Nvidia teased the Hopper H100 GPU featuring 120GB of HBM2e or HBM3 memory, signaling major memory bandwidth advancement.

NVIDIA's high-end Hopper H100 GPU with 80GB HBM2e memory could soon be joined by a beefier variant featuring 120GB of HBM2e or HBM3 memory. NVIDIA produces several different SKUs of its Hopper H100 GPU in both SXM and PCIe models, with the SXM variant featuring higher-end HBM3 memory. The company's new Hopper H100 GPU currently features 80GB of HBM3 or HBM2e memory, though no 120GB variant has been released to date.

A new PCIe-based graphics card called the "NVIDIA H100 120GB PCIe" has surfaced, which appears to be the beefed-up NVIDIA Hopper H100 GPU with 120GB of HBM2e or HBM3 memory. For now, 80GB of memory is "all" you get, with 120GB variants in PCIe form launching in the coming months. The model reportedly has 120GB of HBM memory, which signals things to come for NVIDIA customers.

In its full implementation, the H100 GPU can feature 6 HBM3 or HBM2e stacks for up to 120GB of memory on a 6144-bit memory bus that delivers 3TB/sec+ memory bandwidth. The SXM5-based NVIDIA Hopper H100 GPU has 80GB HBM3 memory maximum through 5 HBM3 stacks across a 5120-bit memory bus. The full implementation specs include 60MB of L2 cache, 4th Gen NVLink, and PCIe 5.0 support.

A pre-production unit visible in testing featured an impressive array of GPU silicon, including an NVIDIA A100 80GB PCIe, 4 x NVIDIA GeForce RTX 3090 Ti graphics cards, an "NVIDIA GeForce RTX ADLCE Engineering Sample" Ada Lovelace GPU, and the H100 120GB PCIe card. The pre-production Ada Lovelace graphics card featured a lowered TDP, limited to 350W, compared to the final RTX 4090 specs which reach up to 450W default TDP.