“The Aprius Computation Acceleration System delivers massive, high-bandwidth compute acceleration, using open standards for mainstream cluster environments, with the installation and management flexibility IT managers demand,” said Varun Nagaraj, CEO, Aprius Inc. “We are partnering with key manufacturing and channel partners to allow OEMs to customize and get this product to market quickly and cost-effectively.”
The system is the first to include a PCI Express (PCIe) 2.0 fiber optic interconnect, developed by Aprius, that allows the system to accelerate four host servers simultaneously, and be placed up to 50 meters away from the hosts with transparent support for all OS environments. This greatly increases the ability for placement in undedicated rack space, giving data center managers the flexibility to install the system where they want, when they want.
The 4U rack mount chassis is modular, and allows in-rack access to cards and hot swap of key components, for easier physical maintenance and minimized downtime. It can house up to eight PCIe compute accelerator cards, such as the AMD FireStream 9270, delivering up to 9.6 teraFLOPS of processing horsepower via four PCIe 2.0 x16 interfaces for a combined bandwidth of 320Gbps.
“Data center managers need superior performance-per-watt and performance-per-dollar. The AMD FireStream 9270 compute accelerator provides more than 1.2 teraFLOPS of performance, applying the massive acceleration capabilities of the GPU to enable supercomputing-class performance at a fraction of the typical cost and power requirements,” said Patricia Harrell, director of stream computing, AMD (Advanced Micro Devices, Inc.). “We’re thrilled that Aprius has chosen the AMD FireStream 9270 to power what stands to be a true game changing system. The Computational Acceleration System takes one giant step forward in bringing the benefits of GPU acceleration to mainstream environments and users.”
The CA8000 is the first in a family of generalized systems known as Cluster Services Nodes which provide various services, such as network IO, storage and compute resources, to nodes within a cluster. These services can be provisioned, allocated or shared, depending on the type of resource, via the Aprius high-performance (20, 40 and 80 Gbps), low latency PCIe optical interconnect and managed through standard network management interfaces.
Aprius and numerous other components and systems companies believe an optical standard for extending PCIe can enable many new and innovative uses, particularly in the interconnect of many servers in a cluster for high-performance and very low-latency communications between servers. Aprius is a member of the PCI-SIG standards body and plans to work within the PCI-SIG to set standards for PCIe over optics.
Aprius to show off 9.6 teraFLOPS system with AMD GPUs
Posted on Friday, November 14 2008 @ 0:20 CET by Thomas De Maesschalck