Supermicro launches Plug-and-Play SuperCluster for NVIDIA Omniverse

“Supermicro has led the industry in developing GPU-optimised products, traditionally for 3D graphics and application acceleration, and now for AI,” said Charles Liang, president and CEO of Supermicro. “With the rise of AI, enterprises are seeking computing infrastructure that combines all these capabilities into a single package. Supermicro’s SuperCluster features fully interconnected 4U PCIe GPU NVIDIA-Certified Systems™ for NVIDIA Omniverse, with up to 256 NVIDIA L40S PCIe GPUs per scalable unit. The system helps deliver high performance across the Omniverse platform, including generative AI integrations. By developing this SuperCluster for Omniverse, we’re not just offering a product; we’re providing a gateway to the future of application development and innovation."

The SuperCluster for NVIDIA Omniverse broadens Supermicro’s offerings of application-optimised AI rack solutions. A wide range of professionals depend on compute-intensive 3D workflows, with use cases ranging from product design to industrial digital twins. Generative AI has augmented existing 3D workflows and is supercharging a new era of applications. SuperCluster for NVIDIA Omniverse helps simplify the deployment of scale-out infrastructure for the multi-workload needs of 3D and AI.

Supermicro NVIDIA OVX systems serve as the foundational building block of the cluster’s compute power. Each system node hosts up to 8 of the latest NVIDIA PCIe GPUs that deliver the combination of highest 3D graphics and virtual reality performance, plus exceptional generative AI performance via Tensor Cores and Transformer Engine support. Systems are powered by 4x 2700W Titanium Level PSUs, all within a high-airflow chassis, to ensure stability under high-utilisation scenarios. Up to four NVIDIA BlueField®-3 SuperNICs or four NVIDIA ConnectX®-7 NICs per system provide multiple ports of 400Gb/s networking with high scalability and security.

Supermicro’s 4U PCIe GPU systems are certified by NVIDIA for use with NVIDIA Omniverse, passing a rigorous validation process that tests for performance, reliability, scalability, and security. Organisations can maximise performance across the diverse range of workloads within the NVIDIA Omniverse development platform, including the world-building OpenUSD ecosystem and generative AI technologies through Omniverse Cloud APIs.

SuperCluster for NVIDIA Omniverse is a fully interconnected infrastructure solution that helps ensure developers, artists, engineers, and others can access the highest level of GPU computing at the time of need, with seamless access to virtual GPUs or bare-metal access to full system nodes. The 400Gb/s high-performance network fabric, provided by the NVIDIA Spectrum™-X Ethernet platform, allows enterprises developing custom large language models to tap into a combined pool of GPU memory across system nodes, essential for training large AI models.

Supermicro’s validated rack solutions range from 4 GPUs to a 256 GPU Scalable Unit, which can be further multiplied to fit enterprises of any size. Customers receive thoroughly validated plug-and-play racks, tested at the L12 level and ready for use on day one.

A Highly Customisable Solution, Sized from a Single Rack to an Enterprise Scale

A Supermicro SuperCluster for NVIDIA Omniverse can be deployed from a range of available sizes and options depending on the customer’s requirements. System nodes can be equipped with either 4 GPUs per system or 8 GPUs per system. Deployments can be sized from a single rack with 4 systems to a scalable unit with 32 systems in 5 racks. Large deployments can be further incremented via scalable units to build clusters of virtually any size.

SuperCluster for the NVIDIA Omniverse Scalable Unit contains:

32 Supermicro SYS-421GE-TNRT (Dual-Root) or SYS-421GE-TNRT3 (Direct-connect) PCIe GPU System nodes

256 or 128 NVIDIA L40S GPUs

3 Supermicro SYS-121H-TNR Hyper System control nodes

3 400G 64-port NVIDIA Spectrum™ SN5600 Ethernet compute fabric switches

2 400G 64-port NVIDIA Spectrum SN5600 Ethernet storage/control fabric switches

2 1G 48-port NVIDIA Spectrum SN2201 Ethernet management switch

NN NVIDIA BlueField-3 SuperNICs or NVIDIA ConnectX-7 NICs

5 Racks: 48U 750mmx1200mm

SuperCluster for NVIDIA Omniverse can be configured in deployment sizes as small as a single rack. A single-rack configuration contains:

4 Supermicro SYS-421GE-TNRT or SYS-421GE-TNRT3 PCIe GPU system nodes

16 or 8 NVIDIA L40S GPUs

2 Supermicro SYS-121H-TNR Hyper System control nodes

1 400G 64-port NVIDIA Spectrum SN5600 Ethernet compute fabric switches

1 400G 64-port NVIDIA Spectrum SN5600 Ethernet storage/control fabric switches

1 1G 48-port NVIDIA Spectrum SN2201 Ethernet management switch

NN NVIDIA BlueField-3 SuperNICs or NVIDIA ConnectX-7 NICs

1 Rack: 48U 750mmx1200mm