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

[Aristotelian]

Scali explained it better than I did. Anyways, welcome to my ignore list

(See, I can do bold too!)

Why do people feel the need to announce that they have put someone on ignore? it just comes off as a juvenile "getting the last word".

And Phil1977, I'm with you on this (so far). Scali has commented that we are likely to see a lot more tessellation in the future, so if that turns out to be the case, perhaps the effects of much higher tessellation levels (and the hardware to support it) will be far more obvious to the relevant gamer.

 

[Arkadrel]

time will tell... currently its not a issue though.
And supposedly its fixed in the 6950-6970 so it can scale like the 4xx cards do.

it all comes off as just nvidia guys rubbing it in for no reason (when games arent currently a issue with it).

 

[3DVagabond]

I believe the reason we take a performance hit, in current games, with tessellation on is because the models used are more complex than they need to be so they don't appear as flat blocky objects for those w/o Dx11 hardware. Tessellation adds detail. That's always going to take a performance hit, all else being equal. Once games are made to work only with tessellation (IE, when you turn tessellation off the game turns into Lego world, model complexity wise) you'll see overall game performance go up. That won't happen until consoles support it for 99% of games. Right now they'd have to build 2 sets of models, one for use with tessellation and one for use without for us to see a performance improvement with tessellation on.

 

[Arkadrel]

Well I'm not convinced of tesselation (yet).

The few games that support it, to be honest I couldn't tell the difference. Ever wondered why all the tesselation demos use wireframes?

I also found that the performance impact is quite large, especially when other DX11 features are used (e.g. tesselation and advanced shadows in Alien vs. Predator)

I tend to agree.. though Ungine shows promis... esp with the brick tiles on houses, and the stones that make up a road. But yeah more tessellation isnt nessarly a good thing unless its used wisely so it had a image improvement effect. Sadly that is not the case with alot of the tessellation we see, its being used for minor stuff that overall has very little image qualty effect.

I believe the reason we take a performance hit, in current games, with tessellation on is because the models used are more complex than they need to be so they don't appear as flat blocky objects for those w/o Dx11 hardware. Tessellation adds detail. That's always going to take a performance hit, all else being equal. Once games are made to work only with tessellation (IE, when you turn tessellation off the game turns into Lego world, model complexity wise) you'll see overall game performance go up. That won't happen until consoles support it for 99% of games. Right now they'd have to build 2 sets of models, one for use with tessellation and one for use without for us to see a performance improvement with tessellation on.

This... you just worded what I wanted to say.

once games start becomeing dx11 ONLY (no dx9 or 10 support) we ll see tessellation DONE RIGHT.
Until then its a pipedream for anything else, and nvidia fanboys pushing this on forums is way overhyped. Sadly some people probably wont admit to that.

1) However there are still alot of people that use windows XP, which doesnt support dx11 (mircosoft is doing that on purpose to drive sales on win7). So that ll slow down how fast the gameing industry takes on useing tessellation the right way.

2) Some people have old grafics cards.... tessellation done right, would mean alot of the older non 4xx or 5xxx+ series wouldnt be able to buy games or play them. This might be a issue for some game developers that want to sell more copies by allowing non- dx11 cards to be able to play the game at a decent experiance.

3) most games sold are to the consols, which dont have tessellation = game ports take alot of work to add tessellation and then its on a already overdone model so gains are small = tessellation NOT done the right way. This means it ll first be around 2012 when we see heavy tesellation games appear (if consol refresh go with a chip that has good tessellation).


to me this all sums up to:
Nvidia jumped the gun with tessellation, their overshooting currently for something that wont be needed for another 2 years+ time.

 

[PingviN]

1) However there are still alot of people that use windows XP, which doesnt support dx11 (mircosoft is doing that on purpose to drive sales on win7)

More like you get new features when you buy new stuff. WDDM needed to be reworked, it's not surprising Microsoft didn't invest time in updating it for XP too. You can't expect old products to get everything newer do. I seriously doubt that DX11 (and 10, for that matter) is of such important to most gamers. If they're happy with XP, they'll stick with XP. I think it's reasonable for Microsoft not to do extensive reworking of XP code when they're releasing a new operating system (Vista, in this case).

 

[Arkadrel]

yeah but it means we wont see "tessellation done the right way"... for a few years because people are still useing old tech.

and by old tech I mean non dx11 stuff like winxp/2xx or lower cards + 4xxx series or lower.

*** edit: I love win7.... I skipped vista (I friend bought it, saw how it was on his pc).. was useing XP until win7. Vista + Windows ME >_> ms what where you thinking?

 

[3DVagabond]

I think when consoles support tessellation, we'll start seeing it done right. Until then, we'll see game models designed to look good without tessellation, and that will keep tessellation an IQ feature, not anything that will boost performance.

 

[Scali]

That won't happen until consoles support it for 99% of games. Right now they'd have to build 2 sets of models, one for use with tessellation and one for use without for us to see a performance improvement with tessellation on.

You can build the non-tessellated set from the tessellated one by simply performing the tessellation offline.
So you really don't have to wait for consoles or a wide adoption of DX11 hardware in the PC market. You can have your cake and eat it too.

Basically it's no different from what has always been done: have a set of different level-of-detail models, and the highest details are only enabled for high-end systems. So you always had geometry that was not used on certain (most?) systems.

I don't see why people try to create a false dilemma about tessellation.

 

[3DVagabond]

That's if they are designing the game for Dx11 and then adapting it for consoles. So far, seems more like the other way around. They are adding superficial Dx11 effects to games.

 

[Scali]

That's if they are designing the game for Dx11 and then adapting it for consoles. So far, seems more like the other way around. They are adding superficial Dx11 effects to games.

Well yes, but that doesn't mean that's the ONLY way.
Obviously any 'DX11' games so far have been in development since long before DX11 hardware arrived.
Some of these games have geometry which can easily be adapted to DX11 tessellation, such as the terrain in Civ5 and HAWX 2.
For other games, it depends on how the art pipeline was designed years ago, and how much they are willing to adapt to support DX11.
The irony is that generally the modeling software applies tessellation anyway, since you generally don't model everything down to the polygon level. So traditional geometry has always been generated from lowpoly meshes by means of tessellation.
With technologies such as PolyBump, they just moved back the other way... first make a detailed model, then reduce the geometry by encoding it into bumpmaps.
This approach could also work for DX11 by the way, using displacement maps rather than bumpmaps.

I think it's pretty safe to say that any games desgined from now on, will be based around adaptive tessellation.

 

[Arkadrel]

@Scali no it doesnt mean thats the only wat to do it, but its unrealistic to expect it done any other way until 2012 when consols start haveing tessellation. And when more people are useing Win7/dx11 gfx cards

 

[jvroig]

Scali has commented that we are likely to see a lot more tessellation in the future, so if that turns out to be the case, perhaps the effects of much higher tessellation levels (and the hardware to support it) will be far more obvious to the relevant gamer.

Properly done tessellation actually won't have to be obvious at all. It could simply result in the same quality (for example, the high/max-detail settings of the game), except that it will have better performance with tessellation active due to memory and bandwidth savings.

It will only be obvious (or have to be obvious) if the max-detail settings that the developers want in the game are too expensive memory-wise to be viable in anything other than tessellation. Due to the wide array of video card selection, this does not seem very plausible to me.

So when tessellation is done right (that is, all games started to be developed now, and perhaps even those this year or maybe even last), it doesn't have to be obvious by details/graphics, but it will be obvious by performance - turning tessellation on/off will result in "on" being better, except when the hardware chokes on it due to the tessellation unit(s) being the bottleneck. This is why having powerful tessellation units are a godsend to devs, it simplifies the decision-making up to this point. The fact that there is an additional variable ("but will it run ok on AMD's currently weaker tessellator in existing offerings") screws up the entire picture, and does not offer help to simplify the decision-making process of determining how much tessellation to support (which goes into either offering an additional "ultra-high detail setting" or simply increasing performance in high settings, etc).

With the news that AMD's soon-to-launch parts will fix their tessellation performance is definitely welcome news not just for games, but also for devs / game studios.

 

[jvroig]

once games start becomeing dx11 ONLY (no dx9 or 10 support) we ll see tessellation DONE RIGHT.

False premise. Tessellation isn't something that AMD and nVidia came up with just for us this year or last year. It is common practice in 3D modeling. All the hype around it now is just the fact that we (rather, our consumer video cards) are getting the capability for it in a meaningful way.

So it doesn't matter if there are DX9 PCs, or that consoles are the main target. These are irrelevant in leveraging the DX11 feature of tessellation. The way the models are constructed are more indicatve of whether tessellation can be easily leveraged or not, and this is something that is easily fixed, and all it needs is decent dev relations work. Doesn't matter if the game studio starts with consoles first, and then PC later. All that matters is that the art is done from low-poly, then tessellated into higher details (done during dev, not in-game), and the resulting higher-detailed models are packaged into and used for consoles and non-DX11 PC settings, while packaging the low-poly models for DX11-powered PCs.

It doesn't have to be an AMD vs nVidia thing, so the "fanboy" terms being thrown around here have no place and are just generally indicative of lack of understanding of the topic.

to me this all sums up to:
Nvidia jumped the gun with tessellation, their overshooting currently for something that wont be needed for another 2 years+ time.

They didn't jump the gun. There are now offering game studios / devs a way to either:
a.) increase detail significantly beyond what would have been the non-tessellation-mode highest detail
b.) increase performance while retaining the same quality (that is, not using tessellation to solely increase detail; stick with the same detail levels, thereby utilizing tessellation as a performance increasing factor)

It is exceedingly beneficial to both gamers AND devs, such that it is a no-brainer. The decision-making process up to this point of the devs/studios (how much support to give to tessellation) is only complicated slightly by AMD hardware not being in general parity with nVidia hardware (it would be ok if the delta is not so wide).

 

[jvroig]

windows XP, which doesnt support dx11 (mircosoft is doing that on purpose to drive sales on win7)

This is not tessellation, so this is rather off-topic, but I just feel I have to challenge this because nobody seemed interested enough to take you up on your statement.

Bolded part: absolutely not true. You say it as if all you need is download DX11 and XP can use it but MS decided to put a software check on it to exclude XP users. DX 10 and 11 are built on top of a new WDDM. So it isn't that Microsoft CHOSE not to release DX11 for XP. But since DX11 was built on top of the WDDM introduced in Vista, there is no way to support it in XP.

There are several reasons why backwards-compatibility has to be sacrificed at some point or another (the list generally includes: better performance, better features, better reliability, scalability and maintainability, fixing mistakes of the past, etc; whichever among these was MS' goal/s is beyond me to state, but it is probably a few of those), so dismissing it with a wave of your hand to be "drive Win7 sales" betrays your depth of understanding of the technical issues involved.

 

[jvroig]

2) Some people have old grafics cards.... tessellation done right, would mean alot of the older non 4xx or 5xxx+ series wouldnt be able to buy games or play them. This might be a issue for some game developers that want to sell more copies by allowing non- dx11 cards to be able to play the game at a decent experiance.

3) most games sold are to the consols, which dont have tessellation = game ports take alot of work to add tessellation and then its on a already overdone model so gains are small = tessellation NOT done the right way. This means it ll first be around 2012 when we see heavy tesellation games appear (if consol refresh go with a chip that has good tessellation).

I think when consoles support tessellation, we'll start seeing it done right. Until then, we'll see game models designed to look good without tessellation, and that will keep tessellation an IQ feature, not anything that will boost performance.

That's if they are designing the game for Dx11 and then adapting it for consoles. So far, seems more like the other way around. They are adding superficial Dx11 effects to games.

All these don't matter when you really think about it.

For one thing, games already have different model sets (generally, low-detail, medium-detail, high-detail). So having "low-detail model" to be used as the target for in-game tessellation isn't something that will break the bank for game devs. They are already doing it. It is not too much extra work that will never benefit their prized console market. Which also answers why it doesn't matter if they make games for consoles first before PC.

For example:
1.) Make low-detail model.
2.) Tessellate low-detail model to create high-detail ("complete") model.
3.) Package high-detail model into console version, release console game.
4.) Package high-detail and low-detail model into PC version. Use high-detail (that is, "complete" model) for non-DX11 settings. Use low-detail model for DX-11 enabled settings, and let on-the-fly tessellation take care of it to produce the high-detail in the game.

This is simplified, because I only want to show that supporting tessellation ("ohs noes, multiple models!!") isn't impossible or an unnecessary burden. In reality, even before DX11, every game already has multiple models anyway (low-detail, medium-detail, high-detail, etc) since video-card capabilities greatly vary (low-end to high-end / SLI / Crossfire).

It is a non-issue, in short. All it needs is a little dev relations elbow-grease to ensure industry-wide adoption now that the hardware has arrived in force. In fact, since DX11 and hardware supporting on-the-fly tessellation arrived, I would be surprised if most game studios don't already include it in their most current and future plans (that means, of course, not the games already being done, since they are not "plans").

 

[Zstream]

I'll respond to you, Scali, in a few days. I feel you are misguided on several points so we'll talk later.

 

[Scali]

@Scali no it doesnt mean thats the only wat to do it, but its unrealistic to expect it done any other way until 2012 when consols start haveing tessellation. And when more people are useing Win7/dx11 gfx cards

Read my posts more clearly:
I explained why there is no reason NOT to incorporate tessellation in your content pipeline, as there is no detrimental effect for non-tessellation hardware.
And jvroig has just done the same above.

 

[Scali]

This is not tessellation, so this is rather off-topic, but I just feel I have to challenge this because nobody seemed interested enough to take you up on your statement.

Yes, I'd like to add that tessellation IS supported on XP, under OpenGL.
You can run the Unigine Heaven benchmark with tessellation in OpenGL mode under XP.
I haven't heard of any upcoming OpenGL titles that use tessellation though, but who knows...

 

[Arkadrel]

Nvidia for how many years has been fighting ATI on this... then get a slightly better working tessellation unit and its not suddenly a great thing. It sounds like a case of "its only great when we re better than you". Whats odd is nvidia gets away with it and we have fanatical nvidia fanboys in threads like these over hypeing it likes its the greatest thing ever and ATI by haveing a slightly slower tessellation unit is a really bad choice for cards. And that their somehow holding technological advancements back... what about all those years when nvidia made fun/belittled tessellation? and made sure it wasnt inplemented in directx 9.

Until games need more than current gen cards offer in tessellation its all just a moot point, and both nvidia and amd will get better at tessellation with time in their newer cards.

 

[Scali]

Nvidia for how many years has been fighting ATI on this... then get a slightly better working tessellation unit and its not suddenly a great thing. It sounds like a case of "its only great when we re better than you". Whats odd is nvidia gets away with it and we have fanatical nvidia fanboys in threads like these over hypeing it likes its the greatest thing ever and ATI by haveing a slightly slower tessellation unit is a really bad choice for cards. And that their somehow holding technological advancements back... what about all those years when nvidia made fun/belittled tessellation? and made sure it wasnt inplemented in directx 9.

Until games need more than current gen cards offer in tessellation its all just a moot point, and both nvidia and amd will get better at tessellation with time in their newer cards.

Oh yea, let's pull the fanboy card again, and throw some FUD at it, rather than countering the arguments that have been presented.
Tessellation in fact *was* implemented in DX9, as it was inherited from DX8's RT/N-patch implementation.
I think you're giving nVidia too much credit if you blame the failure of this technology on them. Heck, I haven't seen them attacking the feature in public, they didn't really need to. It was not a very mature technology, and ATi didn't exactly do much to promote it. In fact, ATi disabled TruForm(RT-patches) in the drivers by default, on the 9000-series.

Perhaps the reason why tessellation is now a topic worth discussing again has more to do with the fact that it is now a better, more mature standard, supported by both OpenGL and DX11, and by both vendors.
Supporting the better implementation doesn't necessarily make one a fanboy. I don't think I can say the same for people using the fanboy-argument to defend the lesser implementation though.

 

[buckshot24]

Scali, why can't you wait to see how Cayman handles tessellation? You act like there aren't new cards coming out in a few weeks. You are jumping the gun big time and will look pretty bad if the tess performance is great on the new top end cards coming out.

 

[Scali]

Scali, why can't you wait to see how Cayman handles tessellation? You act like there aren't new cards coming out in a few weeks. You are jumping the gun big time and will look pretty bad if the tess performance is great on the new top end cards coming out.

What are you talking about?
I haven't said a thing about Cayman's performance (go ahead, search my posts and find a quote, I dare you). We'll just have to wait and see, won't we.
I'm just saying *why* we should want better tessellation performance, as it seems that a lot of people are trying to downplay the importance of this technology.

So why don't YOU stop with the personal attacks. It makes you look pretty bad, right away.

 

[buckshot24]

That is the point. You haven't mentioned Cayman's performance and complained about tess performance of midgrade cards.

This is a thread about Cayman and you keep harping on about how nVidia's tess is superior when the whole dynamic of the discussion could change in a few weeks.

Saying that you will look bad if something happens isn't a personal attack, it's fact. If the situation doesn't change in a few weeks then you can complain all you want.

 

[SickBeast]

IMO tessellation is irrelevant until we have a killer app for it. It's exactly like the whole PhysX argument awhile back. Without useful apps that show a profound difference in gameplay, features like this are nothing but smoke, mirrors, and marketing drivel for the time being.

I'm not saying that tessellation is a bad thing to have; I'm simply saying that it's by-and-large an irrelevant feature for most consumers. By the time its useful, you will need a new GPU to enjoy the game in all its glory. Kind of like how the G80 supported PhysX, but could never run PhysX games acceptably.

 

[Scali]

That is the point. You haven't mentioned Cayman's performance and complained about tess performance of midgrade cards.

How can I mention Cayman's performance when the card isn't officially released yet? Nobody knows its true performance yet.

Saying that you will look bad if something happens isn't a personal attack, it's fact. If the situation doesn't change in a few weeks then you can complain all you want.

You are making it a personal issue. I am just having a technical discussion about tessellation. This isn't about me, nor is it really about AMD vs nVidia (it seems that people are mostly making it console vs PC). It's just about why tessellation is (or will be) important.
I won't look bad, because I support good tessellation performance. If the Cayman delivers that, all the better.