Fermi - will gamers get value for money?

[jolle]

Well the picture in that link of the GT300 shows no tesselation unit either and that comes straight from Nvidia,and here's a partial picture of the ATI cypress GPU.


http://images.anandtech.com/reviews/video/ATI/5870/GE.jpg


Bottom line is that if it had one,and with that particular feature being part of the DX11 standard,i don't think Nvidia would keep it a secret.
The image in that link comes from Beyond3d, atleast its copyrighted by them, and I think it contains some meassure of speculation as noone has talked about ROPS and such afaik.
And Nvidia havent shown anything about ROPS in their own images either, are we to assume they´re not going to have any?
Looking at Anand, they´re really just looking at the shader core there, and all the images from NV show just that.

Since everyone is stating this as fact, there must be something more then the abscence of confirmation that it has fixed function HW for it behind this...
Im not saying it isnt true, I just fail to see how people are so convinced, since it´d be pretty wierd to just take the performance loss of using the shaders to do tesselation instead of a presumably pretty transistor cheap fixed function part.

[shadow001]

The image in that link comes from Beyond3d, atleast its copyrighted by them, and I think it contains some meassure of speculation as noone has talked about ROPS and such afaik.
And Nvidia havent shown anything about ROPS in their own images either, are we to assume they´re not going to have any?
Looking at Anand, they´re really just looking at the shader core there, and all the images from NV show just that.

Since everyone is stating this as fact, there must be something more then the abscence of confirmation that it has fixed function HW for it behind this...
Im not saying it isnt true, I just fail to see how people are so convinced, since it´d be pretty wierd to just take the performance loss of using the shaders to do tesselation instead of a presumably pretty transistor cheap fixed function part.


Maybe because the chip is already packing 3 billion transistors,and even at 40nm,will still be quite big,not to mention that for the GP-GPU market,there likely is no need for a hardware based tesselator....Given the emphasis on GP-GPU first,and that Nvidia's CEO declared that potential market as the future for the company,and there's still no other information regarding gaming performance at all right now,and we're almost in november,i'm starting to believe that while it's gaming performance should be at least higher than the GT200 gpu,gaming isn't the focus anymore.

[jolle]

Yes Im just saying that it´s a strange assumption that there wont be one, if there is no official word or anything on the matter.

ATi even included their Xenon one on their last gen GPU and the chances of anyone using that were really slim, and noone did, as it wasnt part of the DX spec.
So it cant be that big a cost to include one, while it will likely be a big performance impact NOT to, when tesselation is used at any scale.. presumably.
And that performance impact can carry a big loss in terms of sales.

And its fairly sure that they arent throwing the gaming market out of the window TODAY, just cause they´re investing in the HPC segment.
GPGPU performance is still relevant to gaming with Physics, fluid simulations, Postprocessing, AI , etc.

[LydianKnight]

Well... ATI's Tesselator is in its fifth generation or fourth, don't remember well... it all began with something called TruForm or something like that... wheres looks like NVIDIA never got something like that implemented, or at least, haven't heard anything about it in the past.

About NVIDIA, it depends... Fermi shows NVIDIA is clearly going to GPGPU route, but that's not harmful for gaming, on the contrary, we get an even more programmable graphics cards, so Tim Sweeney's prediction will surely be true in the short-term future... as programmability increases with each hardware generation, APIs (and some technologies) will become less and less important... same goes for Intel and Larrabee (failure-to-deliver aside).

The real question here is... do we really need a hardware-based tesselator? Because we can't forget DirectX is Microsoft's way to force the gaming industry and (partially) the hardware industry to follow their vision about what computing should be, not necessarily what currently is or should be...

Of course, it would be great to have one, just to be able to play Aliens VS Predator and many other games which make use of the tessellation process, but seems like we're talking about a brand new feature, and it's not... OpenGL already had a tessellator since the dark ages... inside GLU with the GLUtesselator object.

What I mean with all of this is... there isn't just a single tesselation algorithm but many of them... implementing all in hardware would be costly, in this case, I think a highly optimized version of a tesselator in software running on the graphics hardware can be enough...

(Unless a game wants to exploit the tesselator for every piece of geometry everytime, in that case... don't know what to say...)

Anyway, for the people who has been able to test the Unigine demo with ATI hardware... does it run smooth enough for a real-time game?