Possible huge ATI tessellation performance boost with 10.5 drivers

[jigger]

Inst why he isn't often wrong, down the fact he's right 99% of the time.


I personally think he's just lucky. He's like a journalistic version of Forest Gump.

 

[drunkenmaster]

lol, yes he guessed about bumpgate, because he just out of thin air decided that Nvidia's bump material would fail in the future

This has entirely nothing to do with Charlie and I notice the only people saying its probably Charlie(not Rroff) are complete Nvidia Zealots. He gets everything mostly right, and its rubbish that he hedges his bets, because he gets the information from the horses mouth.


I can ask Fatboy when the next gen Nvidia cards are due, he can only guess. if I asked a guy in the design team, I'd have the real answer.

The only time that he's ever been wrong( I should say, not 100% on the money) is when he's given a range of numbers to hide the number given by his source.

As AMD went into great detail to tell us, corporate espionage is a HUGE problem, leaks are a huge problem and they went to incredible lengths to hide the real details behind eyefinity, and the 4870 real specs.

People give people different info, Nvidia like anyone else can identify sources and leaks by telling people slightly different specs and seeing what turns up on the web.

Anyone link to the twitter feed, do we know what he's refering to as being full of misinformation exactly? Are we sure its the AMD driver info, and not some other random nonsense in some forum somewhere thats wild speculation alongside the driver info?

 

[Rroff]

The twitter comment links to a quote of the same quote as in the first post of this thread and another quote along similiar lines - but also has info about "CUDA emulation on ATI hardware".

When I say hedging his bets its not really the right way to describe it but I can't think of a better description. Its not uncommon for him to say stuff that on its own is still true even when it no longer applies to the original topic.

 

[rancid_elf]

Sorry to say it but at the moment I am more interested in stable drivers which cover all aspects of what they are meant to do. Inproving things like tessellation which is hardly used outside bench marking at the moment is low on my expectation list.

 

[Biffa]

Yes truform tessellates the mesh and then smooths it but its not the same as whats called "tessellation" nowadays - which is (silly ATI) rather badly named as its only a component of the whole process... I prefer what nVidia have done and referenced it as a polymorph engine. Gonna have to look up what the umbrella term for the whole process is really called.

Nvidia talk of the Polymorph engine doing two things. Tessellation and Displacement mapping. The Tessellation smooths the limited geometry of most game models by adding extra geometry. That produces the smooth look, the tessellated object is then displacement mapped to give the depth of detail.

 

[Rroff]

Yah but I still don't know what the term for the whole process is and I probably should.

You have tessellation(sub-division)->smoothing->displacement in the whole process thats become known as tessellation :S

 

[RavenXXX2]

From CatalystMaker Twitter

@raven1xx1 If i didnt say it, it probably aint true

https://twitter.com/CatalystMaker

 

[Rroff]

never mind... if the rumours about how they are implementing tessellation in NI are true you could be looking at an upto 8x tessellation performance increase... but don't quote me on that, its just putting 1+1 together and possibly making 3.

(This is based on some small snippets of information on the NI architecture and a basic understanding of how tessellation is implemented in hardware).

 

[Baboonanza]

Yah but I still don't know what the term for the whole process is and I probably should.

You have tessellation(sub-division)->smoothing->displacement in the whole process thats become known as tessellation :S

surely yiou mean 'tessellation(sub-division)->displacement->smoothing'? There wouldn't be any point in smoothing a flat sub-divided polygon and equally no point in smoothing it if you were just going to displace it afterwards.

 

[Rroff]

Your smoothing the mesh profile not an individual polygon, depends what settings your using anyhow - seems they've moved away from truform style smoothing with this generation.

 

[Mr Paul]

Lets call it tessemelation, just because

 

[metalmackey]

Why not just use a high detailed mesh in the first place? Tessellation slows performance more (probably) than a better quality mesh would.

 

[Duesix]

Because tessellation provide better realism