NVIDIA GF100 & Friends speculation

Two theories on that, "walk softly/carry a big stick" and "weaksauce":

"Walk softly/carry a big stick": This card isn't just good, it's all that and a bag of chips! No need to push the news, as soon as everyone finds out how good it is it will be the story.

"Weaksauce": The other end of the spectrum, the card is hot/slow/noisy. Anything they say now will either sound bad or be dis-proven soon, best to just wait and let the chips fall where they may.

Could also be that nVidia honestly doesn't know what Cayman can do yet so they're not wanting to put out any final clocks/performance numbers until they know what they're shooting for....that would be under the "Cayman is delayed" theory.

 

I think Nvidia is trying to have their cake and eat it. They want to use 580 as a distraction for the upcoming analyst days when they will talk about it to the investors, but they also want to bring it out again as a spoiler against the Cayman launch.

So I think you'll see a launch for the analysts talking about what a good investment product 580, how much money Nvdia will make at Christmas from the HPC market that Nvidia will magic up, etc. Then you'll get a spoiler launch against Cayman that talks about how "kewl!" and "rad!" it is, how it great it play games with ten times too much tessellation, bring out some stone giant benchmarks, etc

 

Geeks3D posted what they say is Nvidia's GF 580 demo, Endless City: 600 million tri/s, procedurally generated, 500k light sources, SSAO, dynamic lighting. It apparently requires CUDA, so no Radeons tested yet, only a 460 and a (2x faster) 480.

(I also posted this in the Barts tessellation thread. Seems relevant here, too.)

 

Oh, so that's where the Gargoyle statues are coming from!

 

Quite nice and impressive tech demo.

EDIT - To be honest, I think that they downgraded too much the non-tesselated parts in demo, to show the difference between non-tesselated and tesselated...

 

To be honest , this is the best tessellation demo I have ever seen , the city just looks amazing , especially in wire-frame .

Agreed.

 

I wonder if I could run it on an AMD gpu if I had an 8800 in as a physX card.....

 

Isn't PhysX disabled when there's an AMD GPU present? Dunno about CUDA…

 

Guess what makes PhysX "run"

 

I don't think the point of the demo was to show how awesome tessellation is vs today's un-tessellated models. It was to show how you can take an extremely simple base mesh and dramatically increase detail with tessellation and displacement mapping. Of course, games will never ship with meshes that simple until tessellation is as common as texture mapping on graphics hardware.

 

Yeah I know, but I'm not sure if CUDA is disabled entirely (which would explain why PhysX is too) of if it's just PhysX.

 

What an ugly fella, reminds me of 4200i days ... god I hate non-breaking phong shading. Can we have Oren-Nayar in 2010, please?

 

Probably not. Also, the sexier BRDFs, whilst intellectually interesting, don't necessarily have a good cost/benefit ratio in the context of games, IMHO.

 

It seems fairly neat, but needs something more, not sure what.

The lighthouse thing is way cooler IMO

 

In other news, Nvidia is again picking out websites where to seed the review samples. Apparently, TechPowerUp is not a website where review samples have gone.

Wizzard on TPU forums

http://forums.techpowerup.com/showpost.php?p=2079210&postcount=83

 

Yeah, PCWorld.fr doesn't have a card either. It seems supply is really tight.

 

Not sure what you find so amazing TBH, I think that looks just terrible. The colors are all dull and washed-out, those 500M light sources could just be one single flat ambient light, there's no shading or shadows worth mentioning... Simply put, terrible graphics through and through.

 

The pics of that demo give me the impression of a low poly count

 

But also no briefing or NDA either? Looks like Nvidia is cherry-picking sites to get good reviews, as tight supply would be no reason not to brief sites.

 

a) Paid reviews;
b) Paper launch, maybe only two or three chips are good enough for GTX580...