4850 Terascale graphics engine?

[sdf]

Anyone heard or know what the terascale graphics engine is or does? I did a search on it and keep coming up with intels teraflops results. Does this even have any benefits of any kind or just a marketing gimic? Not a bad price for a 4850 though. Here's the card.

http://www.newegg.com/Product/Product.aspx?Item=N82E16814102770

 

[cjl]

Marketing.

 

[one-shot]

It is some fantastic marketing brought to you by AMD. They are getting their act together in the Graphics dept but they need some help in the cpu sector. My 4850 also has the "Tera-Scale bla bla blah". It's processing power is high but it doesn't reflect like that in games.

 

[JAYDEEJOHN]

When I read the thread topic, I thought you meant this http://www.techpowerup.com/69789/GECUBE_Unleashes_HD4850_1.0GB_Overlocking_Series.html At first glance, it doesnt look like much, but look at the shader clocks

 

[blashyrkh]

How is that possible? The shaders on ATI cards operate on the same frequency as the GPU. If you OC the GPU then you OC the shaders. How did Gecube manage to up the Shader freq while Ati couldn't do it since the 2900 cards??? I need GPU-Z shot to believe it!!

 

[TheGreatGrapeApe]

It is some fantastic marketing brought to you by AMD.

And intel before them;
http://techresearch.intel.com/articles/Tera-Scale/1421.htm

There've been talking about Terascale since about 2005, with presentations in 2006 about the 80-core processors, the talk that launched the Larrabee/Nehalem stuff.
It seems to me that it's a semi-generic term about Tera-flop computing.

 

[TheGreatGrapeApe]

How is that possible? The shaders on ATI cards operate on the same frequency as the GPU. If you OC the GPU then you OC the shaders.

They aren't specific about whether or not they OC the other components, it might just be talking about overclocking the shader/gpu versus both it and memory, or it may be specific to the computational shader component. It's nothing new or excluvie for different parts to have different clock domains, ther are many within the chip that do it. ATi has made no secret that they are working on different clocks domains for shaders and such, similar to nV, but they chose not to implement it this time for whatever reason (they weren't clear when mentioning that they were not going to do it for the HD4K at launch, but they also said it might come in the future [they also wren't specific about for the same chip]).

How did Gecube manage to up the Shader freq while Ati couldn't do it since the 2900 cards???

It is unlikely it would be GeCube alone that would do this, they would need help from ATi, but there is no reason they couldn't if they have the tools provided by ATi. If ATi simply doesn't do it from the plant, but leaves the option open, then it's something that would be designed by ATi, but enabled by the OEMs, this happens to other 'unofficial' features. But that it was not widespread to me means there is a limitation that affect rendering/3D apps, but has no negative impact in the less complex and interdependant functions of compute/GPGPU apps. If a compoentn like the texture units or ROPs don't react well to asynchronous clocks, then that would make it problematic for 3D, but do nothing for plain math calculations.
Also think in the context the disabling of parts of the GPU, this has been discussed as well by both companies. If ATi is unable to effectively get different clock domains to work for the shaders versus the rest of the GPU, but they are able to turn off those other unneeded parts, then it would be something that would once again have no application to graphics, but be great for computational work, since you don't waste energy keeping useless parts idle even. That's definitely the direction that AMD want to go, but even their new FireStream 9250 makes no mention of separate domains or disabling other parts of the VPU.