An overview of HBM

[Tzupy]

Article at bit-tech, I think it's worth reading. A new generation of aftermarket gfx coolers will be needed for cards with HBM.
The GPU package becomes larger and the HBM is in it, so it doesn't need extra memory heatsinks (VRMs will still need).
http://www.bit-tech.net/hardware/memory ... mory-hbm/1

Edit: you can find the same article replicated on multiple tech sites.

[CA_Steve]

I just read the Anandtech version. It'll be a neat thing to see. I wonder how well they solved the thermal stress issues (temp cycling/3 layers of differing materials, etc)?

[boost]

As I understand is the layout is:

HBM                HBM
HBM    GPU    HBM
I  n  t  e  r  p  o  s  e  r
Graphics   card   PCB

If I understand it correctly, the HBM stacks, GPU and interposer are basically silicon chips of the same material and the PCB is some sort of laminate.
Shouldn't the top layers up to the interposer have the same coefficient of thermal expansion?
Is the interface between interposer and PCB any different from the the interface in current graphics card where it is between GPU and PCB directly?
Why would thermal stress be a bigger issue in this setup (other than maybe bigger surface area)?
[edit for nbsp]

[CA_Steve]

Thinking it over, I'm not that worried about the Si on Si contacts. The interesting bit will be due to the change from heatsink's direct contact with the gpu changing to an IHS. What will they use for the thermal interface between the chips and the IHS? Thermal paste? Solder? Intel moved to thermal paste as their die got too small for solder (shearing/microcracking issues). While GPU and RAM are fairly large die...what happens when you have many of them soldered to one IHS? ..and if you have to use thermal paste, that'll be a lot of heat to try and transfer compared to a CPU. <shrugs>