E4 Ships ARM HPC Servers with Cavium ThunderX

Print Friendly, PDF & Email

e4Today Italy’s E4 Computer Engineering announced public availability and deployment of Cavium ThunderX-based ARKA servers. Customers can place orders today for one-socket and two-socket ARKA platforms in multiple chassis configurations based on Cavium ThunderX ARMv8 workload-optimized CPUs.

The ARKA family of servers is powered by 64-bit ThunderX ARMv8 processors, consists of one-socketand two-socket systems in single and multi-node chassis, and includes support for NVIDIA Tesla GPUs. For compute-intensive workloads E4 offers the ARKA RK204, with four two-socket server nodes in a 2U chassis. GPU capability for floating point-intensive workloads is available in the one socket RK004 system, and the two-socket RK201 system, both with option for integrated NVIDIA Tesla Accelerated Computing Platform.

E4 began delivering pre-production ARKA ThunderX-based servers to customers earlier this year for testing on HPC, cloud and big data workloads. The servers are now publicly available, and E4 customers are deploying them to increase their computing performance while reducing TCO, including Barcelona Super Computer, Centro Swizzero di Calcolo Scientifico (CSCS) and ETH Zurich.

The focus in my group is on Energy-efficient computing systems and we have been following with interest the path of ARMv8-based solutions toward the green HPC market. Thanks to E4, we finally have the opportunity to access real-life machines. We are working now to deploy our advanced power and thermal management solutions on E4’s ARKA servers, and we look forward squeezing the best FLOPS/W on this new wave of energy-efficient HPC systems,” said Luca Benini, Professor of Digital Circuits and Systems ETH Zurich.

E4 Computer Engineering specializes in the production and architectural integration of complete and customized solutions, for environments ranging from high performance computing to design of turn-key datacenters. E4 HPC customers include leaders such as BSC, CSCS, CERN and ENEA, where they have delivered solutions for workloads ranging from processing data from CERN’s Large Hadron Collider to wave modeling in the Mediterranean Sea.

ThunderX SOCs are uniquely suited to drive performance for these systems, with 48 custom cores, high memory bandwidth and capacity, onboard workload accelerators and integrated network and storage capability. Focused on workloads such as high-energy physics and molecular dynamics, the ThunderX-based systems will leverage the high-performance, energy efficient Tesla GPU accelerators to deliver best-in-class performance with lower cost and power consumption than delivered by comparable systems.

Visit E4 booth# 189 at SC15.

See our complete coverage of SC15 * Sign up for our insideHPC Newsletter