Here's what AMD didn't want us to see - HAWX 2 benchmark

[Scali]

It is was larger and still only competed with the leftovers (HD5830).

It also competed quite well with the 5850, and it wasn't exactly 'miles behind'. Die size, performance and power consumption well all quite close, especially with the 1 GB model. This, combined with very friendly pricing, made it very competitive.
Perhaps you need a refresher: http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king

Supply problems correct. The smoothness lies in a huge increase in marketshare and a launching full range from top to bottom in three months.

Huge increase in marketshare and supply problems at the same time?
Not really. The marketshare increase didn't happen at launchtime, but mostly many months later, when the launch problems were finally ironed out.

 

[tannat]

Tessellation factors have a range of 1..64
They improved here: 1 <------> 11
But not here: 12 <---------------------------------------------------------------------------------------> 64

And computers have much more room for RAM with 64 bit operating systems. Anyway motherboard companies forcefully limit the potential use to 16 or 32GB

What's your point?

Noone says that nVidias brute force Tesselation solution isn't more powerfull for larger chips.

If HD6970 shows up to have scalable tessellation. Will you attribute this to AMD copying nVidia?. You know AMD certainly does a lot of modeling to investigate where breakpoints are in efficiency relative to die size, power, overall performance level and typical workloads.

 

[Scali]

What's your point?

My point is that the sweet-spot of tessellation (better performance AND better IQ, is not in the 1-11 range).

Will you attribute this to AMD copying nVidia

Why don't we stick to technical discussions, rather than this sort of 'personal speculation'.
I hate it when people do that. As if they can't discuss in a normal, polite matter, while sticking to the subject, and only have to demonize their discussion-partner.

 

[tannat]

It also competed quite well with the 5850, and it wasn't exactly 'miles behind'. Die size, performance and power consumption well all quite close, especially with the 1 GB model. This, combined with very friendly pricing, made it very competitive.
Perhaps you need a refresher: http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king

Die size larger - check
Power- higher
Performance - lower
Income per sold chip - lower

But anyway you're comparing a 2010 card with a 2009 card. Ignoring the ones that overlap closely in launch date.

But I'm pretty sure you know this.

Huge increase in marketshare and supply problems at the same time?
Not really. The marketshare increase didn't happen at launchtime, but mostly many months later, when the launch problems were finally ironed out.

Launch was great. Supply was constrained for long time. Due to demand and problem with getting enough chips from TSMC.

But I'm pretty sure you know this.

 

[Scali]

Die size larger - check
Power- higher
Performance - lower
Income per sold chip - lower

It's about the 'miles behind' remark, which is totally inaccurate. The differences are marginal.

But anyway you're comparing a 2010 card with a 2009 card. Ignoring the ones that overlap closely in launch date.

Yea, whatever.

But I'm pretty sure you know this. For me it seems that your grasping for straws and are only trying to blow smoke.

Ah yes, more personal attacks.

 

[tannat]

My point is that the sweet-spot of tessellation (better performance AND better IQ, is not in the 1-11 range).

Where is the sweetspot if it is not in the 1-15 range then?

Why don't we stick to technical discussions, rather than this sort of 'personal speculation'.
I hate it when people do that. As if they can't discuss in a normal, polite matter, while sticking to the subject, and only have to demonize their discussion-partner.

Sure, my bad. I only thought I saw a trend in your postings
My bad again.

 

[Scali]

Where is the sweetspot if it is not in the 1-15 range then?

Roughly in the 1-25 range, at this point, 1-32 on high-end cards. The key being that it's a range. Not any single value or subset.

 

[tannat]

It's about the 'miles behind' remark, which is totally inaccurate. The differences are marginal.

But they multiply. Miles behind is not an exaggeration. The success in GTX460 lies in that Nvidia sold it extremely cheap in comparison to it's die size. Not by getting close to AMD technology.

Yea, whatever.

But no, GTX460 was till far behind RV870. It is miles behind Barts though.

Ah yes, more personal attacks.

Sorry again, I'm just trying to figure out what your point is. You know the stance that AMD is working against tesselation is quite extreme and obviusly unfounded in my eyes.

 

[tannat]

Roughly in the 1-25 range, at this point, 1-32 on high-end cards. The key being that it's a range. Not any single value or subset.

Interesting. How is the sweetspot range defined and what input the tesselation sweetspot is defined by? I've never seen it suggested to be this large before.

 

[Scali]

But they multiply. Miles behind is not an exaggeration. The success in GTX460 lies in that Nvidia sold it extremely cheap in comparison to it's die size. Not by getting close to AMD technology.

You're talking nonsense.
The die-sizes of GTX460 and 5850 are virtually identical.
And both are 'harvested' cores, delivering about the same performance.

Sorry again, I'm just trying to figure out what your point is. You know the stance that AMD is working against tesselation is quite extreme and obviusly unfounded in my eyes.

Well, then spend some time studying the technological background, and you'll see what I mean.

 

[Scali]

Interesting. How is the sweetspot range defined and what input the tesselation sweetspot is defined by? I've never seen it suggested to be this large before.

The sweet-spot is where you get the best performance and image quality gains.
If you've never seen it suggested to be that large before, you've probably only been listening to AMD's side (their DX11 sample is artificially limited to low tessellation factors, and even then you see performance degrading at higher settings. On my GTX460, performance is at roughly 1000 fps, regardless of tessellation settings).

 

[Arkadrel]

Tessellation factors have a range of 1..64
They improved here: 1 <------> 11
But not here: 12 <---------------------------------------------------------------------------------------> 64

I just wanted to say that 6870 has +50% to +100% better performance than a 5870 in factors 1-11.

BUT even at factors 11-22 its still visbly faster than the 5870s. Yes it gets slower and closer to performance like the 5870 the closer to 32 factor we get. However it still does 32 factor tessellation better than the 5870.


Also Uningine is probably the most "fair" tessellation benchmark thats out there, and in that the 6870 is beating the 460's in tessellation (from the benchmarks Ive seen). Which is the card it was supposed to be paired against (even though in games its sometimes closer to 470 performance).

 

[Scali]

(even though in games its sometimes closer to 470 performance).

I think if the card is close to the 470 in games, it should also be close to the 470 in Heaven... Else apparently there's some kind of sudden drop in performance on the 6870 when tessellation is used, which cuts into its value-for-money and that sort of metric.
If we look here:
http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/19

For the "50-100&#37;" tessellation increase, the margin with the 5870 is not terribly large.
Sadly there's no stock GTX460 involved, but the OC model is still well ahead of the 6870, and the 470/480 are completely out of reach.

 

[Arkadrel]

Die size larger - check
Power- higher
Performance - lower
Income per sold chip - lower

But anyway you're comparing a 2010 card with a 2009 card. Ignoring the ones that overlap closely in launch date.

"April 12, 2010: Official release date for most manufacturers, bar EVGA, which released its cards on April 7, 2010." wikipedia.

First sold Fermi was in 2010.... so we re compairing a 2010 card vs a 2010 card. JHH saying that he has people working on a card, doesnt count as a launch... the fermis are not a 2009 card.

 

[Arkadrel]

Interesting. How is the sweetspot range defined and what input the tesselation sweetspot is defined by? I've never seen it suggested to be this large before.

"sweet spot" is the point where performance loss vs noticable image quality improvements are shown. I dont believe people "would" use 64x tessellation on a amd card OR a nvidia card in any game. Simply because Image quality gain would be sooooooooo small beyound the 10 mark or so, vs the huge amount of resources it would take.


Can anyone show me a game that has 64x tessellation?

because if not isnt all of this just nVidia makeing a mountain out of a molehill?

 

[krumme]

The sweet-spot is where you get the best performance and image quality gains.
If you've never seen it suggested to be that large before, you've probably only been listening to AMD's side (their DX11 sample is artificially limited to low tessellation factors, and even then you see performance degrading at higher settings. On my GTX460, performance is at roughly 1000 fps, regardless of tessellation settings).

Except to overtesselated dragons and stuff like that, where should i look in real world games bm for proof that the range is to narrow for AMD solution? with a cost/benefit point of view?

The problem by the overfocusing on very specific themes, fx. on the tesselation, is that it forces in this case AMD to do something about even if there is no real benefit for the gamers. Who knows, perhaps we will se off die buffering for 6970, beating 480 in dragon shows. Its good for nothing. To be fair, AMD is hit by the bs tesselation marketing they did themselves before 5870 launch

I have the impression that the technical solution NV uses for tesselation is because this card was not intended for gaming in the first place. Is that impression completely unknow or wrong in your view?

 

[Scali]

Can anyone show me a game that has 64x tessellation?[/B]

because if not isnt all of this just nVidia makeing a mountain out of a molehill?

With adaptive tessellation it's perfectly possible to go up to 64x in some cases.
It's much like 16xAF. It may happen for some pixels, just not the majority. The average will be generally about 4xAF over the entire screen, I suppose.

 

[Scali]

Except to overtesselated dragons and stuff like that, where should i look in real world games bm for proof that the range is to narrow for AMD solution? with a cost/benefit point of view?

What is 'overtessellated' in your opinion, and why?
Unless we can agree on what 'overtessellated' means, we cannot answer such questions.

I have the impression that the technical solution NV uses for tesselation is because this card was not intended for gaming in the first place. Is that impression completely unknow or wrong in your view?

There is no other use for their tessellation hardware than graphics. It's specifically for gaming.

 

[Arkadrel]

I still think all this is makeing mountains out of molehills.

the 6850-6870 will have scaleable tessellation and be equal to the nvidia products (from rumors)... and its out in like 3 weeks.

and still no real games that can benefit from x64 factor tessellation... its WAY over the top for VERY LITTLE image quality improvements. I think it ll be years before we see that in games.

 

[Scali]

the 6850-6870 will have scaleable tessellation and be equal to the nvidia products (from rumors)... and its out in like 3 weeks.

We'll see if those rumours are true in 3 weeks then.

and still no real games that can benefit from x64 factor tessellation... its WAY over the top for VERY LITTLE image quality improvements. I think it ll be years before we see that in games.

I don't think you understand (even after all that's been explained about tessellation).
Tessellation factor says NOTHING about image quality, detail or triangle size.

 

[blastingcap]

I still think all this is makeing mountains out of molehills.

the 6850-6870 will have scaleable tessellation and be equal to the nvidia products (from rumors)... and its out in like 3 weeks.

and still no real games that can benefit from x64 factor tessellation... its WAY over the top for VERY LITTLE image quality improvements. I think it ll be years before we see that in games.

I think you mean the 6950/70, with scalable tessellator according to the leaked slide. Cypress/Barts is "good enough" but I wouldn't hang onto either of them for more than a year or two, as tessellation factors ramp up. Cayman may have a much longer useful lifespan if it has stronger tessellation.

 

[Arkadrel]

@blastingcap

yeah your right... typo.. ment to say the 6950/6970 will likely have scaleable tessellation equal too the counterparts from nvidia.

Yeah... I think the 69xx cards will have a longer lifespan too, the 68xx where just ment to be cheapo cards with good performance for their cost.

Its hard to say that 2x 6870s is a bad buy though... almost 480 SLI performance isnt really what Id call slow pc grafics.

 

[tannat]

You're talking nonsense.
The die-sizes of GTX460 and 5850 are virtually identical.
And both are 'harvested' cores, delivering about the same performance.

No, 15% larger s not identical and the harvested HD5850 is quite stronger than GtX460 1GB. HD5830 is closer but this a severely harvested core.

But I guess you know that. And you are still comparing 2009 tech with 2010 tech.

Well, then spend some time studying the technological background, and you'll see what I mean.

I have not seen any such indications that AMD is working against tesselation. This is not due to ignorance.

But thanks for sending me back to the school bench

 

[tannat]

"April 12, 2010: Official release date for most manufacturers, bar EVGA, which released its cards on April 7, 2010." wikipedia.

First sold Fermi was in 2010.... so we re compairing a 2010 card vs a 2010 card. JHH saying that he has people working on a card, doesnt count as a launch... the fermis are not a 2009 card.

I know, I try to compare with HD6xxx here but we are getting back to HD5xxx all the time for some reason. That is a 2009 card. 9/23 2009.

 

[Scali]

No, 15% larger s not identical and the harvested HD5850 is quite stronger than GtX460 1GB.

Okay.