Guido Appenzeller
@appenz
At @hotchipsorg 2026 and stacking dies/chiplets in 3D is the new norm.
Diamond Rapids (below) stacks cores on cache and fabric manager. Fujitsu stacks compute on SRAM. d-Matrix stacks DRAM on compute. HBM stacks 4/8/16 layers of RAM. And the GPU talks haven't even started yet.
Diamond Rapids (below) stacks cores on cache and fabric manager. Fujitsu stacks compute on SRAM. d-Matrix stacks DRAM on compute. HBM stacks 4/8/16 layers of RAM. And the GPU talks haven't even started yet.
0 3