Home › Technologies › HBM3
HBM3
Technology
Memory
The H100/MI300X generation: ~0.8 TB/s per stack.
You're looking at a technology record: a building block of AI hardware — what it does, who makes it, and which chips and systems use it.
HBM3 roughly doubled HBM2E bandwidth (6.4 Gb/s per pin, ~819 GB/s per stack) and enabled the 80–192 GB accelerators of the ChatGPT era. SK hynix led qualification at NVIDIA, cementing its market lead.
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Used as memory in
Succeeded by
Successor to
Sources
Record reviewed as of 2026-01-15 — see Methodology.