A few days after NVIDIA’s announcement of the first Arm servers based on the Grace Hopper platform, the Californian company’s project immediately begins to materialize, not only due to the partnerships with OEM partners, but also as regards the adoption of the NVIDIA technology within the scientific community and especially supercomputers. In this regard, in the past few hours NVIDIA has officially announced that the new US supercomputer VENADO will use the Grace Hopper platform for a system that is supposed to guarantee one computing power of over 10 ExaFLOPS in AI.
This machine, built on the basis of the supercomputer HPE Cray EX, will be a heterogeneous system based a mix of Grace Superchip and Grace Hopper Superchip knots; everything will come to life in the Los Alamos National Laboratory and should allow a clear leap forward also in scientific research. NVIDIA Grace Hopper solutions will not be a prerogative of the United States, even the major European centers will marry the new Arm architecture, among these we mention Alpsthe next system of the Swiss National Computing Center.
We remember that Grace Superchip features two Arm architecture based CPUs connected via the NVIDIA NVLink-C2C interconnect; let’s talk about a solution that includes a total of 144 cores paired with a memory subsystem capable of delivering 1 terabyte per second bandwidth. NVIDIA Grace Hopper instead provides the same CPU but flanked by a Hopper GPU that always interfaces with NVIDIA NVLink-C2C; these platforms obviously support the latest standards such as DDR5, PCI-E 5.0, NVIDIA ConnectX-7 and, last but not least, NVIDIA BlueField-3 DPUs for secure HPC and AI workloads.