- Hopper (4th Gen): Introduced 128-thread warp group asynchronous MMA (wgmma), accelerated large global→shared memory transfers with TMA (Tensor Memory Accelerator), enhanced SM collaboration with CGA (Thread Block Clusters).
A data center-focused GPU microarchitecture announced on March 23, 2022, and released in October that year. The product is built on NVIDIA's custom 4N node based on TSMC N5, and designed to support large-scale HPC and AI applications alongside the GPU.
NVIDIA Hopper Architecture Products