#Singapore #AI Platform #NVIDIA #SuperX

SuperX Launches a Revolutionary AI Supercomputing Platform

SuperX AI Technology Limited (NASDAQ: SUPX) has officially introduced its groundbreaking SuperX GB300 NVL72 system, steeped in innovation and filled with advanced technology. This latest platform integrates NVIDIA's cutting-edge GB300 Grace Blackwell Ultra Superchip, pushing the boundaries of what’s possible in rack-scale AI supercomputing. The introduction marks a crucial turning point for the data center industry, offering a new model for training and executing next-generation AI workloads.

Performance Beyond Limits

The SuperX GB300 NVL72 system is engineered to break the physical and computational limits that restrict current data center capacities. With an astonishing performance of up to 1.8 exaFLOPS, this liquid-cooled machinery offers unprecedented gains in both performance density and energy efficiency. The system’s design reflects the needs of modern workloads that conventional air-cooled infrastructures simply cannot meet.

By juxtaposing advanced 800 Voltage Direct Current (800VDC) solutions, SuperX is steering away from traditional alternatives, elevating them from mere benefits to essential features. This shift assures that the gargantuan power requirements for such high-performance systems are met with stability and safety in perspective. Instead of seeing the GB300 as an isolated product, it’s envisioned as a linchpin in a broader SuperX Prefabricated Modular AI Factory solution that includes necessary liquid cooling systems and robust power infrastructures.

The Incredible Grace Blackwell Ultra Superchip

At the heart of the SuperX GB300 System is the innovative blend of 72 NVIDIA Blackwell Ultra GPUs and 36 NVIDIA Grace CPUs, effectively forming a robust 2-to-1 configuration. This synergy develops an extreme bandwidth of 900GB/s, facilitating a smooth connection between the high-memory bandwidth of the Grace CPU and the performance power of the Blackwell Ultra GPUs.

Moreover, the combination ensures a unified memory architecture with a stunning total of 2,304GB of HBM3E paired with LPDDR5X memory from the Grace CPU. This monumental capacity allows handling of extraordinary model sizes and K/V caches without encountering I/O bottlenecks, thus promoting optimal efficiency. The unique power-efficient processing of the Grace CPU and the compute-intensive performance of Blackwell Ultra GPUs together provide significantly enhanced performance per watt.

Unmatched AI Performance Metrics

Designed for massive horizontal scaling, one SuperX GB300 System can expand into the NVL72 rack format, linking 72 Blackwell Ultra GPUs into an impressive single GPU installation. Its groundbreaking capabilities redefine industry benchmarks, achieving as much as 1.8 exaFLOPS of FP4 AI compute—all within a single rack.

The incorporation of advanced liquid cooling design enables continued operation at peak performance by ensuring exceptional density and maximum energy efficiency around the clock. Here are some of the pertinent technical specifications of the GB300 NVL72:

• - CPU Configuration (Per Rack): 36 NVIDIA Grace CPUs (total of 144 Arm Neoverse V2 cores)
• - GPU Configuration (Per Rack): 72 NVIDIA Blackwell Ultra GPUs
• - Total HBM3E Memory: Around 165TB
• - Total LPDDR5X Memory: Approximately 17TB
• - Peak AI Performance: Roughly 1.8 ExaFLOPS (FP4 AI)
• - Networking Features: 4 NVIDIA NVLink Connectors, 4 NVIDIA ConnectX 8 OSFP Ports (800Gb/s), 1 NVIDIA BlueField 3 DPU
• - Dimensions: 48U NVIDIA MGX Rack (2296mm x 600mm x 1200mm)

Target Market

The SuperX GB300 NVL72 system emerges as the quintessential platform for organizations aiming to develop the future of AI, whether that be in national AI infrastructure initiatives, public cloud services, or grand-scale enterprise AI factories. It’s particularly suited for those requiring exascale computing to effectively train and operate sophisticated multi-modal and huge language models. The system further stands to serve academic and governmental entities tackling significant challenges across various scientific disciplines such as physics, materials science, and climate modeling.

Moreover, industries like automotive, manufacturing, and energy benefit from the capacity to build high-fidelity digital twins, leveraging the combined processing power of both the Grace CPU and Blackwell Ultra GPUs.

Conclusion

SuperX AI Technology Limited is not only paving the way for advanced AI deployment but also setting the stage for infrastructural evolution in data engineering across power-hungry applications. By marrying high-performance with efficiency, SuperX offers vital solutions for a future where AI plays an integral role across various sectors. For inquiries related to product offerings or specifications, reaching out to the SuperX sales team is highly encouraged.