VPU codec performance information for QNX TI_J784S4_EVM_Board

Description

This datasheet lists various performance metrics for the QNX VPU encoder and decoder stack.

Attention: This is a preliminary release.

Test Setup

Please refer to the SDK documentation on how to re-run the resouce manager to start collecting performance information for codec use cases.

Encoder

The primary test util used to measure case is the omxil_video_enc_p binary bundled with the SDK release. This utility takes in the same parameters as the omcil_video_enc util. The only difference is that, the _p variant keeps using a subset of the input that exists in the test vector and keeps cycling between those input (10 frames by default) to minimize the file system load while running the test. Further, it discards the out data encoded to further reduce CPU load.

Decoder

The decoder test util omxil_video_dec is used as is for performance measurements. To reduce the load on CPU, the decoded frames are discarded and are not stored anywhere (-o option is not used).

For ease of measurements, the perf instrumentation is captured at the resource manager level. The OMX component overhead is assumed to be minimal and hence not considered

Results

Test Case	Description	FPS
1x 4K Encode@60fps	Encode 1 stream at 60fps	65.15
2x 4K Encode@30fps	Encode 2 streams simultaneously at 30fps	32.70
4x 1080p Encode@60fps	Encode 4 streams simultaneously at 60fps	48.67
8x 1080p Encode@30fps	Encode 8 streams simultaneously at 30fps	23.70
8x D1 Encode@30fps	Encode 8 streams simultaneously at 30fps	41.20
1x 4K Decode@60fps	Decode 1 stream at 60fps	71.55
2x 4K Decode@60fps	Decode 2 streams simultaneously (Multicore) at 60fps	71.19
4x 1080p Decode@60fps	Decode 4 streams simultaneously at 60fps	53.05
4x 1080p Decode@60fps	Decode 4 streams simultaneously (Multicore) at 60fps	49.46
8x 1080p Decode@30fps	Decode 8 streams simultaneously at 30fps	26.26
8x D1 Decode@30fps	Decode 8 streams simultaneously at 30fps	59.10

Buffer consumption stats for the VPU codec

The following table gives a rough guideline on the amount of internal and IO buffers used based on empirical measurements.

**_NOTE:_** The memory usage might vary based on the profile, bitrate and other factors.

Per VPU instance

4K - AVC

Buffer Type	Initial (MB)	Per Channel (MB)
Internal	176	169
External	94	94

4K-HEVC

Buffer Type	Initial (MB)	Per Channel (MB)
Internal	52	46
External	46	46

1080p - AVC

Buffer Type	Initial (MB)	Per Channel (MB)
Internal	97	45
External	28	14

1080p-HEVC

Buffer Type	Initial (MB)	Per Channel (MB)
Internal	67	15
External	44	11