Download PDF >

FPGAs in High Performance Computing
Hardware accelerators such as FPGAs give technical computing users significant computation, I/O and memory bandwidth advantages over traditional CPU-only solutions. FPGAs have recently emerged as an attractive alternative to power-hungry CPUs in the field of High Performance Computing. Parallelism is the key to acceleration. Even though an FPGA operates in the 100's of Megahertz range it can outperform a multi-Gigahertz CPU via parallelism.

One significant barrier to widespread adoption of FPGAs in HPC is the hardware-oriented design flow environment. This development environment is unfamiliar and ill-suited to this market. A higher level of abstraction, via a software-oriented design flow, is much more desirable.

Access FPGAs Through Star-P Connect
Star-P enables this high-level abstraction by coupling familiar desktop tools such as MATLAB® with a high-performance back-end execution engine. FPGA coprocessors are a natural fit in this back-end engine. Through the Star-P Connect library API link, compute-intensive algorithms embedded in hardware appear as standard library functions, and can be easily called from the high-level desktop application. XtremeData in collaboration with Star-P will jointly develop a complete and easy-to-use environment for FPGAs in HPC.

XD1000™ FPGA Co-Processor Module
The XD1000™ allows the user to integrate Altera's leading edge Stratix™ II Field Programmable Gate Array (FPGA) technology into a multi-Opteron™ based system. The XD1000™ can be inserted directly into an Opteron™ 940 socket and uses the motherboard's existing CPU infrastructure to create a full featured environment for FPGA coprocessor functions. The module connects to the CPU's HyperTransport bus and motherboard DIMMs while utilizing the existing power supply and heat sink solution for the CPU. The XD1000™ provides a cost effective platform for FPGA acceleration that is deployable in the densest blade systems.

• Accelerate custom algorithms with FPGAs
• Integrate into multi-Opteron servers and clusters, including high-density blade systems
• Execute task-parallel and data parallel algorithms
• Familiar MATLAB® interface

Broad Algorithm Coverage
Visual Numerics allows MATLAB® and Star-P customers the ability to tap into IMSL's comprehensive library of algorithms. The IMSL Libraries offer broad platform support and native implementations in C/C++/C#, Java, and Fortran.