Official AMD Radeon 6000 Series Discussion Thread

[TheMailMan78]

Wow the more PR slides I see more dissapointed that I am with AMD on this release. First the PR spin of claiming an improvement on perf/die area where it doesn't exist, using HD6870 and HD5850 to fake it out and now these. Despite the improvement on tesselation they cannot compete with Nvidia cards, so they release this BS. Yes AMD, certainly the rasterizer can be a bottleneck that's why Nvidia put 4 of them in Fermi. :shadedshu

But obviously instead of doing something like adding rasterizers or STFU, they have to try to limit the ammount of tesselation in games, so that the weakness of their architecture in regards to tesselation is not exposed. And they are doing it in a pathetic way at that. I mean, ideally you want 16 pixels/polygon? What garbage is that?

Let's see:

1920 x 1200 = 2304000 pixel -> 2304000 pixel / 16 pixel/poly = 144000 poly
1680 x 1050 = 1764000 pixel -> 1764000/16 = 110250 poly

110k-144k polygons? That's the limit they are suggesting?

Nearly all modern games already use 500k - 1 million polys, without ANY tesselation. :shadedshu

Crysis has around 1-2 million. Nanosuit model alone is 60k poly.
GOW2 and ME2 characters are 15k-30k each...

Even moving to the "innefficient" 1 pixel/poly we are talking about 1.76-2.3 million polys, so AMD is saying that games using tesselation shouldn't really surpass the polycount of 3+ years old games. Wonderful AMD, you know what? Nvidia did envision that bottleneck and did something to fix it on their architecture, so either do something or STFU.

tessellation should reduce poly count. Not increase it. Thats all they are saying and they are correct.

ECH linked Tigerdirect and Amazon yesterday. One was $274, the other was $269, if I remember correctly.

Thats to much. That means the 5850 replacement (6950) is going to be around 350 or 360.

 

[Benetanegia]

tessellation should reduce poly count. Not increase it. Thats all they are saying and they are correct.

No. Poly count should be reduced before aplying tesselation. After tesselation has been applied polycount should be much higher than on 3 year old games, otherwise what's the point to even use tesselation.

What they are asking for is reduced tesselation so that it doesn't exceed their rasterization limit and they are calling it doing "Tesselation the right way". BS.

 

[Delta6326]

HIS H687F1G2M Radeon HD 6870 Video Card - 1024MB, GDDR5, PCIe, Dual mini-Display Port, HDMI, Dual DVI
i think it is priced right around $270

 

[Mr McC]

No. Poly count should be reduced before aplying tesselation. After tesselation has been applied polycount should be much higher than on 3 year old games, otherwise what's the point to even use tesselation.

What they are asking for is reduced tesselation so that it doesn't exceed their rasterization limit and they are calling it doing "Tesselation the right way". BS.

What does it matter if the games look better? Why are you so intent on criticising AMD's new method of handling tesselation before any of us have even seen it? Might be smarter to wait and see.

 

[cadaveca]

No. Poly count should be reduced before aplying tesselation. After tesselation has been applied polycount should be much higher than on 3 year old games, otherwise what's the point to even use tesselation.

What they are asking for is reduced tesselation so that it doesn't exceed their rasterization limit and they are calling it doing "Tesselation the right way". BS.

Well considering that ATi has like 10+ years with tesselation hardware, I think they have it right.(it was called TruForm before, and tesselation is from DirectX 8, BTW. Yes, it's taken nvidia 10 years to get working hardware http://www.anandtech.com/show/773)

Now, here's the thing. There's tesselation, and then there's tesselation used effectively. I've said before in the past that tesselation should not be used the way it is currently, and should be used as a way to increase resolution(I mean triangle resolution), and limit used system bandwidth.

We really do need to control data being sent between cpu-gpu. In a big way.

And no, not now exactly, but within 3 years, if things continue as they are now, PCI-E 6.0 won't be enough.

That is ALL tesselation has been about.

nVidia's tesselation is pure overkill. We don't need it now. In two years...OK.

Thats to much. That means the 5850 replacement (6950) is going to be around 350 or 360.

That's actually a nod to 5-series owners, IMHO. If $279 for 6870, and 5870 is a bit faster......I was actually expecting $299, so anything less is great news to me.

 

[Benetanegia]

What does it matter if the games look better? Why are you so intent on criticising AMD's new method of handling tesselation before any of us have even seen it? Might be smarter to wait and see.

It's not their method that I'm criticizing, it's their PR bullshit. I would have no problem if they had released the cards, released some slides, like the graph comparing HD5000 tesselation vs HD6000 tesselation, etc and see what happens. But that thing about using 16 pixels/poly etc is complete BS and it's not even their goal (16 pixel/poly), because as I said games are already doing almost as low as 2-3 pixels/poly, they just want it to be as high as posible so that Nvidia cards never have an advantage.

The fact that they took the time to release this thing is proof enough that their method of doing tesselation is heavily bottlenecked and they know it.

Well considering that ATi has like 10+ years with tesselation hardware, I think they have it right.(it was called TruForm before, and tesselation is from DirectX 8, BTW. Yes, it's taken nvidia 10 years to get working hardware http://www.anandtech.com/show/773)

nVidia's tesselation is pure overkill. We don't need it now. In two years...OK.

I don't care how much time AMD or Nvidia have been doing tesselation. It's irrelevant.

Now, here's the thing. There's tesselation, and then there's tesselation used effectively. I've said before in the past that tesselation should not be used the way it is currently, and should be used as a way to increase resolution(I mean triangle resolution), and limit used system bandwidth.

We really do need to control data being sent between cpu-gpu. In a big way.

And no, not now exactly, but within 3 years, if things continue as they are now, PCI-E 6.0 won't be enough.

That is ALL tesselation has been about.

Then I totally pass about tesselation on games. If all that is going to be used for is to improve performance while making models a little bit more rounded, I don't care about it at all, because performance is already enough.

IMO tesselation is not being used enough, it should be used much more extensively in order to use displacement maps correctly*, because using tesselation to make characters look a little bit more rounded is a total waste of time. Even on Heaven Extreme or Stonegiant you can clearly see that more triangles are needed in order to get the displacement maps looking well. Since walls are always going to be made of two triangles, very high levels of tesselation are needed.

* The goal, to be achieved in the next generation or two should be to have at least 1 polygon per texel, so that there's no aliasing when doing displacement.

 

[EastCoasthandle]

No. Poly count should be reduced before aplying tesselation. After tesselation has been applied polycount should be much higher than on 3 year old games, otherwise what's the point to even use tesselation.

What they are asking for is reduced tesselation so that it doesn't exceed their rasterization limit and they are calling it doing "Tesselation the right way". BS.

No that's incorrect and is not BS in the context in which you are trying to post it. There is an absolute limit as to how much tessellation any given object needs. NV wants to beat AMD by using more than is needed because they have the capability to do so. However, as already demonstrated through Heaven BM more than what's needed doesn't make a better image nor a better gaming experience. There should be a limit as to how much tessellation any object gets. Furthermore, objects at a distance do not need the level of tessellation as an object up close need. Some objects may need tessellation while others may not, etc.

 

[Benetanegia]

No that's incorrect in the context in which you are trying to post it. There is an absolute limit as to how much tessellation any given object needs. NV wants to beat AMD by using more than is needed because they have the capability to do so. However, as already demonstrated through Heaven BM more than what's needed doesn't make a better image nor a better gaming experience. There should be a limit as to how much tessellation any object gets.

No. We are nowhere near the ammount of tesselation that is needed. And Heaven does not use more than needed by any means, like I said displacement maps are really poor in that benchmark. True that they do use more polys than required on other areas, like flat surfaces that have been split into 64 polys, but the amount of polys that are lacking in the areas with displacement heavily outweight those other areas where it's excesive. Overall you want more triangles not less.

The problem is that AMD and fanboys want tesselation to be nothing but a way to make models look a little bit more rounded (i.e Dirt2, Stalker...), and that's a total waste, because developers could just use more polys on those areas and AMD/Nvidia could just devote more silicon to rasterizers and SPs to make up for the difference instead of devoting it to tesselators.

 

[cadaveca]

No offense, Bene, but i don't think you can actually speak for what people want.

Nobody really cares about what tesselation will do...they care about perforamnce, and visual fidelity...not how those things are done.

I guess you didn't watch the 6870 bullet physics demo? There are far more uses for tesselation that you think. Simply adding huge poly counts to objects isn't the only use. That's just a brute-force thing...and nV is the huffing, slobbering moster you hear down the hall, pushing it that way.

Anyway, here's an interesting link for those actually interested in AMD's focus with tesselation:

http://developer.amd.com/documentation/videos/InsideTrack/assets/GDC2010DemosTranscript.pdf

 

[Benetanegia]

No offense, Bene, but i don't think you can actually speak for what people want.

I guess you didn't watch the 6870 bullet physics demo?

I can speak for what I want and for what people NEED and what do NOT NEED versus what AMD/Nvidia want to sell them.

NEEDED: True tesselation. Like the one that has been in use in 3DSmax, Maya, etc as a way to being able to effectively use displacement textures.

NOT NEEDED: gimmicky tesselation. Used to make characters look rounder, when it's not necessary at all to use tesselation for that, 99% of current games use models suitable/limited to consoles hardware after all. A higher polycount without tesselation is posible. If anything Crysis already uses 60k polys on the nanaosuit model, while some other games are using 10k. i.e Stalker without tesselation. If 50k polys are used intelligently there's no need at all for tesselation in order to get rid of those hard edges.

As a 3D artist, I think that I know what I'm talking about.

I guess you didn't watch the 6870 bullet physics demo? There are far more uses for tesselation that you think. Simply adding huge poly counts to objects isn't the only use. That's just a brute-force thing...and nV is the huffing, slobbering moster you hear down the hall, pushing it that way.

No I haven't seen the demo. No I don't think there's more uses to tesselation than the ones that I've been using for a decade...

 

[EastCoasthandle]

No. We are nowhere near the ammount of tesselation that is needed. And Heaven does not use more than needed by any means, like I said displacement maps are really poor in that benchmark. True that they do use more polys than required on other areas, like flat surfaces that have been split into 64 polys, but the amount of polys that are lacking in the areas with displacement heavily outweight those other areas where it's excesive. Overall you want more triangles not less.

The problem is that AMD and fans want tesselation to be nothing but a way to make models look a little bit more rounded, and that's a total waste, because developers could just use more polys on those areas and AMD/Nvidia could just devote more silicon to rasterizers and SPs to make the difference instead of devoting it to tesselators.

There is more to DX11 than just tessellation. Adding more tessellation in most cases will only distort or bloat. This is why adding more is not always better, warranted or needed. Therefore I believe you are incorrect in thinking that more tessellation is needed.

It's apparent in your posts that you are unaware of any other features that DX11 offers. Such as better multithreaded support. The use of HDAO (AMD) while nvidia can use HBAO (NV). In DX11 post processing is done full screen w/o the same performance penalty as found in DX9. Although there are techniques used to compensate for this in DX9 they are not needed in DX11 (however, that still depends on what is actually be post processed). And, HDR in DX9 is done at a floating point color depth of R8G8B8A8. DX11 can do full floating point color depth R16G16B16A16.

Those are just a few that I can think of. But there is more to DX11 than just tessellation. Which is part of the reason why I believe your assessment is not correct. Because there is more to it than just tessellation here. So again, no we don't need more tessellation. We need to make sure that new games incorporate all the options that DX11 offers. Since you've not mentioned that in your recent posts you are clearly unaware there is more involved.

I know your MO is to argue your own opinion be it wrong or otherwise. So I know we will agree to disagree. However, I took the time to post this so that passer bys looking for more information can read it. And, if there is more information about DX11 they can freely post it.

Now having said this, with the release of better DX11 cards I do hope to see better improvements in DX11 as whole. This is important because tessellation is not the only feature that DX11 offers.

 

[Benetanegia]

There is more to DX11 than just tessellation. Adding more tessellation in most cases will only distort or bloat. This is why adding more is not always better, warranted or needed. Therefore I believe you are incorrect in thinking that more tessellation is needed.

It's apparent in your posts that you are unaware of any other features that DX11 offers. Such as better multithreaded support. The use of HDAO (AMD) while nvidia can use HBAO (NV). In DX11 post processing is done full screen w/o the same performance penalty as found in DX9. Although there are techniques used to compensate for this in DX9 they are not needed in DX11 (however, that still depends on what is actually be post processed). And, HDR in DX9 is done at a floating point color depth of R8G8B8A8. DX11 can do full floating point color depth R16G16B16A16.

Those are just a few that I can think of. But there is more to DX11 than just tessellation. Which is part of the reason why I believe your assessment is not correct. Because there is more to it than just tessellation here. So again, no we don't need more tessellation. We need to make sure that new games incorporate all the options that DX11 offers. Since you've not mentioned that in your recent posts you are clearly unaware there is more involved.

I know your MO is to argue your own opinion be it wrong or otherwise. So I know we will agree to disagree. However, I took the time to post this so that passer bys looking for more information can read it. And, if there is more information about DX11 they can freely post it.

And now. Who is talking about DX11? I'm talking about tesselation because that's what the slide is about. Fanboys are fanboys are fanboys :shadedshu

 

[cadaveca]

No I haven't seen the demo. No I don't think there's more uses to tesselation than the ones that I've been using for a decade...

So, you aren't aware of new tech, or how it can be used, but you can speak for others about what they need from it?


Anyway, let's take your example of a 2 triangle wall. Applied to the surface is a bump map and displacement map to create the resemblance of depth.

But that depth isn't real.

So, player comes along, and blows up that wall...

Wait...how do you blow up and break apart just two triangles? I mean, you didn't need more than 2 triangles to create the wall...

So, obviously, you use tesselation to create the extra geometry so you can deform the wall.

But...to apply tesselation before-hand, when it's not needed yet, is the wrong approach.

Nvidia would like you to apply the tesselation first, and then the deform is no big deal...ATI says sure, you can tesselate the object, but it only makes sense(or rather, more optimized) to do so only when you need it, and that doesn't really require alot of tesselation power, becuase it can be done in real-time.

Oh, and that's how the rest of DX11 ties in. Tesselation is but one part of the big puzzle...and a small part of blowing that wall up. Tesselation, used correctly with the rest of DX11, is how it's done right...tesselation doesn't really have any use otherwise.

Tesselation is about real-time geometry creation, not pre-rendered geometry, and as an artist, you are not looking at it correctly, IMHO.

I understand your tact, but I see bigger and better possibilities for tesselation than simply giving bumps to stuff.

 

[inferKNOX]

Now we need to know info on the 6900s!

 

[crazyeyesreaper]

Atis method makes alot more sense and it allows for it to be scaled across hardware Nvidias method would require everyone to buy high end gpus but i dont see $1000 gpus going for $100 at launch and being sold in every PC as such ATis method has been more effective and generally easier on hardware. Could alot of the tessellation we see today be done with real time geometry? yes is it practical and scalable across a range of gpus and thereby allows Developers games to be played across multitudes of hardware? fuck no, and while i like the Idea of Tessellation it dosent fix issues like stretching on objects that are animated in games it doset magically fix clipping issues or other such common problems today. Tessellation is a just a forward thinking approach to slowly replace the need for Normal maps in some situations, It can be used to allow fully destructible lvls in real time with a proper physics implementation among other things it has many uses just few we will see today tomorrow or the next day.

 

[Benetanegia]

So, you aren't aware of new tech, or how it can be used, but you can speak for others about what they need from it?


Anyway, let's take your example of a 2 triangle wall. Applied to the surface is a bump map and displacement map to create the resemblance of depth.

But that depth isn't real.

So, player comes along, and blows up that wall...

Wait...how do you blow up and break apart just two triangles? I mean, you didn't need more than 2 triangles to create the wall...

So, obviously, you use tesselation to create the extra geoemtry so you can deform the wall.

But...to apply tesselation before-hand, when it's not needed yet, is the wrong approach.

Nvidia would like you to apply the tesselation first, and then the defrom is no big deal...ATI says sure, you can tesselate the object, but it only makes sense(or rather, more optimized) to do so only when you need it, and that doesn't really require alot of tesselation power, becuase it can be done in real-time.

Like I said I'm very well aware of "new tech" although I would hardly call tesselation new tech. lol

Any posible "new" way of using tesselation in games has probably been in use in "offline" rendering for years if not decades...

I would like to see the demo anyway, so if you have a link, I'd apreciate it.

Regarding the wall.

A wall with a 1024x1024 pixel displacement texture would need 1024x 1024 = 1048576 polys (yeah that's 1 million) ideally to look perfect in a close-up (close enough to see the entire wall) and the bare minimum by my standard would be 512x512 = 262144 (262k).

So 262k-1M for a single wall. AMD is calling out for 110k-144k polygonn scenes, that's what I'm calling BS. I've never really said anything about their method of doing tesselation on HD6000, because we have not seen how it performs yet, but if they felt they needed to ask for 110k to 2 million polys scenes in a bunch of PR slides tit's clear they have a bottleneck and they want to avoid it.

 

[inferKNOX]

What do you guys think, are 6800 prices going to rise with time (like 5800s) or fall?

 

[the54thvoid]

After reading through the past few posts i think it's absolute crap that people are getting so hyped up about this and that. Tesselation, frankly, over exaggerated in heaven to make cobbles like bubbles. On the above mentioned bullet physics demo, shows it working on broken fragments to give depth (nice use).

I'm sure there are many people here that do a lot of tech work but having spats back and forth which come down to opinions and statements like, "well, i've been a 'whatever' for so long" mean jack shit.

As long as AMD and NV are dishing out the drugs, we're all stuck on SOMA - so just get used to it people.

And ffs! Some people need to take a step back and stop being such loyalists. Even when people say, I'm not anti-"insertcompanyhere" BUT.... it probably means you are. LIke that saying - "I'm not a racist BUT..."

And on topic - I think Cayman is going to spank butt.

 

[Benetanegia]

Atis method makes alot more sense and it allows for it to be scaled across hardware Nvidias method would require everyone to buy high end gpus but i dont see $1000 gpus going for $100 at launch and being sold in every PC as such ATis method has been more effective and generally easier on hardware.

http://www.geeks3d.com/20100913/ope...campaign=Feed:+TheGeeksOf3d+(The+Geeks+Of+3D)

As you can see even the GTS450 decimates the HD5870 so no, AMD's method is not better.

 

[cadaveca]

So 262k-1M for a single wall. AMD is calling out for 110k-144k polygonn scenes, that's what I'm calling BS. I've never really said anything about their method of doing tesselation on HD6000, because we have not seen how it performs yet, but if they felt they needed to ask for 110k to 2 million polys scenes in a bunch of PR slides tit's clear they have a bottleneck and they want to avoid it.

Yes, I understand what you are saying.

But, like you mentioned, using tesselation for that sort of rendering is wasting power...gpus and cpus today are way overpowered for what we use them for. What you are asking it to be used for is no different than making a straight line curved.

Again, what you are suggesting is that tesselation is used to create geometry as the scene is rendered, but really, tessleation should be used to create extra geometry where either system bandwidth requires it, or in realtime, for things like physics effects, and not applied to the whole scene.

It could be used for that, of course, but I do not see that as practicle, just like you seem to...you say it's OK, because nv offers the power to do so, but again, to me, that's a "brute force" approach, and to me, is far from elegant.

To me, what AMD is saying is that only geometry with such a low order needs tesselation...not that that's what's required to use it.

 

[EastCoasthandle]

What do you guys think, are 6800 prices going to rise with time (like 5800s) or fall?

That will depend on supply and demand.

 

[crazyeyesreaper]

also were talking 1 aspect Nvidia uses shaders for everything

Physx Tesselation regular day to day rendering in game direct compute effects etc etc so what happened when you combine all that in a brute force approach hmmm? would it still be viable sure but why would you pay say $1200 for the power to do that when by doing it only when needed can allow something in the $200 range to do the same and match 95% of the image quality.

And for someone that talks of 3D and knows what there talking about you should also know the golden rule of 3D to use only what you need to get the job done.

 

[Benetanegia]

also were talking 1 aspect Nvidia uses shaders for everything

Nope. They use shaders for shading and physics/direct compute (like AMD does) and the tesselators for tesselation. The ammount of tesselation is not affected by any other thing nor does affect any other task, other than the ones that belong to the pipeline like hull and domain shaders (and those affect AMD too) and the increased requirements due to higher detail (which is the same on AMD).

And for someone that talks of 3D and knows what there talking about you should also know the golden rule of 3D to use only what you need to get the job done.

And what is that? We could argue forever, but after everything had been considered we might as well agree that a Nintendo Wii is what you need to get the job done, as that is what the majority of people use to play.

Aside from this opinion that is based on statistics and could be considered as objectively reasoned, anything above that is subjective. DX9 as seen in XB360 or PS3 or better yet as seen in Crysis might as well be enough. DX10? DX11 with tesselation? Sure, we need tesselation, both AMD and Nvidia say so after all. So how much tesselation do we need? And for instance, how much of dynamic lighting do we need? We want all our in-game lights to be dynamic, right? We spend money on capable cards after all. But with 1 dynamic and several filling lights Valve does wonders and makes games that run confirtably on 5 year old DX9 cards, so why do we need these new generation of cards (any) after all?

 

[cadaveca]

No, I'm not asking something like making a straight line curved. I'm asking tesselation to be a replacement for normal mapping in those places where normal mapping or POM do not really work, like walls, bricks, rock grain/deformity/holes, sand, grass. Obviously the wall/rock would still have the overall shape done in actual geometry, it's the imprefections that tesselation should be used for. In characters it sould be used to create wrinkes and other facial features, old people in current game simply suck big balls. Those are good uses for tesselation on top of what you already mentioned. Apart from that I don't want tesselation, I fail to see the advantage of having tesselation.

Imprefections and wrinkles are nothing more than making straight lines curved. Part of the problem in facial rendering is a huge lack of straight lines in a real face.


It seems ot me what you are after is an easier way to get your work done...if you don't need to draw every polygon, and the renderer takes care of that for you using tesselation, then you'd be happy, as your work would be easier.

And nV kinda offers that.

AMD takes the exact opposite approach...they are saying that in high-order geometry, the usefulness of tesselation is questionable, because it can greatly distort and image. Too many polygons in a given space, and you get nothing but a big smudge.

I do not think tesselation should be used for those things you mention. I agree that current techniques need to change, and are full of limitations, but more often than not, it's not the hardware causing the limitation..it's software.

And that was the issues 10 years ago for ATi...tesselation needed software support, and nV wasn't having none of it. Today, DX11 takes alot of the software side out the equation, and now nV has capable hardware.

And really, as the64htviod sai, tesselatin isn't really important in the context of this thread. how it's implemented is stil llargely dependant on software, and as such, has little place in a hardware thread...and even more so now, as currently tesselation isn't exactly present in many apps, so is of low-priority. It's there for devs working on next-gen console titles, and other apps that are at least a year off...and won't be common-place until at least after that.

In the meantime, both Fusion and SandyBridge will come out, and those technologies may greatly change what's needed for any given app.

 

[EastCoasthandle]

And now. Who is talking about DX11? I'm talking about tesselation because that's what the slide is about. Fanboys are fanboys are fanboys :shadedshu

Thanks for conceding the debate with that ad homeinem. But here is some food for thought. The author thinks that nv is grasping at straws with that tactic. But I'll leave it at that.