The BDTImark2000™ is a summary measure of processors’ signal-processing speed. The score is distilled from a processor’s results on the BDTI DSP Kernel Benchmarks™, a suite of 12 key DSP algorithms. A higher score indicates a faster processor.
Because it is based on realistic benchmarks, the BDTImark2000 characterizes a processor’s signal processing speed far more accurately than simplified measures such as millions of multiply-accumulates per second (MMACS). Moreover, the benchmarks themselves are implemented in hand-coded assembly language, just as would be done for critical DSP inner loops in a performance-critical application.
Note that BDTImark2000 scores represent the signal-processing performance of a single processor core only. Where it is practical and meaningful to do so, BDTI makes first-order projections of best-case multi-core and/or multi-threaded performance. For example, on a four-core device, some applications may be able to achieve performance that approaches four times the performance of a single-core device. In this case, we project best-case performance of a four-core device by multiplying the single-core BDTImark2000 score by a factor of four.
For more information about the BDTImark2000, please see our white paper The BDTImark2000: A Measure of DSP Execution Speed.
BDTImark2000 scores (and its related BDTIsimMark2000 score, see below) can be used to derive metrics other than speed. BDTI uses these scores to derive the following metrics:
- Cost effectiveness (BDTImark2000/$ and BDTIsimMark2000/$)
- Silicon area efficiency (BDTImark2000/mm² and BDTIsimMark2000/mm²)
- Power efficiency (BDTImark2000/mW and BDTIsimMark2000/mW)
The BDTImemMark2000™ is a summary measure of processors’ memory efficiency on signal-processing applications. The BDTImemMark2000 is based on the same suite of signal-processing benchmarks as the BDTImark2000 and BDTIsimMark2000. A higher BDTImemMark2000 score indicates a more efficient processor. That is, a higher BDTImemMark2000 score indicates lower memory use. Memory efficiency is important for two reasons: First, a processor’s memory efficiency has a significant impact on overall system cost and energy consumption. Second, a processor may experience significant performance degradation if frequently-accessed application code and data do not fit in level-one memory.
The BDTImark2000, BDTIsimMark2000, and BDTImemMark2000 summarize performance on the twelve BDTI DSP Kernel Benchmarks. These summary metrics are useful for making first-order assessments of processors. System designers can obtain a more accurate assessment—one that takes into account the nature of their applications—by inspecting the results for each of the BDTI DSP Kernel Benchmarks. System designers can obtain these individual results from BDTI.
|
Measurement |
Fixed-point chips |
Floating-point chips |
Licensable cores (130nm) |
Licensable cores (65nm) |
|---|---|---|---|---|
|
Execution Speed |
|
|
||
|
Cost Effectiveness |
n/a |
n/a |
||
|
Silicon Area Efficiency |
n/a |
n/a |
||
|
Power Efficiency |
||||
|
Memory Use |
BDTImark2000 vs. BDTIsimMark2000: What's the Difference?
BDTI’s policy is to verify benchmark results on silicon before issuing a BDTImark2000 score. This policy helps to ensure that the score accurately reflects the performance that can be expected from actual silicon available today.
Unfortunately, it is not always practical to verify benchmarks on hardware. For example, a chip designer may need to evaluate a licensable core before the core has been fabricated. To meet such needs, BDTI publishes the BDTIsimMark2000™. This metric is calculated in the same manner as the BDTImark2000, but is based on simulated results instead of hardware measurements. Although BDTIsimMark2000 and BDTImark2000 scores are calculated in the same manner, they should be compared with caution. In addition, caution should be used when comparing scores for chips to scores for cores. Note that the best-case multi-core and multi-threaded performance projections are not verified on hardware.
The following table explains how BDTI determines whether a processor receives a BDTImark2000 or BDTIsimMark2000 score. Although BDTIsimMark2000 and BDTImark2000 scores are calculated in the same manner, they should be compared with caution. This is particularly true when comparing a processor that has not been fabricated to a processor that has been fabricated. First, the pre-silicon processor may not achieve the expected clock speed. Second, in the time it takes for the pre-silicon processor to reach production, competing vendors may achieve higher clock speeds or introduce new architectures.
|
Processor Type |
Fabrication Status |
Measurement Methodology |
Clock Speed Used |
Score Type |
|---|---|---|---|---|
|
Chip |
Available in at least sample quantities |
Hardware |
Clock speed of fastest available chip |
BDTImark2000 |
|
Available in at least sample quantities |
Simulator |
Clock speed of fastest available chip |
BDTIsimMark2000 |
|
|
Not yet available |
Simulator |
Projected clock speed of fastest chip |
BDTIsimMark2000 |
|
|
Core |
Has been fabricated |
Hardware |
Projected worst-case clock speed in a 0.13μm process |
BDTImark2000 |
|
Has been fabricated |
Simulator |
Projected worst-case clock speed in a 0.13μm process |
BDTIsimMark2000 |
|
|
Has not been fabricated |
Simulator |
Projected worst-case clock speed in a 0.13μm process |
BDTIsimMark2000 |
It is also hazardous to compare scores for chips to scores for cores. For chips, vendors guarantee that the processor will achieve a certain clock speed. For cores, the clock speed depends on the fabrication process, synthesis targets, and other factors. Hence, the clock speed of a core may vary dramatically from one design to the next.
For the sake of consistency, BDTI calculates scores for licensable cores using projected worst-case clock speeds in a 0.13 µm process. In this context, “worst-case clock speed” means the clock speed projected for a core assuming worst-case process, voltage, and temperature variations.
BDTI uses worst-case speeds for licensable cores because this is the speed with which chip designers typically are most concerned. Although BDTI relies upon processor vendors to supply the projected clock speed, BDTI evaluates the stated speed based on a variety of factors (such as the speed of similar processors) and adjusts the projection as necessary to obtain what it believes to be a realistic clock speed projection.
Because clock speed does not affect memory use, the BDTImemMark2000 does not distinguish between pre-silicon processors and those that are in production. Similarly, BDTI does not require hardware verification before it issues a BDTImemMark2000 score.