Cirrascale Unveils Rackmount Multi-Device Peering Platform 

Cirrascale Corporation, a premier developer of build-to-order, open architecture blade-based and rackmount computing infrastructure, today announced the RM4600 Series of rackmount workstations and servers designed around its proprietary 80-lane Gen3 PCIe switch-enabled risers. The new rackmount series supports the recently released Intel Xeon processor E5-2600/1600 v3 product families and up to four PCIe Gen 3.0 devices enabling multi-device peering on a single PCIe root complex and making it a perfect solution for highly parallel applications and libraries like those used for deep learning, machine learning, and molecular dynamics such as Torch 7, Theano, and AMBER.

“We’re finding a large cross-section of customers that are beginning to discover a whole new set of technical computing challenges and want their highly parallel applications to perform better,” said David Driggers, CEO, Cirrascale Corporation. “Our new RM4600 rackmount products tackle these challenges by providing the ability to scale and peer multiple PCIe devices, maintain those devices on a single root complex for increased performance, and do it with core components that are setting new levels of reliability and performance such as the Intel Xeon processor E5-2600/1600 v3 product families.”

The Cirrascale RM4600 implements the new low power, high speed DDR4 memory technology providing increased bandwidth and power efficiency while enabling higher overall memory bandwidth with lower density DIMMs. Additionally, the new systems provide improved performance of various high performance computing, professional imaging and feature detection applications and include next-gen networking improvements, new virtualization and security features. Overall, the Cirrascale RM4600 is perfectly aligned for various scientific, analytics, and engineering applications worldwide, including those for deep learning, machine learning, and molecular dynamics.

“Cirrascale peer to peer technology, like that used in the new RM4600, allows us to unlock the true power of GPUs for scientific computation,” said Ross Walker, Associate Research Professor, San Diego Supercomputing Center. “This is the first platform of its kind and provides the unprecedented bandwidth needed to scale an individual molecular dynamics simulation with AMBER across all the GPUs and devices within a node. This previously unattainable performance will have impact on many fields.”