NVIDIA Hopper Architecture In-Depth
NVIDIA specifies H100 SXM5 with 80 GB HBM3 across five HBM3 stacks.
- Published
- 2022-03-22
- Retrieved
- 2026-09-08
4 entities and 3 documented connections shown.
Drag the map or use pan buttons to move. Tab to a node or connection and press Enter to inspect its evidence. Arrows follow the actual claim direction.
| Subject | Relationship | Object | Status | Evidence |
|---|---|---|---|---|
| NVIDIA H100 SXM | HBM3 | SUPPORTEDSource-reported snapshot: 2022-03-22 | Why are these connected?Variant-specific: NVIDIA also describes PCIe H100 configurations with HBM2e. NVIDIA Hopper Architecture In-DepthNVIDIA specifies H100 SXM5 with 80 GB HBM3 across five HBM3 stacks.
| |
| AMD Instinct MI300X | HBM3 | SUPPORTEDSource-reported snapshot: 2023-12-06 | Why are these connected?No memory-chip vendor is asserted. AMD Delivers Leadership Portfolio of Data Center AI Solutions with AMD Instinct MI300 SeriesAMD specifies MI300X with 192 GB of HBM3 memory.
| |
| AMD Instinct MI300A | HBM3 | SUPPORTEDSource-reported snapshot: 2023-12-06 | Why are these connected?No memory-chip vendor is asserted. AMD Delivers Leadership Portfolio of Data Center AI Solutions with AMD Instinct MI300 SeriesAMD specifies MI300A with 128 GB of HBM3 memory.
|
From NVIDIA H100 SXM · up to two relationship steps.
A path retains its original claims. Shared technology does not establish a supply contract.
Relationships may describe different dates. A path does not establish simultaneous activity. Bounded source-backed collection. Missing or filtered connections do not establish absence. Supported does not mean independently verified.