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

[Scali]

No, they aren't. These newly released cards are the new budget mid-range cards. The cutting-edge stuff is coming out in a month.

Cutting-edge means: supporting the latest features. Whether that's midrange or not, is not relevant.
As far as I know, Cayman does not support anything that Barts doesn't. It's just faster and more expensive.

 

[betasub]

Other than unpredictable leaks, I guess we are waiting for the 6800 feeding-frenzy to die down before AMD PR throw us a titbit.

 

[Skurge]

The difference is, those nVidia cards you speak of, are not DX11 cards. They were never marketed as such, and people were never promised any tessellation.
People buying these cards know (or should know) full well that they have not bought the latest technology, and will be missing out.

The story with 6800-series is a bit different. They are supposed to be cutting-edge right now, the latest DX11 cards, and AMD has been promoting DX11 and tessellation for a while now. So it would be a bit of a letdown if they turn out not to be able to run DX11 games very well. After all, that is what the people were promised by AMD.
Or the other way around, if developers bow down to AMD's wishes, then the nVidia users won't be able to enjoy their DX11 cards to the fullest either. But somehow I don't see that happening.

The 6800 is a great card for right now, its not forward looking like the 400series, that's why it does a great job in today's games. I don't mind that It runs DX11 games well, DX11 is not only tessellation. A card can have all the tessellation power in the world, if its not good at anything else then its of no use to anyone cause a DX11 game has more than just tessellation in it. I don't see you blasting nV and AMD for not having multithreaded drivers that CIV5 uses in addition to tessellation and other features like directcompute which the 6800 does fine with.

Similar to the situation with the R580 and G70, the X1900 had every DX9 feature that was there it had a forward looking architecture. One thing it could do was run HDR and AA at the same time. The 7900 couldn't, that didn't make the cards obsolete overnight. They were still great cards. Having a setting to have ether HDR or AA or both made sure everyone got the best experience on their hardware. So I don't know whats wrong with doing it the same way with tessellation.

 

[lavaheadache]

This argument is getting nowhere fast. Bottom line aside from synthetic benchmarks, CURRRENT games are not particularly faster on nv hardware due to superior tessalation power. This is fact. Now, whether or not future games will be the same, nobody knows. I'm sure we can all make guesses, as this is what is happening in here now. Honestly though, both solutions are lacking power for tomorrows games anyhow.

 

[Skurge]

As far as I know, Cayman does not support anything that Barts doesn't. It's just faster and more expensive.

Yes, as far as you know, and that isn't much. Also, you said it yourself, Barts isnt lacking anything, it just doesn't do it fast enough, so whats the problem if Cayman is faster than Barts in regard to tessellation

 

[Chiropteran]

Cutting-edge means: supporting the latest features. Whether that's midrange or not, is not relevant.
As far as I know, Cayman does not support anything that Barts doesn't. It's just faster and more expensive.

What are you complaining about if not speed? Cayman is "just faster". Isn't the entire basis of your argument against AMD video cards that nVidia is "just faster" at tessellation? Suddenly speed isn't relevant at all when it doesn't fit with your argument? Funny how that works.

 

[OCGuy]

Other than unpredictable leaks, I guess we are waiting for the 6800 feeding-frenzy to die down before AMD PR throw us a titbit.

I am still waiting to see numbers to confirm that there actually is a 6800 "feeding-frenzy."

 

[Kenmitch]

Yes, as far as you, and that isn't much. Also, you said it yourself, Barts isnt lacking anything, it just doesn't do it fast enough, so whats the problem if Cayman is faster than Barts in regard to tessellation

Caymen is allowed/suppost to be faster than Barts in tessellation.

The problem lies with the GTX 480 as for some unkown reason it's suppost to be the king of tessellation and to only be beaten by Nvidia's next offering....Still trying to figure out the logic behind this one!

Without knowing more of the inner workings and what was changed in Caymen it's kinda hard to base it's hypothetical tessellation performance at this time....But it's not out the realm of possiblility that it's tessellation performance could be about the same or even greater than that of the GTX 480's

 

[lavaheadache]

I am still waiting to see numbers to confirm that there actually is a 6800 "feeding-frenzy."

lol, I'm not budging from my 5870's till the 6900 feeding frenzy

 

[Scali]

So I don't know whats wrong with doing it the same way with tessellation.

I'm not the one who says there's something wrong with that. On the contrary, I have been promoting this option all week.
It's the AMD camp that doesn't want to settle for second-best, so they don't want an 'extreme' tessellation setting in games, which would only work properly on nVidia hardware.

 

[Scali]

What are you complaining about if not speed? Cayman is "just faster". Isn't the entire basis of your argument against AMD video cards that nVidia is "just faster" at tessellation? Suddenly speed isn't relevant at all when it doesn't fit with your argument? Funny how that works.

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.

 

[Skurge]

Can someone explain this graph to me?

Now AMD say that tesselation in barts is faster than Cypress up until a factor of 11 or something then after that they should be about equal.

Scali, you claim that HAWX2 uses more tessellation than either cypress or barts can handle. If that is true than they should be the same speed in that game with tess on right?

But the 6870 is much faster than the 5870 and 6850, the latter having a better tesselation unit, but shouldnt make a difference as they should both me at the same speed. Now as the res is increased the gap between the 6870 and the 5870 begins to become smaller. How does that make any sense If HAWX2 is bottlenecked by tessellation?

EDIT: They are also closing the gap to the 400 cards A LOT.

 

[Aristotelian]

Tesselation is a major component of DX11. So is GPGPU. Why is it AMDs DX11 chips arent good performers compared to the competition in 2 of 3 major selling points of DX11? What purpose does it serve to put DX11 bullet point on the chip if it isnt great at performing that role? Feels like a DX9\10 chip with a DX11 checkbox.

Have any benchmarks to back that up? DX11 performance is important in games, as far as I can tell.

 

[Scali]

Scali, you claim that HAWX2 uses more tessellation than either cypress or barts can handle. If that is true than they should be the same speed in that game with tess on right?

No, because it's adaptive.
The tessellation factor is not constant over the entire screen. 6800 mainly gets its gains in areas with low tessellation factors.

How does that make any sense If HAWX2 is bottlenecked by tessellation?

Sounds like you bought AMD's story about how these benchmarks don't use adaptive tessellation, and render 1-pixel triangles and all that.
HAWX2 doesn't, it's adaptive, and as such, only parts of the screen are bottlenecked.
This explains why the 6800 can still outperform the 5800-series, while not getting anywhere near the performance that nVidia gets. nVidia mainly gets its gains in the areas with higher tessellation factors.

If you were to do a synthetic test, where you ONLY use high tessellation factors, THEN you'd probably see little or no performance difference between 6800 and 5800.

 

[Arkadrel]

I dont think Tom Clancys H.A.W.X. 2 is a value measure of tessellation performance.
Neither are Nvidias own made tessellation benchmarks to test cards vs one another.

Why because their made to run the best possible on nvidia, and the worst possible on amd.


Uningine is a much more realistic representation of the tessellation power between cards.
Someone find some tessellation scores of the 68xx and 58xx and the 4xx series and lets see those instead.

 

[Scali]

I dont think Tom Clancys H.A.W.X. 2 is a value measure of tessellation performance.
Neither are Nvidias own made tessellation benchmarks to test cards vs one another.

Why because their made to run the best possible on nvidia, and the worst possible on amd.

Anandtech used the DX11 detail tessellation sample in their 6850/6870 review. This sample was written by AMD, before Fermi was even around.
The results speak for themselves:
http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/19

 

[Arkadrel]

from that Anandtech, link Scali posted:

Tessellation doesnt seem that bad to me on the 6870's.

Yes the 460s in tess at high levels beat the 6870s by about 20%... but in low levels the 6870 are 50% faster than the stock 460s.

Now if games use low levels of tessellation, haveing a amd card might actually mean faster tessellation. Unless game developers start useing insane amounts of tessellation.

 

[Scali]

Tessellation doesnt seem that bad to me on the 6870's.

As you can see, there's a much larger drop on AMD's cards from medium to high.
At medium settings, the 6870 can keep up with the 470, but at high settings, the 470 is nearly twice as fast(!), and the 460 also overtakes the 6870 by a margin.
Doesn't look bad? I think it does.
At high settings, AMD manages only about 1/3rd of the framerate of the medium settings.
nVidia manages about 2/3rds. Quite a difference.

And remember, this sample is made by AMD, so 'high' is not all that high really. I believe it went up to 11 only. This is still in the 'good' area of the 6800. Imagine what it'd look like when it'd go to 64.

 

[Genx87]

Have any benchmarks to back that up? DX11 performance is important in games, as far as I can tell.

Have you seen the tesselation and GPGPU benchmarks?

 

[Creig]

And you can't say for sure what will happen with the games that will be coming out soon. Let's say that people who buy a 6800-series card now, will be using it for about 2-3 years before they upgrade. That's quite some time, and quite a lot of games to be released.
We'll know for sure in 2-3 years, whether AMD or nVidia took the right approach with the current generations.

Exactly. So you can hardly make the claim that AMD's tessellator is lacking. Two or three years from now, we may look back on the 6800 launch and find that AMD was bang on and that it was Nvidia who ended up wasting die space on a super fast tessellator that wasn't actually needed.

 

[Skurge]

Sounds like you bought AMD's story about how these benchmarks don't use adaptive tessellation.

What story is that? All I heard about HAWX was that AMD said its performance could be improved with optimization. I said I would wait to see how that goes before passing judgment.

 

[buckshot24]

Oh My goodness, a $240 part doesn't do something as well as a $500 part? Really? Wait til cayman comes out then complain if the tessellation is still lacking compared to nvidia.

Seriously it's like complaining about your meal when you have just eaten a bread stick.

Also let's wait to see comparison screenshots when AMD get's their driver fix for hawx 2 done and the consumer can decide if the detail makes a meaningful difference between the two methods. I'm guessing it won't be noticeably different.

 

[Creig]

And remember, this sample is made by AMD, so 'high' is not all that high really. I believe it went up to 11 only. This is still in the 'good' area of the 6800. Imagine what it'd look like when it'd go to 64.

And is 64 likely to be showing up in games anytime soon? Wouldn't 64 be in the "overkill" range? At that setting, I'd imagine neither Nvidia nor AMD would put up very impressive numbers.

 

[Grooveriding]

Oh My goodness, a $240 part doesn't do something as well as a $500 part? Really? Wait til cayman comes out then complain if the tessellation is still lacking compared to nvidia.

Seriously it's like complaining about your meal when you have just eaten a bread stick.

Also let's wait to see comparison screenshots when AMD get's their driver fix for hawx 2 done and the consumer can decide if the detail makes a meaningful difference between the two methods. I'm guessing it won't be noticeably different.

Yeah agree. I mean really, I thought this thread was about a purported leak of 6970 performance. There are several other threads steered on the course of whining about AMD and tessellation, does this one need to go that way as well.

We get it. Wait for 6970 to actually come out and then if this leak is false, commence bitching about it, otherwise there are other threads to whine over the 6870 not performing as well as a GTX 460 in synthetic benchmarks.

 

[Scali]

And is 64 likely to be showing up in games anytime soon? Wouldn't 64 be in the "overkill" range? At that setting, I'd imagine neither Nvidia nor AMD would put up very impressive numbers.

'Overkill' depends entirely on what it is you're tessellating, and how.
nVidia's hardware actually does reasonably well at the 64x setting, you can try it with TessMark. It scales up beautifully through the entire tessellation factor range.