The industry’s first and only independent benchmarks for digital signal processing
Overview
The BDTI DSP Kernel Benchmarks are the most widely used digital signal processing benchmarks in the world. Nearly every major vendor or buyer of processors for DSP uses this benchmark suite.
Why? Because they are the only vendor-independent suite of digital signal processing benchmarks. Each benchmark is carefully designed by BDTI’s team of DSP and benchmarking experts. These benchmarks are recognized industry-wide as trustworthy, reliable, and accurate. And because they are so widely used, BDTI provides a vast pool of benchmark data—giving you the context you need to evaluate a processor’s results.
Motivation for the BDTI DSP Kernel Benchmarks
When BDTI designed the first version of the BDTI DSP Kernel Benchmarks in 1994, it was motivated by a need to realistically evaluate the digital signal processing capabilities of processors in order to help processor vendors understand their competitive standing and to assist system designers in making informed assessments of their options. In BDTI’s view, realistic evaluation of processor architectures on digital signal processing workloads required a benchmark suite based on realistic workloads and realistic programming techniques, and BDTI emphasized these factors in its design of the suite. The market has validated BDTI’s approach: since 1994, the BDTI DSP Kernel Benchmarks have been the most widely used and most trusted benchmarks for evaluating digital signal processing performance.
However, applications and processors change over time. BDTI updated the benchmarks in 2000. Version 2000 modified some of the benchmarks and replaced others in order to better reflect the evolution of digital signal processing application workloads. Now, in 2012, BDTI has released a new version of the BDTI DSP Kernel Benchmarks to keep the suite in step with the latest applications and processors.
Version 2012 of the DSP Kernel Benchmarks is comprised of eleven kernels (see the table below). As with previous versions, Version 2012 of the BDTI DSP Kernel Benchmarks is used to generate BDTI’s composite metrics, which include:
- Speed: The BDTImark2012™
- Memory use: The BDTImemMark2012™
- Cost efficiency: The BDTImark2012/$™
- Area efficiency: The BDTImark2012/mm2™
- Energy efficiency: The BDTImark2012/Watt™
Why Use the BDTI DSP Kernel Benchmarks
The BDTI DSP Kernel Benchmarks are essential for both processor vendors and processor users; each derives unique benefits from using BDTI’s benchmarks.
Benefits for Processor Vendors
If you're a processor vendor you need to convince your customers that your processor has the performance they’re looking for, so you’ll need to provide them with benchmark results. But customers are invariably skeptical of benchmark results from companies that are trying to sell them something. They have no way of knowing whether the results are real and accurate, and whether they are comparable to results from other vendors.
If you use the BDTI DSP Kernel Benchmarks these problems disappear. Processor users recognize the integrity of BDTI’s benchmarks, and they are far more likely to believe your marketing pitch if you have independently certified results to back it up.
BDTI’s benchmarks evaluate all of the key performance metrics, including speed, energy efficiency, cost-performance, and memory use. And because BDTI has benchmarked dozens of processors with this suite, you’ll be able to present apples-to-apples performance comparisons to your competitors—and seal the deal.
Of course, BDTI’s benchmarks are useful for more than marketing; many vendors use the BDTI DSP Kernel Benchmarks to evaluate and improve processor designs that are still in progress.
Benefits for End System and Product Designers
If you're an end product designer trying to select the right processor for your design, you know that choosing the best processor for your product is one of the most important design decisions you’ll make. If you get it wrong, your product may not work as well as you’d hoped—or it may not work at all.
To make a good decision, you’ll need benchmark results. But not just any results will do—you need to know that the results you’re using are trustworthy, unbiased, and accurate.
Most processor vendors provide results for their own processors, but there is no way to know whether those results are comparable to results from other vendors. Some vendors are more careful and fair than others, but few have the resources to create high-quality and technically sound benchmarks, and then implement them across a range of competitor processors.
BDTI has over a decade of experience in developing and implementing signal processing benchmarks, and we are recognized worldwide as the independent benchmarking experts. If you use our benchmark results you’ll be confident that you’re basing your decisions on unbiased performance data. Better yet, you can compare results for many different processors—from many different vendors—easily, quickly, and reliably.
Technical Details
Version 2012 of the BDTI DSP Kernel Benchmarks comprises 11 digital signal processing algorithm kernels:
|
The BDTI DSP Kernel Benchmarks™ Version 2012 |
||
|---|---|---|
|
Benchmark |
Description | Example Application |
|
Real Block FIR |
Finite impulse response filter that operates on a block of real (not complex) data. |
Speech processing (e.g., G.728 speech coding). |
|
Real Single-Sample FIR |
FIR filter that operates on a single sample of real data. |
Speech processing, general filtering. |
|
LMS Adaptive FIR |
Least-mean-square adaptive filter; operates on a single sample of real data. |
Channel equalization, servo control, linear predictive coding. |
|
IIR |
Infinite impulse response filter that operates on a single sample of real data. |
Audio processing, general filtering. |
|
Vector Dot Product |
Sum of the pointwise multiplication of two vectors. |
Convolution, correlation, matrix multiplication, multi-dimensional signal processing. |
|
Vector Add |
Pointwise addition of two vectors, producing a third vector. |
Graphics, combining audio signals or images. |
|
Vector Maximum |
Find the value and location of the maximum value in a vector. |
Error control coding, algorithms using block floating-point. |
|
Control |
A sequence of control operations (test, branch, push, pop, and bit manipulation). |
Virtually all DSP applications include some control code. |
|
256-Point In-Place FFT |
Fast Fourier Transform converts a time-domain signal to the frequency domain. |
Radar, sonar, MPEG audio compression, spectral analysis. |
|
Bit Unpack |
Unpacks variable-length data from a bit stream. |
Audio decompression, protocol handling. |
|
Histogram |
Estimates the statistical distribution of a variable by counting the number of vector elements that fall into equally-spaced ranges. |
Image processing, entropy coding |
Each kernel is implemented in hand-optimized assembly language on the target processor.
License to the BDTI DSP Kernel Benchmarks
A license to the BDTI DSP Kernel Benchmarks provides:
- The BDTI DSP Kernel Benchmarks™ Specification for use in benchmarking one processor
- Benchmark reference C source code
- Test vectors for verifying the functionality of BDTI DSP Kernel Benchmarks implementations
- One-year license to use the above items for one project
- Training and technical support
Please see Overview of Benchmark Licensing for more information.
Next Steps
To license the BDTI DSP Kernel Benchmarks, initiate an analysis engagement, or discuss your benchmarking needs, please call us at +1 925 954 1411 or contact us via the web.