Biggest cpu/gpu die ?

[Royaleagle]

I have two question and it would help me if you had some answers...

 

what is the biggest area for a cpu or gpu die (no xeon phil), is it the nehalem eight core processor (684mm sq) ?

 

what is the max bandwidth of 4 gddr5 x32 8Gb/s chip that ar in parrallel ? (64, 128, 256, 512 or 1024 Go/s)

 

thank you very much I appreciate !

[The_Denks]

2 minutes ago, Royaleagle said:

I have two question and it would help me if you had some answers...

 

what is the biggest area for a cpu or gpu die (no xeon phil), is it the nehalem eight core processor (684mm sq) ?

 

what is the max bandwidth of 4 gddr5 x32 8Gb/s chip that ar in parrallel ? (64, 128, 256, 512 or 1024 Go/s)

 

thank you very much I appreciate !

There is a dedicated thread for CPU's on this forum.

Please do not post these sorts of questions here.

[Royaleagle]

but it's also gpu  that's why I didn't knew what to do...

[Royaleagle]

2 minutes ago, UberGamerKing said:

There is a dedicated thread for CPU's on this forum.

Please do not post these sorts of questions here.

how do I move it ?

[mikat]

6 minutes ago, Royaleagle said:

how do I move it ?

report your own post and say that you want it moved to cpus  but i don't think there is a real need for it to be moved

oh wait i thought you put it in gpus, nvm, do report your post and ask for it to be moved

[Royaleagle]

1 minute ago, mikat said:

report your own post and say that you want it moved to cpus  but i don't think there is a real need for it to be moved

thanks, the moderator are king

[Coaxialgamer]

biggest cpu die seems to be that 8 core nehalem chip you mentioned ( frankly 684mm² is HUGE ). the biggest gpu die ever made is gp100 , used in the tesla p100 ( not yet out yet , but demo'ed by nvidia about a month or two ago ). The problem is the yield to die size ratio isn't proportional. Many factors affect yield , including process maturity etc , but generally , the yields decrease exponentially the bigger you make the chip .

The formula to calculate gddr5 bandwidth is [effective memory clock] * [bus width] /8 *2= [bandwith in MB/s ]. In this case , that would be 8000*128 ( because 4*32 = 128) /8*2 = 256GB/s.

 

 

[Royaleagle]

1 hour ago, Coaxialgamer said:

biggest cpu die seems to be that 8 core nehalem chip you mentioned ( frankly 684mm² is HUGE ). the biggest gpu die ever made is gp100 , used in the tesla p100 ( not yet out yet , but demo'ed by nvidia about a month or two ago ). The problem is the yield to die size ratio isn't proportional. Many factors affect yield , including process maturity etc , but generally , the yields decrease exponentially the bigger you make the chip .

The formula to calculate gddr5 bandwidth is [effective memory clock] * [bus width] /8 *2= [bandwith in MB/s ]. In this case , that would be 8000*128 ( because 4*32 = 128) /8*2 = 256GB/s.

 

 

thank you very much  it is for a contest

it is enormous yeah ^^ I learn new things every day !

[BrightCandle]

Depends on the process and its maturity. Typically 500mm^2 has been the usual limit most companies were willing to push to but the lithography can potentially allow larger chips given some designing around the bigger die. I didn't know they had got up to 684mm^2, you rarely see anything go above 500mm^2 for purely economic reasons.