GPU silicon, its sliced or made from scratch?

[chaosmassive]

this things really makes me curious,
lets see at older chip first GM104 and GM106
and current chip GP104 and GP106

both of them have numbers exactly half from flagship chip
so, Nvidia sliced bigger chip into half to create lower end, or they start from zero to make those ?

 

[cadaveca]

from scratch. It is much easier that way, and gives them a bit more leeway in figuring out how the silicon itself behaves. GPUs are highly modular, so it is only natural that smaller GPUs seem to be just part of a larger one.

 

[xorbe]

No not quite. The basic blocks are the same, but they re-assemble into a smaller design. The smaller gpu is not an all-new from scratch design separate from the big guy.

 

[RejZoR]

this things really makes me curious,
lets see at older chip first GM104 and GM106
and current chip GP104 and GP106

both of them have numbers exactly half from flagship chip
so, Nvidia sliced bigger chip into half to create lower end, or they start from zero to make those ?

Depends. You can make a big chip and turn all the bad ones (which are salvageable) into lower end models. This is usually used in the beginning. They often go the dedicated route later. As manufacturing of the bigger GPU gets more refined and they can't deliver high supply of lower end models anymore.

 

[cdawall]

Depends. You can make a big chip and turn all the bad ones (which are salvageable) into lower end models. This is usually used in the beginning. They often go the dedicated route later. As manufacturing of the bigger GPU gets more refined and they can't deliver high supply of lower end models anymore.

That is only typical with specific models. 1070/1080, 980/970 it is very atypical to see a really highend GPU cut down for mid or low end. The physical BGA area of the high end stuff is much larger.

 

[xorbe]

570 -> 560Ti448
Titan X -> 980Ti

Low end ... man, they'll just rebrand the old stuff 1/2 the time