One Stop Systems and AMD Collaborate on 1U Accelerator Expansion System Utilizing ATI Stream Technology

Intel’s long-time nemesis, CPU maker AMD (Advanced Micro Devices (News - Alert)), is also rapidly becoming known for its GPU Technology for Accelerated Computing. In particular, its ATI Stream technology is a set of advanced hardware and software technologies enabling the AMD Graphics Processor (News - Alert) Unit (GPU), working in concert with the system’s CPU, to accelerate many applications having nothing to do with graphics (so long as such non-graphic “parallel problems” can also be mapped onto the GPU architecture). Being able to plug-in GPU acceleration means that heavy computation can be offloaded from the CPU, thus allowing for better balanced platforms capable of running demanding computing tasks as fast as possible in a given-sized box. This availability of such a “shot in the arm” enables new applications on new architectures, along with the transition from fixed-function to programmable pipelines. Research and industry efforts are demonstrating various proof-of-concept projects under the name GPGPU.

 

AMD (News - Alert) is building a complete ATI Stream computing ecosystem in conjunction with major third party industry partners and academic institutions around the globe.

 

For example, over in Escondido, California, One Stop Systems (OSS), a major provider of embedded computing modules and systems, has announced that it has collaborated with AMD to apply its ATI Stream technology to produce the OSS Computational Accelerator Expansion System. The first joint collaboration of the two companies, the OSS Accelerator Expansion System is a new (and novel) product family that supports the greatest compute density in datacenters, allowing users to deploy up to 4 single-precision TFLOPS and 800 double-precision GFLOPS of compute performance in a 1U (1.75-inch)-high package.

 

The 1U accelerator expansion system employs either the AMD FireStream 9270 or 9250 graphics processor-based accelerator boards to provide a system tailor-made for high-end computational applications. The 1U 9270 system contains one or two AMD 9270 boards while the 1U 9250 system employs up to four AMD 9250 boards. The 1U accelerator expansion system is equipped with a PCI Express (PCIe) x16 Gen 2 host cable adapter which installs in a host computer's PCIe x16 expansion slot, delivering data transfers of up to an amazing 80 Gbps to the host system. A 1-meter standardized PCIe x16 cable is included. The GPU expansion system off-loads the host CPU engines and will be used by research laboratories, video imaging systems, and gaming applications.

 

You can install the 1U rack-mountable system up to 10 meters from its host system using the standard PCIe cable. Greater distances from the host system can be achieved using a fiber optic version that will be released by OSS later this year. Critical system functions such as fan rotation, power voltages, and temperature can be remotely monitored through Ethernet, making the solution easy to manage in large datacenter environments.

 

The 1U 9270-2 Accelerator Expansion System with two FireStream 9270's lists for USD$6,845 (P/N OSS-PCIe-1U-ACC-9270-2) and is available immediately. The 1U 9250-4 Accelerator Expansion System with four FireStream 9250's lists for USD$7,495 (P/N OSS-PCIe-1U-ACC-9250-4) and is available immediately. OEM quantities can be quoted on request. Call toll-free at 1-877-438-2724 to order or to receive more information.

 

AMD FireStream 9270 is a dual-slot, solution with 2GB GDDR5 memory and up to 1.2TFLOPS single-precision and 240GFLOPS double-precision floating point performance. The AMD FireStream 9250 is a single-slot, lower power solution with 1GB GDDR3 memory and up to 1 TFLOPS single precision and 200 GFLOPS double-precision floating point performance. Both boards run familiar 32 and 64 bit Linux and Microsoft (News - Alert) Windows environments. The C-like development environment connects via extensions to open source Brook+. Stream software supports multiple commodity GPUs per system. These commodity GPU components were originally designed for PC gamers, which means that there’s already a reliable GPU and memory roadmap available for developers.