Technical White Paper: Understanding Peak Floating-Point Performance Claims

Technical White Paper: Understanding Peak Floating-Point Performance Claims

Digital signal processors, graphics processing units (GPUs), and FPGAs serve as accelerators for the CPU, providing a performance boost. Given the variety of architectures available, designers need a uniform method to compare the performance of these different architectures. Typically, the benchmark used is peak giga floating-point operations per second (GFLOPs).

However, there are several challenges involved in calculating the peak floating-point capacity of an FPGA when implemented in programmable logic. So, just how reliable are the GFLOPs numbers claimed by FPGA vendors for such floating-point implementations?

Learn how Altera can reliably claim up to 1.5 TeraFLOPs of performance in Arria® 10 devices and 10 TeraFLOPs in Stratix® 10 devices using an industry-standard method, and compare this claim against the claims of another FPGA vendor before making your design choice.

Download this technical white paper to:

  • Compare the floating-point performance of digital signal processors, GPUs, and FPGAs
  • Learn about the challenges of determining the performance of traditional FPGA floating-point implementations
  • Examine the applicability of competitive floating-point performance claims

Related Resources

Webcast: Accelerating Design Development Time with Hard Floating-Point DSP Blocks in FPGAs

Watch this 15-minute webcast for:

 

  • An overview of current floating-point implementation challenges
  • An introduction to Altera’s hard floating-point digital signal processing
    (DSP) blocks
  • An overview of how you can achieve unprecedented DSP performance,