Single and dual socket servers exploit AMD’s EPCY processors
To expand data centres performance, single and dual socket servers which support the 4th generation AMD EPYC processors have been released by Supermicro. They are intended for cloud, artificial intelligence / machine learning (AI / ML), high performance computing (HPC), hyper-converged infrastructure (HCI) and enterprise applications.
They contain up to 192 cores with up to 12Tbyte DDR5-4800MHz 12-channel memory and up to 160 PCIe lanes.
According to Supermicro, the servers will transform how enterprises analyse massive amounts of data, perform complex simulations, and can lower the total cost of ownership (TCO). They are designed to support the fastest and highest-core AMD EPYC 9004 series processors, each with up to 96 cores and 3Tbyte of memory per CPU. Up to 160 PCIe 5.0 lanes are available in dual processor systems, with eight lanes of PCIe 3.0 or 4.0 for less demanding peripheral connections. All 4th Gen AMD EPYC processors include security features that help keep data secure, whether in use, in-flight, or in-store.
The configurable systems are designed for the highest TDP (thermal design power) CPUs that can be liquid cooled, increase performance per Watt, tailored for a customer’s specific requirements and have a reduced TCO. The GPU servers can be used for AI applications, support Nvidia H100 Tensor Core GPUs and are suitable for metaverse-type applications, said Supermicro.
The portfolio includes the GrandTwin which has multi-node architecture with front and rear I/O. It is designed for maximum density and purpose-built for single-processor performance per node on the latest AMD EPYC 9004 series processors, and contains 12 DIMM slots. The modular design is optimised for applications, with a front I/O option fully accessible from the cold aisle, which simplifies installations and services space-constrained environments.
The company’s CloudDC servers are single-processor systems optimised for I/O flexibility and with 12 DIMM slots for many cloud-focused applications. They offer serviceability with toolless brackets, hot swap drive trays and redundant power supplies for rapid deployment and ease of maintenance in data centres. According to Supermicro, they are designed for cost-effective service delivery in cloud computing, including web hosting, email services, public and private cloud computing and content-delivery networks.
Supermicro Hyper enterprise servers have dual processors and 12-channel 24 DIMMs optimised for supporting the highest TDPs, said the company, offering a range of computing, networking, storage and I/O expansion capabilities.
The company also offers 4U GPU servers which are 4U dual processor systems, supporting up to 10 full height, full length, double-width PCIe GPU cards, including the latest AMD Instinct MI200 Series and Nvidia H100 GPUs. The 4U GPU-optimised systems provide maximum acceleration, flexibility, and balance for AI, deep learning and HPC applications and contain 24 DIMMs, said Supermicro.
There are also 8U universal GPU servers for machine learning and equipped with eight NvidiaA HGX H100 accelerators and dual processors, with 24 DIMMs. They are configurable with dual zone cooling and end-to-end non-blocking bandwidth to the GPU provide high utilisation and low TCO and training cycles.
Supermicro servers with 4th Gen AMD EPYC 9004 series processors through its JumpStart programme.