At this time, the performance capabilities of the ERDC deployment would rank the system at no. 15 on the TOP500 ranking of the world's most powerful high-performance computing (HPC) platforms. The systems collectively represent 32 petaflops of performance and NVIDIA A100 GPU resources can be quickly added to or removed from compute systems for unprecedented flexibility and agility.
"The global challenges that public sector HPC organisations face require hardware resources to be much more muscular and efficient than traditional data centre architectures, which is why we are excited to work with NVIDIA to deliver to the DoD the first fully composable HPC system with the remarkable NVIDIA A100 GPU at its core", stated Sumit Puri, CEO & Cofounder, Liqid. "In collaboration with NVIDIA and other industry-leading partners, we hope to lead a wave of architectural innovation to provide a platform to solve some of the most difficult computing problems researchers face to build 21stcentury physical infrastructure, advance genomic research and development, and address the most pressing national security concerns."
The NVIDIA A100 delivers unprecedented acceleration at every scale for AI, data analytics, and HPC to address the world's most difficult computing challenges. The A100 can efficiently scale to thousands of GPUs or, with NVIDIA Multi-Instance GPU technology, be partitioned into up to seven isolated instances to accelerate workloads of all sizes. And with third-generation Tensor Core technology, the A100 offers up to 20x more performance than its predecessor, accelerating every precision for diverse workloads and speeding time to insight and time to market.
The NVIDIA HDR InfiniBand network provides extremely low latency, high data throughput and smart in-network computing acceleration engines. The HDR InfiniBand Quantum switches provide in-network computing through the Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) technology. This technology enables all active data center devices to accelerate the communications frameworks using embedded hardware, resulting in order of magnitude application performance improvements. As a standard technology, InfiniBand guaranteed backward and forward compatibility protects the data centre and software development investments.
Liqid's adaptive architecture enables the DoD to change A100 configurations as necessary and better accommodate the performance requirements of uneven, unpredictable workloads associated with graphics-intensive artificial intelligence (AI) applications. When deployed in the composable HPC system from Liqid across the HDR InfiniBand Quantum switching technology, the A100 delivers massive performance with all the data agility of composable disaggregated infrastructure, without sacrificing any data speed. Ultra-fast A100 GPUs and Liqid composable NVMe storage can be aggregated and deployed via software without regard to physical limitations, and shared across intelligent fabrics in the exact ratios required for a given workload, at massive scale or down to the level of individual MIG deployment. This means that data operations such as GPU-over-Fabric (GPU-oF) can be done with the same efficiency as those that take place up and down the hardware stack for industry-leading performance with the tightest possible physical footprint.
"The NVIDIA A100 GPU is the most advanced visual computing accelerator on the market, delivering the power and flexibility required by today's most complex AI models", stated Paresh Kharya, Senior Director of Product Management and Marketing at NVIDIA. "NVIDIA Partner Network members like Liqid are using the performance and functionality of the A100 GPU and HDR InfiniBand to build solutions that meet the needs of AI and HPC workloads."