Embedded Processing & DSP

FPGAs’ DSP Performance, Revealed

Jeff Bier, BDTI

FPGAs are increasingly being considered for use as processing engines in high-performance digital signal processing applications, such as wireless base stations.
In these applications, FPGAs frequently work side-by-side with DSP processors. Having more
choices is good, but it means that system designers need a clear picture of the signal
processing performance of FPGAs–both relative to each other, and relative to high-end DSP
processors. Unfortunately, the most commonly used performance figures are unreliable and
confusing.

For example, because digital signal processing applications often rely heavily on
multiply-accumulate (MAC) operations, DSP processor vendors and FPGA vendors sometimes use
peak MACs per second as a simple metric for comparing
digital signal processing performance. But MAC throughput is a lousy predictor of digital
signal processing performance, for FPGAs and DSPs alike. Here are a few reasons why.

The MAC performance numbers for FPGAs often assume the hard-wired digital signal
processing
elements are operating at their highest possible clock rate. In practice, typical FPGA
designs will operate at lower speeds. On the other hand, using hard-wired elements is not
the only way to implement MACs on FPGAs; additional MAC throughput can be achieved using
programmable logic resources and distributed arithmetic. Furthermore, not all signal
processing algorithms are MAC-intensive.
Viterbi decoding, for example, is a key DSP algorithm used in telecommunications
applications
that makes no use of MACs at all.

Another approach for assessing signal processing
performance is to use common DSP functions (like FIR filters). But this approach can have
drawbacks, too. One problem is that each vendor typically uses a different implementation
of these functions–perhaps using different data widths, a different algorithm, or
different implementation parameters (such as latency). This means that results from
different vendors are generally not comparable. In addition,
small kernel functions typically aren’t effective for FPGA benchmarking, because the way
you’d implement a function within a full FPGA application is often quite different from
the way you’d implement the function alone. (For processors, in contrast, these little
benchmarks
are usually pretty good at predicting overall DSP application performance.) In addition,
benchmarks implemented by processor or FPGA vendors often lack independent verification,
making it difficult for engineers to make confident comparisons between devices.

Several years ago BDTI recognized the increasingly
urgent need for independent, accurate, apples-to-apples performance comparisons among
FPGAs and processors targeting digital signal processing applications. (See sidebar: Who
is BDTI?) To address this need, BDTI developed a new application-oriented benchmark,
the BDTI Communications Benchmark (OFDM)™, that is based on an orthogonal frequency
division multiplexing (OFDM) receiver.

Recently BDTI used The BDTI Communications Benchmark (OFDM) to evaluate several new
high-performance FPGAs and DSP processors. The full set of benchmark results and analysis
are published in BDTI’s report, “FPGAs for DSP: Second Edition.” Figure 1 shows sample
normalized,
low-cost results for a Xilinx SX25 and a typical high-performance DSP processor.

As shown in this figure, BDTI’s benchmark results provide a dramatic demonstration of
the potential cost advantages of using FPGAs for high-performance DSP applications–the
SX25 is more than an order of magnitude more cost-effective than a typical
high-performance DSP processor on this benchmark.

Designers also need to understand how the choice of processing engine will affect their
development flow, implementation effort, and system design. For this reason, BDTI’s report
explores the qualitative factors that influence the decision of whether to use an FPGA, a
DSP, or both, and provides guidance on how to make an informed choice. The report
highlights key open questions that will affect the long-term success of FPGAs in high-end
DSP applications,
such as FPGA energy efficiency and the effectiveness of new high-level synthesis tools for
FPGAs.

Who is BDTI?

BDTI (www.BDTI.com) is the most respected source for signal processing benchmarks. BDTI has been benchmarking the signal processing performance of processors
for nearly 15 years, and in recent years has expanded its benchmarking activities to
include FPGAs, multi-core chips, and other technologies.