Oct. 15, a global provider of enterprise computing, storage, network solutions and green computing technologySuper Micro Computer, Inc. (SMCI) announces the launch of passenger transportationNVIDIAAmpere architecture GPU and the third generation of built-in AI acceleratorsIntelA new system of Xeon scalable processors (Supermicro X12 series).
These servers are designed for AI applications that require low latency and high application performance. The 2U NVIDIA HGX A100 4-GPU system is suitable for large-scale deployment of modern AI training clusters with high-speed CPU-GPU and GPU-GPU interconnections. The Supermicro 2U 2-Node system reduces energy consumption and costs and carbon emissions by sharing power supplies and cooling fans, while selecting a variety of GPU accelerators to match workloads. Both systems contain advanced security features implemented by the latest Intel Software Guard Extensions (Intel SGX).
Charles Liang, President and CEO of Supermicro, said, "Supermicro engineers have once again created a new product for our rich and diverse range of high-performance GPU servers, which can save cost, space and energy consumption compared to other designs on the market. With innovative design, we offer our customers a 2U high-density GPU system with four NVIDIA HGX A100 (code: Redstone) GPU accelerators for their AI and HPC (high performance computing) workloads. In addition, our 2U 2-node system is uniquely designed to share power supply and cooling components, thereby reducing operating costs and environmental impact. "
This 2U NVIDIA HGX A100 server uses a third-generation Intel Xeon extensible processor platform, supports Intel Deep Learning Boost Technology, and is optimized for analysis, training, and reasoning workloads. The system can carry 4 A100 GPU and is fully interconnected with NVIDIA NVLink technology, providing up to 2.5 petaflops AI performance. Its GPU memory up to 320GB accelerates breakthroughs in enterprise data science and AI. For complex conversational AI models such as BERT large-scale reasoning, this system is 4 times faster than the previous generation GPU calculus, and has a 3-fold performance improvement in BERT large-scale AI training.
In addition, these systems use advanced cooling and cooling designs, making them ideal for high-performance clusters that give priority to node density and power efficiency. These systems can also use liquid cooling systems to save more operating costs. The platform also supports Intel Optane persistent memory (PMem), and larger models can be stored in memory close to CPU before processing on the GPU. For applications that require multi-system interaction, the system can also be equipped with four NVIDIA ConnectX-6 200Gb/s-speed InfiniBand cards to support GPUDirect RDMA at 1:1 GPU-DPU.
The new 2U, 2-node is an energy-efficient and resource-saving architecture designed to support up to 3 double-width GPU per node. Each node is also equipped with a third-generation Intel Xeon scalable processor with up to 40 cores, built-in AI and HPC accelerators. A wide range of AI, rendering, and VDI applications can benefit from this balanced design of CPU and GPU. The system is equipped with Supermicro's Advanced Istroke O Module (AIOM) expansion slot with fast and flexible network connectivity to load large amounts of data traffic while running workloads and dealing with deep learning patterns, meeting high-demand AI/ML applications, deep learning training and reasoning. This system is also an ideal choice for multi-individual advanced cloud games and many other computing-intensive VDI applications. In addition, the virtual content delivery network (vCDN) can meet the growing demand for streaming media services. The backup power supply built into the system can be used by any node to use the power supply of adjacent nodes in the event of failure.