AMD today announced the FireStream 9270 workstation graphics card, this card seems to be based on the Radeon HD 4870 and will be available in late Q4 2008 for $1,499. The AMD FireStream 9270 features 800 stream cores, 2GB GDDR5 memory, 108.8GB/s memory bandwidth, 256-bit memory interface, 1.2 teraFLOPS single and 240 gigaFLOPS double computing power, typical power consumption of 160W and peak power consumption of under 220W.
Today’s announcements also advance the company’s corporate brand and strategy, called Fusion, by further enabling AMD graphics processors (GPUs) to work in concert with CPUs to deliver balanced platforms capable of increasing performance, energy-efficiency, and cost-effectiveness.
“The demands of the high-performance datacenter are intense, with dramatic increases in problem complexity and size every year. Customers need to achieve ever higher performance-per-watt and performance-per-dollar in order to address these problems with the same infrastructures. The AMD FireStream 9270 compute accelerator was designed to address all of these concerns and more,” said Rick Bergman, senior vice president and general manager, Graphics Products Group, AMD. “Delivering two times the double-precision floating point performance of competing offerings, and armed with 2GB of ultra-fast, ultra-high bandwidth memory, the AMD FireStream 9270 is the ideal solution for dramatically accelerating technical applications.”1
The AMD FireStream 9270 delivers supercomputing-class performance, deployable in a wide variety of server and workstation configurations. It has a typical board power requirement of only 160 watts, yet provides over 1.2 teraFLOPS of single-point precision performance — the most available in a single-GPU solution today — and over 240 gigaFLOPS of double-point precision performance.1 It includes two gigabytes of Graphics Double Data Rate, version 5 memory (GDDR5) for greater overall data throughput and processing of larger data-sets. It comes with a three-year limited warranty.
There was also a bit more information about the firm's GPGPU initiatives in the press release:
System and Software Developers
AMD is working with leading system and software developers to help ensure broad availability of GPU accelerated end-solutions for a wide range of uses.
One such company is Aprius, which develops high bandwidth server interconnect systems that bring new levels of scaling, resource sharing and low-latency performance to the data center.
With help from AMD, Aprius plans to bring to market a solution for connecting up to eight AMD FireStream 9270 computer accelerators within a rack mounted chassis, hosted in a server cluster via multiple 80 Gbps PCI Express (PCIe) optical links. This results in 9.6 teraFLOPS of processing horsepower and 16GB of high speed memory over up to four PCIe buses. Dubbed the Aprius Computational Acceleration System, it provides new innovations in native PCIe interconnect over optical cables that allows PCIe 2.0 connections up to 50m in length, easy in-rack installation and maintenance, and transparent support for all OS environments. The technology is expected to be demonstrated at Supercomputing 2008 in Austin, Texas next week.
AMD is also working closely with long-time partner HP on the HP Accelerator Program, to ensure AMD FireStream compute accelerators are validated for use in HP ProLiant servers.
“Increased competitive pressure is driving companies to adopt high-bandwidth, compute-intensive solutions to facilitate better business decision making,” said Ed Turkel, product marketing manager for the Scalable Computing and Infrastructure organization at HP. “The combination of AMD FireStream compute accelerators with HP ProLiant servers enables customers to accelerate application performance, ultimately enabling faster development and time to market of products.”
AMD is also working closely with Brown Deer Technology, a company that provides consulting and software development services to clients solving complex and computationally challenging problems in high-performance computing.
“Using the ATI Stream SDK and Brook+ compiler, we’ve obtained GPU accelerated benchmarks on a range of algorithms, from electromagnetic and seismic wave solvers to particle-based calculations. These are important algorithms for modeling and simulation that impact wireless communication, oil and gas exploration, and physics and chemistry research,” said David Richie, president of Brown Deer Technology. “We’ve seen tremendous speed up, in some cases more than a 120x.2 The results have generated a lot of interest from customers.”
ATI Stream SDK v1.3
Version 1.3 of the free ATI Stream SDK includes extensive enhancements to help ease development of ATI Stream-enabled applications capable of running on a broader range of AMD graphics hardware, including AMD FireStream™ 9250 and AMD FireStream™ 9270 compute accelerators, ATI FirePro™ V5700 and ATI FirePro™ V8700 workstation graphics accelerators, and ATI Radeon™ HD 4000 series graphics cards.
It is expected to be released in conjunction with the ATI Catalyst™ v8.12 software driver, which is planned to include ATI Stream software enablement allowing millions of users of ATI Radeon graphics cards to run ATI Stream-enabled applications. Version 1.3 of the ATI Stream SDK also includes significant improvements to the Brook+ runtime and kernel language, allowing for increased stability, flexibility and performance when compared to prior version of the SDK.
As previously announced, AMD is working with industry partners to develop the OpenCL programming standard. The continued evolution of the Stream SDK is expected to guide developers using Brook+ code paths to OpenCL upon its release, ensuring that work done today can be leveraged in future software.