AI Architecture & Systems

GPU Generational Evolution: Compute, Data Supply, and Cooperation Scope

SourcesSurvey §5, Tables 3–4 and Figs. 11–15; NVIDIA and AMD corpus records. Preliminary specifications retain that status.

4.24.1NVIDIA: Pascal → Volta → Turing → Ampere → Hopper → Blackwell#

4.24.2NVIDIA: Cross-Generation Comparison of Tensor Core Datapaths and Control Granularity#

4.24.3NVIDIA: Rubin Disclosures in the Survey and Preliminary Specification Boundaries#

4.24.4AMD: MI50 → CDNA/MI100 → MI200 → MI300 → MI350/MI400#

4.24.5AMD: Cross-Generation Comparison of Matrix Compute, On-Chip Memory, and Interconnect#

4.24.6The Divergent Evolution of Scalar/Vector Throughput and Low-Precision Matrix Throughput#

4.24.7Co-Evolution of Arithmetic Capability, Memory Supply, Execution Coordination, and Deployment Form#