VDEC Performance Specifications

Atlas training products do not support the VDEC function described in this section.

Performance Specifications

This section describes only the performance specifications of Ascend AI Processors of each product, which may differ based on the memory and CPU configurations of the respective product. For details about the product performance specifications, see the official descriptions at the Ascend community.

This section uses a single Ascend AI Processor as an example to describe the performance indicators in basic scenarios. The performance indicators of other resolutions can be estimated in the same way.

In the following table, one channel corresponds to one thread, 720p indicates a resolution of 1280 x 720 pixels, 1080p indicates a resolution of 1920 x 1080 pixels, and 4K indicates a resolution of 3840 x 2160 pixels.

On the Atlas A3 training products/Atlas A3 inference products, an NPU has two devices, and the maximum specifications can be reached only when the two devices work in parallel.

**Table 1** AI processor performance specifications of Atlas A3 training products/Atlas A3 inference products
Scenario	Total Frame Rate
n-channel 720p (1 ≤ n ≤ 4)	n*480fps
n-channel 720p (n > 4)	1920fps
n-channel 1080p (1 ≤ n ≤ 4)	n*240fps
n-channel 1080p (n > 4)	960fps
n-channel 4K (1 ≤ n ≤ 4)	n*60fps
n-channel 4K (n > 4)	240fps

**Table 2** AI processor performance specifications of Atlas A2 training products/Atlas A2 inference products
Scenario	Total Frame Rate
n-channel 720p (1 ≤ n ≤ 2)	n*480fps
n-channel 720p (n > 2)	960fps
n-channel 1080p (1 ≤ n ≤ 2)	n*240fps
n-channel 1080p (n > 2)	480fps
n-channel 4K (1 ≤ n ≤ 2)	n*60fps
n-channel 4K (n > 2)	120fps

**Table 3** Performance specifications of Atlas inference products
Scenario	Total Frame Rate
n-channel 720p (1 ≤ n ≤ 12)	n * 480fps NOTE: For Atlas 300V video analysis cards, when 1 < n ≤ 12, the total frame rate is n x 400 FPS.
n-channel 720p (n > 12)	5760fps NOTE: For Atlas 300V video analysis cards, when n > 12, the total frame rate is 4800 FPS.
n-channel 1080p (1 ≤ n ≤ 12)	n * 240fps NOTE: For Atlas 300V video analysis cards, when 1 < n ≤ 12, the total frame rate is n x 200 FPS.
n-channel 1080p (n > 12)	2880fps NOTE: For Atlas 300V video analysis cards, when n > 12, the total frame rate is 2400 FPS.
n-channel 4K (1 ≤ n ≤ 12)	n * 60fps
n-channel 4K (n > 12)	720fps

**Table 4** Atlas 200I/500 A2 inference products performance specifications
Scenario	Total Frame Rate
n-channel 720p (1 ≤ n ≤ 4)	n * 600fps
n-channel 720p (n > 4)	2400fps
n-channel 1080p (1 ≤ n ≤ 4)	n * 300fps
n-channel 1080p (n > 4)	1200fps
n-channel 4K (1 ≤ n ≤ 4)	n * 75fps
n-channel 4K (n > 4)	300fps

**Table 5** Atlas A2 training products/Atlas A2 inference products performance specifications
Scenario	Total Frame Rate
n-channel 720p (1 ≤ n ≤ 2)	n * 480fps
n-channel 720p (n > 2)	960fps
n-channel 1080p (1 ≤ n ≤ 2)	n * 240fps
n-channel 1080p (n > 2)	480fps
n-channel 4K (1 ≤ n ≤ 2)	n * 60fps
n-channel 4K (n > 2)	120fps

The following table uses 1080p input streams as an example to describe the formula for calculating the maximum buffer used by each VDEC channel. The buffer usage is calculated as follows:

Buffer size for the input streams: ≥ 3/4 of the decoding channel size (width x height).
Frame buffer size for the decoded images: Call acl.himpi.vdec_get_pic_buf_size to obtain the value, which is related to the input stream resolution.
TMV buffer size for decoded video images: Call acl.himpi.vdec_get_tmv_buf_size to obtain the TMV buffer size for decoded video images. The value is related to the resolution of the input stream.
The input stream buffer size, frame buffer size of decoded images, TMV buffer size of decoded video images, and number of reference frames are set when the acl.himpi.vdec_create_chn API is called to create a channel.

The number of buffered frames after decoding is set during the acl.himpi.vdec_set_chn_param call.

Formula	Scenario	Memory Usage (MB)
6 MB + Buffer size for the input streams x 2 + (Frame buffer size for the decoded images + TMV buffer size for the decoded images) x (Number of reference frames + Number of buffered frames after decoding + 1)	Input stream format: H.264 Input stream resolution: 1080p Buffer size for the input streams: 4 MB Frame buffer size for the decoded images: 3 MB TMV buffer size for the decoded images: 0.5 MB Number of buffered frames after decoding: 2	52.5 MB (eight reference frames) 31.5 MB (two reference frames)
Input stream format: H.265 Input stream resolution: 1080p Buffer size for the input streams: 4 MB Frame buffer size for the decoded images: 3 MB TMV buffer size for the decoded images: 1 MB Number of buffered frames after decoding: 2	58 MB (eight reference frames) 34 MB (two reference frames)

Formula

Scenario

Memory Usage (MB)

6 MB + Buffer size for the input streams x 2 + (Frame buffer size for the decoded images + TMV buffer size for the decoded images) x (Number of reference frames + Number of buffered frames after decoding + 1)

Input stream format: H.264
Input stream resolution: 1080p
Buffer size for the input streams: 4 MB
Frame buffer size for the decoded images: 3 MB
TMV buffer size for the decoded images: 0.5 MB
Number of buffered frames after decoding: 2

52.5 MB (eight reference frames)

31.5 MB (two reference frames)

Input stream format: H.265
Input stream resolution: 1080p
Buffer size for the input streams: 4 MB
Frame buffer size for the decoded images: 3 MB
TMV buffer size for the decoded images: 1 MB
Number of buffered frames after decoding: 2

58 MB (eight reference frames)

34 MB (two reference frames)

Parent topic: VDEC/JPEGD APIs