DVSDK Codec Performance Measurements

Performance Measurements in DVSDK

The following topic presents different performance measurements taken for different DVSDK releases for a particular platform. Its main purpose is to serve as a reference of the system performance expectations when building an application using DVSDK for video codecs. It measures the time it takes to process a video frame from an application perspective for certain environment conditions; given than video requires high computational and memory resources compared to other codecs.

Tools used to measure the performance

This section explains the different tools and environments that were used when evaluating the System Performance for a particular video codec.

DSP Only

These results are the same results that are presented in the associated Codec Data Sheet for the respective codec used. They reflect the performance that is obtained for a particular codec (given the environment settings) when only the DSP is used and only the codec code runs; it also means that DSP has unlimited access to shared resources like the memory. Therefore, it represents the baseline from which real applications should me measured against. The unit is converted from MHz to milliseconds since it is based on the environment settings used for a particular platform/DVSDK combination.

File To File

A simple application (like DVTB or a DMAI example application) is used to process a file, to produce the results into another file, without the use of any capture/display. Contrary to the DSP Only measurement, the use of a Codec Engine, a Codec Server, and a multi-processor system with shared resources is evaluated here.

Out of the Box Demonstration Applications

The demo application is a non-optimized application, it is intended to be an out of the box demonstration of the capabilities of a particular platform. It tries to present as best as possible a real-time implemenatation of the system capabilities, like encoding, decoding or simultaneous encode & decode, using platform available devices like cameras, displays, ethernet connections, memory cards, etc. There are several differences among the different demos distributed per platform/DVSDK and further optimizations than what is reported may be possible. Thus, the performance results should be used to understand the impact of running more functions in the system and not what is achievable.

Important Notes

1. For the DECODING scenarios, only video clips with constant bit rates (CBR) were used for a valid assessment of the impact that the bit-rate has in the performance.
2. MV represents the number of Motion Vectors used per Macro Block.
3. The Frames per Second value (FPS) is only valid when enough frames are available, typically more than 3 seconds; hence sometimes this column is Not Available.
4. Some of the clips used are not commercially available, and were used for internal validation purposes only.

Results and Analysis

1. OMAP3530: DVSDK 3.00 (Linux)
2. DM365: DVSDK 2.10 (Linux)

See Also

• C64x+ and Davinci codec performance tables