QCT QuantaGrid S74G-2U
QCT's Grace Hopper machine, and its entry in the first generation of NVIDIA's MGX modular architecture, which QCT says it partnered with NVIDIA to introduce. A single GH200 Superchip pairs 72 Arm Neoverse cores with a Hopper GPU over NVLink-C2C, giving one coherent memory pool across CPU and GPU rather than a host and a device that copy between them. QCT lists up to 480 GB of embedded LPDDR5X alongside 96 GB of HBM3 GPU memory, four E1.S NVMe drives by default, and three full-height full-length dual-width PCIe 5.0 x16 slots in a 2U chassis 900 mm deep. The memory figure here counts the HBM only, since the LPDDR5X is the Grace side of the pair. Power is one-plus-one 2,000 W Titanium supplies, which is capacity rather than draw, and QCT publishes no throughput figure.
We will point you at suppliers who have it. Free, and no signup.
System Details
GPU Configuration
System Specifications
Powered by NVIDIA GH200
This system utilizes 1 × NVIDIA GH200 96GB HBM3 GPUs, each delivering exceptional performance for AI and HPC workloads.
Per GPU TDP
1000W
Per GPU Memory
96 GB
Process Node
4nm
Architecture
Hopper
Documentation & Resources
Official Datasheet
QCT QuantaGrid S74G-2U technical specifications
QCT QuantaGrid D75H-10U
QCT • Latest version
Typical Use Cases
The QCT QuantaGrid S74G-2U runs 1× GH200 GPUs over NVLink-C2C joins the Grace CPU and Hopper GPU into one coherent memory pool; three PCIe 5.0 x16 full-height full-length dual-width slots for expansion.