What happened to HBM Memory?

[Modred189]

When Fury was released, HBM was supposed to be the next gen of VRAM. There were even rumors Nvidia was looking at it.
What happened to it? Everything seems to be GDDR6/6x.

 

[Snowdog]

When Fury was released, HBM was supposed to be the next gen of VRAM. There were even rumors Nvidia was looking at it.
What happened to it? Everything seems to be GDDR6/6x.

It's more expensive so it gets used where needed. HBM was in a recent Mobile Navi part for Macbook Pro.

I am thinking it will be needed for Big Navi to be competitive.

 

[MangoSeed]

HBM is alive and well. It's just too expensive for most consumer graphics cards. GDDR6X is Nvidia's way of delaying the inevitable. HBM is still the future.

 

[melk]

HBM is utilized in the Datacenter and HPC space.

https://www.nvidia.com/en-us/data-center/a100/

For example.

 

[Logan M]

As everyone said HBM is still very much a thing it’s just prohibitively expensive.

 

[LazyGamer]

When Fury was released, HBM was supposed to be the next gen of VRAM. There were even rumors Nvidia was looking at it.
What happened to it? Everything seems to be GDDR6/6x.

The bigger the die, the bigger the HBM interposer, the lower the yields for the combined part... the higher the costs per working part and the fewer working parts available for sale.

 

[jeremyshaw]

The bigger the die, the bigger the HBM interposer, the lower the yields for the combined part... the higher the costs per working part and the fewer working parts available for sale.

Well, Intel has EMIB, but only for their own products (mostly FPGAs and an orphaned AMD GPU), Lets see if TSMCs... whatever it's called (COWOS-L?) can deliver on EMIB's promises (basically, no full sized silicon interposer needed, only a tiny strip to bridge the HBM stack to the memory controller / die).

 

[spine]

When Fury was released, HBM was supposed to be the next gen of VRAM. There were even rumors Nvidia was looking at it.
What happened to it? Everything seems to be GDDR6/6x.

My belief, so back when 'Poor Volta' AMD PR was hitting strong, Nvidia just dumped a big wad of cash to Micron to make GDDR faster.

+ also secured HMB2, then played it low.