targets to hit performance wise /price wise / die size wise vs what amd was able ot put out with barts.

Anyway this is a discussion that will go no where but in circles cause there are no tessellation games that don't play well on amd's hardware. mabye we will get to that point at some time. But isn't that true for all hardware ? No matter what gpu it is , the rage 3d , voodoo 2 , r300 , gf 100 , cypress at some point the hardware will no longer be sutiable for a targeted play experiance.

 

[trinibwoy]

There are tessellation demos from amd.

Where?

Anyway this is a discussion that will go no where but in circles cause there are no tessellation games that don't play well on amd's hardware.

That was already addressed. It's not about whether AMD's hardware can handle the weak displays of tessellation to date. After all many of those games were developed on AMD hardware and could very well have been limited by the available performance. Of course, if you're impressed with tessellation usage in current games then you should be happy. As the owner of a card with geometry power to burn I want to see more detailed models and not just rounded heads or super tessellated flat ground.

Btw, the "new feature" excuse doesn't work when you're boasting that it's your 7th generation of the hardware implementation.

 

[Harison]

Perhaps they wanted tessellation to be a useful option for enhancing games as opposed to some epeen inflating synthetic that allows you to make more triangles than is in anyway useful.

Exactly. Some fans are ecstatic over few selected Nvidia-paid games/benches, some of those discard 75% of the tes. work done, and the only reason this nonsense is green-lighted with green-dollars is because Geforce have lesser penalty than Radeons. And then certain people blame AMD because they didnt planned for this nonsense

Like in other thread I gave a hypothetical example of 2560x with 8x AA as good enough for gamers (AMD approach), yet if lets say Nvidia gets less penalty at 24x AA, they push it as a default AA, and its fans are claiming 2560x 8x AA is "inferior", "weak" and "not good enough" In the end if it becomes a public perception, HW maker have to make 24x AA as a default too, even though there would be no visible gains for the end-users. Real life example - 3DMark, its almost useless for gamers, yet vendors spend a lot of time optimizing cards for it, simply because of the public perception.

 

[AlphaWolf]

Where?

Is your google broken?

Try Froblins.

 

[trinibwoy]

Is your google broken?

Try Froblins.

Is your english broken? Froblins is a DX10.1 demo. An unimpressive one at that. Really if that's the best you can come up with you're essentially making my point for me Please don't turn this into a defend AMD at all costs campaign. It's clear for everyone to see that tessellation has not been brought to market effectively.

 

[AlphaWolf]

Is your english broken? Froblins is a DX10.1 demo. An unimpressive one at that. Really if that's the best you can come up with you're essentially making my point for me Please don't turn this into a defend AMD at all costs campaign. It's clear for everyone to see that tessellation has not been brought to market effectively.

So you expect AMD to demonstrate every feature in every new DX version?

 

[trinibwoy]

So you expect AMD to demonstrate every feature in every new DX version?

Yes, of course! They've had the hardware before everyone else and it's not just some little feature. It's something that's been hyped for ages, something for which they've had hardware languishing unused for years. Flaunting it with some awesome looking demos is the least I would expect, wouldn't you?

Anyway, AMD isn't solely responsible for content that highlights tessellation. However, if they had their own demos out showcasing how great tessellation could be on Evergreen hardware then we could blame developers for not effectively using the hardware. If they can't even showcase their own stuff what message does that send?

Bottom line is I don't think what we've gotten so far content wise is impressive in the least and I think we can make good use of lots more polygons. The only reason I'm scapegoating AMD here is because there were previously the champions of tessellation and are the ones I would have expected the most from but now they seem to be standing on the sidelines.

 

[AlphaWolf]

Yes, of course! They've had the hardware before everyone else and it's not just some little feature. It's something that's been hyped for ages, something for which they've had hardware languishing unused for years. Flaunting it with some awesome looking demos is the least I would expect, wouldn't you?

Anyway, AMD isn't solely responsible for content that highlights tessellation. However, if they had their own demos out showcasing how great tessellation could be on Evergreen hardware then we could blame developers for not effectively using the hardware. If they can't even showcase their own stuff what message does that send?

Bottom line is I don't think what we've gotten so far content wise is impressive in the least and I think we can make good use of lots more polygons. The only reason I'm scapegoating AMD here is because there were previously the champions of tessellation and are the ones I would have expected the most from but now they seem to be standing on the sidelines.

I think you're being absolutely ridiculous. It's just another damn feature. We'd be way further ahead if more than one vendor would have adopted the feature years ago than if AMD had released 5203958723 demos of it. The hardware is out there, the know how is out there, its in the hands of developers at this point.

 

[Razor1]

The question isn't if they are weak or not. Your the one saying lets not compare to other hardware. So the question should be , is the tessellation amd's hardware capable of , worth while. I have a 5850. On heaven benchmark there are diffrent levels of Tessellation disabled , moderate , normal , extreme.
My card runs the benchmark at high frames with normal tessellation at 1920x1080 and can run it quite well at 5760x1080 with moderate tesselation. Both of these modes offer a leap in image quality . Why are we dismising what these modes can do ?

AMD has been increasing tessellation performance. Barts offers better tessellation performance than cypress at lower costs and a smaller chip. Cayman may again offer better performance than bart. So its not like developers can't start using tesselation.

Once the tessellation algorithm is implemented its easy to increase complexity very easily from a developers point of view. If AMD provided increased tessellation performance developers could have increased detail levels in games. Pretty simple to do. Its not like developers have to create something new, its just a few lines of code.

And Barts doesn't really provide increased tessellation performance, at least not real world performance once you increase the iterations, they need additional tessellation units, Cayman will probable address this.

 

[Razor1]

I think you're being absolutely ridiculous. It's just another damn feature. We'd be way further ahead if more than one vendor would have adopted the feature years ago than if AMD had released 5203958723 demos of it. The hardware is out there, the know how is out there, its in the hands of developers at this point.

Demos give examples for developers to develop their own versions, demos are very helpful, that's wy the GPU gems and nV's developer's site is so popular, they provide alot of demos with code, gives a nice staging ground for devs.

 

[trinibwoy]

I think you're being absolutely ridiculous.

If you think it's ridiculous to aggressively market and promote one of the most visible advances in DX11 with a 6 month lead to market over your competition then you must think having common sense or any sort of business or marketing acumen is ridiculous as well. It's one feature but it's a huge one and has been touted as the second coming for years. Trying to diminish it doesn't help your argument.

But let's cut to the chase. Are you impressed with tessellation thus far and are you happy with the tacked on implementations by console developers that we've seen to date? If your answer is yes to both questions then I don't think we'll ever get on the same page. Personally, I want to see more cool shit.

 

[AlphaWolf]

If you think it's ridiculous to aggressively market and promote one of the most visible advances in DX11 with a 6 month lead to market over your competition then you must think having common sense or any sort of business or marketing acumen is ridiculous as well. It's one feature but it's a huge one and has been touted as the second coming for years. Trying to diminish it doesn't help your argument.

But let's cut to the chase. Are you impressed with tessellation thus far and are you happy with the tacked on implementations by console developers that we've seen to date? If your answer is yes to both questions then I don't think we'll ever get on the same page. Personally, I want to see more cool shit.

Tessellation has gotten as much (if not more) use (in the first year of both major IHV supporting) than any other new hardware feature.

 

[3dcgi]

Demos give examples for developers to develop their own versions, demos are very helpful, that's wy the GPU gems and nV's developer's site is so popular, they provide alot of demos with code, gives a nice staging ground for devs.

AMD didn't provide a "pretty" demo, but did provide a demo that is included in the DX11 SDK which shows developers how to use tessellation. The one with the stone ground.

 

[DavidGraham]

What about adaptive tessellation ? I am sick to the bones for seeing geometry and detail pop ups , it destroys immersion in my games , THAT is something I want it to be addressed as early as possible .

 

[MfA]

For good adaptive tesselation the tesselation factor has to depend on the screenspace length of the triangle edge, and be performed in screen space ... which is a patently obvious idea and has thus of course been patented. This is not going to help speed adoption.

I wouldn't have expected that from Tomas Möller

 

[Sxotty]

What about adaptive tessellation ? I am sick to the bones for seeing geometry and detail pop ups , it destroys immersion in my games , THAT is something I want it to be addressed as early as possible .

I agree popping is terrible, better to be lower polygon count if that is the tradeoff. (Obviously it should not be the tradeoff).

 

[Tchock]

Would this be of any help? Current situation anyway.

3DMark 11 eXtreme模式

HD6870___2302

HD6850___2269

GTX 460 1GB___1772