Skip to knowledge
VLSI Hub / Knowledge

Knowledge / memory technology

HBM2

Canonical entity with source-backed relationships and retained history.

urn:vlsi:entity:28f91da6-231c-46ef-8fbd-4d1cb14cac4b

Clear dates

A source-reported date is a historical snapshot. Unknown validity is excluded when a date is selected. “Known to VLSI” uses the end of that UTC day. This entity’s heading uses its current label.

Recorded relationships

SUPPORTED1 evidence record · Revision 1

NVIDIA A100 Tensor Core GPUUses memory technologyHBM2

This is the cited 40 GB configuration; later A100 variants use HBM2e.

Source-reported snapshot: 2020-05-14

Inspect evidence, confidence and history

Evidence index 85/100 · A rule-based index, not the probability this claim is true. 1 independent source origin(s).

Dated source-reported snapshot. Publication date is not an inferred manufacturing start, contract start or legal-close date.

NVIDIA Ampere Architecture In-Depth

NVIDIA Technical Blog · Tier A · Primary source · supports

Evidence summary: NVIDIA specifies 40 GB of HBM2 for the A100 discussed in the architecture article.

Location: lines 82-84 / 40 GB HBM2 and 40 MB L2 cache

Published: 2020-05-14 · Retrieved: 2026-09-08 · Extraction: manual-source-review

Store factual paraphrase, metadata and link only. No source document redistribution or inferred license; no source bytes retained.

First observed by VLSI: 2026-09-08T15:57:21.457Z. No independent human verification recorded.

Review: Evidence ingested by curated-primary-sources/1.0.0. This is the cited 40 GB configuration; later A100 variants use HBM2e.

Full claim history →